chore: release v0.1.2 (#55)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

.dockerignore

@@ -0,0 +1,8 @@
+target/
+.git/
+.env
+.env.*
+*.md
+!CLAUDE.md
+node_modules/
+tools-src/

.env.example

@@ -1,15 +1,15 @@
 # Database Configuration
-DATABASE_URL=postgres://ironclaw:password@localhost:5432/ironclaw
+DATABASE_URL=postgres://localhost/ironclaw
 DATABASE_POOL_SIZE=10
 
 # LLM Provider (NEAR AI)
 # NEAR AI provides a unified interface to all models with user authentication
-# Session token is stored in ~/.near-agent/session.json and managed automatically.
+# Session token is stored in ~/.ironclaw/session.json and managed automatically.
 # On first run, the agent will open a browser for OAuth authentication.
 NEARAI_MODEL=claude-3-5-sonnet-20241022
 NEARAI_BASE_URL=https://cloud-api.near.ai
 NEARAI_AUTH_URL=https://private.near.ai
-# NEARAI_SESSION_PATH=~/.near-agent/session.json  # optional, default shown
+# NEARAI_SESSION_PATH=~/.ironclaw/session.json  # optional, default shown
 
 # Channel Configuration
 # CLI is always enabled

.github/workflows/code_style.yml

@@ -0,0 +1,21 @@
+name: Code Style
+on:
+  pull_request:
+
+jobs:
+  codestyle:
+    name: Code Style (fmt + clippy)
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repository
+      uses: actions/checkout@v4
+    - name: Install Rust
+      uses: dtolnay/rust-toolchain@stable
+      with:
+        profile: minimal
+        components: rustfmt, clippy
+    - name: Check formatting
+      run: |
+        cargo fmt --all -- --check
+    - name: Check lints (cargo clippy)
+      run: cargo clippy -- -D warnings

.github/workflows/release-plz.yml

@@ -0,0 +1,65 @@
+name: Release-plz
+
+on:
+  push:
+    branches:
+      - main
+
+jobs:
+
+  # Release unpublished packages.
+  release-plz-release:
+    if: ${{ github.repository_owner == 'nearai' }}
+    name: Release-plz release
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - &checkout
+        name: Checkout repository
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - &install-rust
+        name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+      # Generating a GitHub token, so that PRs and tags created by
+      # the release-plz-action can trigger actions workflows.
+      - name: Generate GitHub token
+        uses: actions/create-github-app-token@v2
+        id: generate-token
+        with:
+          # GitHub App ID secret name
+          app-id: ${{ secrets.GH_RELEASES_MANAGER_APP_ID }}
+          # GitHub App private key secret name
+          private-key: ${{ secrets.GH_RELEASES_MANAGER_APP_PRIVATE_KEY }}
+      - name: Run release-plz
+        uses: release-plz/[email protected]
+        with:
+          command: release
+        env:
+          GITHUB_TOKEN: ${{ steps.generate-token.outputs.token }}
+          CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+
+  # Create a PR with the new versions and changelog, preparing the next release.
+  release-plz-pr:
+    if: ${{ github.repository_owner == 'nearai' }}
+    name: Release-plz PR
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+      pull-requests: write
+    concurrency:
+      group: release-plz-${{ github.ref }}
+      cancel-in-progress: false
+    steps:
+      - *checkout
+      - *install-rust
+      - name: Run release-plz
+        uses: release-plz/[email protected]
+        with:
+          command: release-pr
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}

.github/workflows/release.yml

@@ -0,0 +1,329 @@
+# This file was autogenerated by dist: https://axodotdev.github.io/cargo-dist
+#
+# Copyright 2022-2024, axodotdev
+# SPDX-License-Identifier: MIT or Apache-2.0
+#
+# CI that:
+#
+# * checks for a Git Tag that looks like a release
+# * builds artifacts with dist (archives, installers, hashes)
+# * uploads those artifacts to temporary workflow zip
+# * on success, uploads the artifacts to a GitHub Release
+#
+# Note that the GitHub Release will be created with a generated
+# title/body based on your changelogs.
+
+name: Release
+permissions:
+  "contents": "write"
+
+# This task will run whenever you push a git tag that looks like a version
+# like "1.0.0", "v0.1.0-prerelease.1", "my-app/0.1.0", "releases/v1.0.0", etc.
+# Various formats will be parsed into a VERSION and an optional PACKAGE_NAME, where
+# PACKAGE_NAME must be the name of a Cargo package in your workspace, and VERSION
+# must be a Cargo-style SemVer Version (must have at least major.minor.patch).
+#
+# If PACKAGE_NAME is specified, then the announcement will be for that
+# package (erroring out if it doesn't have the given version or isn't dist-able).
+#
+# If PACKAGE_NAME isn't specified, then the announcement will be for all
+# (dist-able) packages in the workspace with that version (this mode is
+# intended for workspaces with only one dist-able package, or with all dist-able
+# packages versioned/released in lockstep).
+#
+# If you push multiple tags at once, separate instances of this workflow will
+# spin up, creating an independent announcement for each one. However, GitHub
+# will hard limit this to 3 tags per commit, as it will assume more tags is a
+# mistake.
+#
+# If there's a prerelease-style suffix to the version, then the release(s)
+# will be marked as a prerelease.
+on:
+  pull_request:
+  push:
+    tags:
+      - '**[0-9]+.[0-9]+.[0-9]+*'
+
+jobs:
+  # Run 'dist plan' (or host) to determine what tasks we need to do
+  plan:
+    runs-on: "ubuntu-22.04"
+    outputs:
+      val: ${{ steps.plan.outputs.manifest }}
+      tag: ${{ !github.event.pull_request && github.ref_name || '' }}
+      tag-flag: ${{ !github.event.pull_request && format('--tag={0}', github.ref_name) || '' }}
+      publishing: ${{ !github.event.pull_request }}
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+          submodules: recursive
+      - name: Install dist
+        # we specify bash to get pipefail; it guards against the `curl` command
+        # failing. otherwise `sh` won't catch that `curl` returned non-0
+        shell: bash
+        run: "curl --proto '=https' --tlsv1.2 -LsSf https://github.com/axodotdev/cargo-dist/releases/download/v0.30.3/cargo-dist-installer.sh | sh"
+      - name: Cache dist
+        uses: actions/upload-artifact@v4
+        with:
+          name: cargo-dist-cache
+          path: ~/.cargo/bin/dist
+      # sure would be cool if github gave us proper conditionals...
+      # so here's a doubly-nested ternary-via-truthiness to try to provide the best possible
+      # functionality based on whether this is a pull_request, and whether it's from a fork.
+      # (PRs run on the *source* but secrets are usually on the *target* -- that's *good*
+      # but also really annoying to build CI around when it needs secrets to work right.)
+      - id: plan
+        run: |
+          dist ${{ (!github.event.pull_request && format('host --steps=create --tag={0}', github.ref_name)) || 'plan' }} --output-format=json > plan-dist-manifest.json
+          echo "dist ran successfully"
+          cat plan-dist-manifest.json
+          echo "manifest=$(jq -c "." plan-dist-manifest.json)" >> "$GITHUB_OUTPUT"
+      - name: "Upload dist-manifest.json"
+        uses: actions/upload-artifact@v4
+        with:
+          name: artifacts-plan-dist-manifest
+          path: plan-dist-manifest.json
+
+  # Build and packages all the platform-specific things
+  build-local-artifacts:
+    name: build-local-artifacts (${{ join(matrix.targets, ', ') }})
+    # Let the initial task tell us to not run (currently very blunt)
+    needs:
+      - plan
+    if: ${{ fromJson(needs.plan.outputs.val).ci.github.artifacts_matrix.include != null && (needs.plan.outputs.publishing == 'true' || fromJson(needs.plan.outputs.val).ci.github.pr_run_mode == 'upload') }}
+    strategy:
+      fail-fast: false
+      # Target platforms/runners are computed by dist in create-release.
+      # Each member of the matrix has the following arguments:
+      #
+      # - runner: the github runner
+      # - dist-args: cli flags to pass to dist
+      # - install-dist: expression to run to install dist on the runner
+      #
+      # Typically there will be:
+      # - 1 "global" task that builds universal installers
+      # - N "local" tasks that build each platform's binaries and platform-specific installers
+      matrix: ${{ fromJson(needs.plan.outputs.val).ci.github.artifacts_matrix }}
+    runs-on: ${{ matrix.runner }}
+    container: ${{ matrix.container && matrix.container.image || null }}
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      BUILD_MANIFEST_NAME: target/distrib/${{ join(matrix.targets, '-') }}-dist-manifest.json
+    steps:
+      - name: enable windows longpaths
+        run: |
+          git config --global core.longpaths true
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+          submodules: recursive
+      - name: Install Rust non-interactively if not already installed
+        if: ${{ matrix.container }}
+        run: |
+          if ! command -v cargo > /dev/null 2>&1; then
+            curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
+            echo "$HOME/.cargo/bin" >> $GITHUB_PATH
+          fi
+      - uses: swatinem/rust-cache@v2
+        with:
+          key: ${{ join(matrix.targets, '-') }}
+          cache-provider: ${{ matrix.cache_provider }}
+      - name: Install dist
+        run: ${{ matrix.install_dist.run }}
+      # Get the dist-manifest
+      - name: Fetch local artifacts
+        uses: actions/download-artifact@v4
+        with:
+          pattern: artifacts-*
+          path: target/distrib/
+          merge-multiple: true
+      - name: Install dependencies
+        run: |
+          ${{ matrix.packages_install }}
+      - name: Build artifacts
+        run: |
+          # Actually do builds and make zips and whatnot
+          dist build ${{ needs.plan.outputs.tag-flag }} --print=linkage --output-format=json ${{ matrix.dist_args }} > dist-manifest.json
+          echo "dist ran successfully"
+      - id: cargo-dist
+        name: Post-build
+        # We force bash here just because github makes it really hard to get values up
+        # to "real" actions without writing to env-vars, and writing to env-vars has
+        # inconsistent syntax between shell and powershell.
+        shell: bash
+        run: |
+          # Parse out what we just built and upload it to scratch storage
+          echo "paths<<EOF" >> "$GITHUB_OUTPUT"
+          dist print-upload-files-from-manifest --manifest dist-manifest.json >> "$GITHUB_OUTPUT"
+          echo "EOF" >> "$GITHUB_OUTPUT"
+
+          cp dist-manifest.json "$BUILD_MANIFEST_NAME"
+      - name: "Upload artifacts"
+        uses: actions/upload-artifact@v4
+        with:
+          name: artifacts-build-local-${{ join(matrix.targets, '_') }}
+          path: |
+            ${{ steps.cargo-dist.outputs.paths }}
+            ${{ env.BUILD_MANIFEST_NAME }}
+
+  # Build and package all the platform-agnostic(ish) things
+  build-global-artifacts:
+    needs:
+      - plan
+      - build-local-artifacts
+    runs-on: "ubuntu-22.04"
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      BUILD_MANIFEST_NAME: target/distrib/global-dist-manifest.json
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+          submodules: recursive
+      - name: Install cached dist
+        uses: actions/download-artifact@v4
+        with:
+          name: cargo-dist-cache
+          path: ~/.cargo/bin/
+      - run: chmod +x ~/.cargo/bin/dist
+      # Get all the local artifacts for the global tasks to use (for e.g. checksums)
+      - name: Fetch local artifacts
+        uses: actions/download-artifact@v4
+        with:
+          pattern: artifacts-*
+          path: target/distrib/
+          merge-multiple: true
+      - id: cargo-dist
+        shell: bash
+        run: |
+          dist build ${{ needs.plan.outputs.tag-flag }} --output-format=json "--artifacts=global" > dist-manifest.json
+          echo "dist ran successfully"
+
+          # Parse out what we just built and upload it to scratch storage
+          echo "paths<<EOF" >> "$GITHUB_OUTPUT"
+          jq --raw-output ".upload_files[]" dist-manifest.json >> "$GITHUB_OUTPUT"
+          echo "EOF" >> "$GITHUB_OUTPUT"
+
+          cp dist-manifest.json "$BUILD_MANIFEST_NAME"
+      - name: "Upload artifacts"
+        uses: actions/upload-artifact@v4
+        with:
+          name: artifacts-build-global
+          path: |
+            ${{ steps.cargo-dist.outputs.paths }}
+            ${{ env.BUILD_MANIFEST_NAME }}
+  # Determines if we should publish/announce
+  host:
+    needs:
+      - plan
+      - build-local-artifacts
+      - build-global-artifacts
+    # Only run if we're "publishing", and only if plan, local and global didn't fail (skipped is fine)
+    if: ${{ always() && needs.plan.result == 'success' && needs.plan.outputs.publishing == 'true' && (needs.build-global-artifacts.result == 'skipped' || needs.build-global-artifacts.result == 'success') && (needs.build-local-artifacts.result == 'skipped' || needs.build-local-artifacts.result == 'success') }}
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+    runs-on: "ubuntu-22.04"
+    outputs:
+      val: ${{ steps.host.outputs.manifest }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+          submodules: recursive
+      - name: Install cached dist
+        uses: actions/download-artifact@v4
+        with:
+          name: cargo-dist-cache
+          path: ~/.cargo/bin/
+      - run: chmod +x ~/.cargo/bin/dist
+      # Fetch artifacts from scratch-storage
+      - name: Fetch artifacts
+        uses: actions/download-artifact@v4
+        with:
+          pattern: artifacts-*
+          path: target/distrib/
+          merge-multiple: true
+      - id: host
+        shell: bash
+        run: |
+          dist host ${{ needs.plan.outputs.tag-flag }} --steps=upload --steps=release --output-format=json > dist-manifest.json
+          echo "artifacts uploaded and released successfully"
+          cat dist-manifest.json
+          echo "manifest=$(jq -c "." dist-manifest.json)" >> "$GITHUB_OUTPUT"
+      - name: "Upload dist-manifest.json"
+        uses: actions/upload-artifact@v4
+        with:
+          # Overwrite the previous copy
+          name: artifacts-dist-manifest
+          path: dist-manifest.json
+      # Create a GitHub Release while uploading all files to it
+      - name: "Download GitHub Artifacts"
+        uses: actions/download-artifact@v4
+        with:
+          pattern: artifacts-*
+          path: artifacts
+          merge-multiple: true
+      - name: Cleanup
+        run: |
+          # Remove the granular manifests
+          rm -f artifacts/*-dist-manifest.json
+      - name: Create GitHub Release
+        env:
+          PRERELEASE_FLAG: "${{ fromJson(steps.host.outputs.manifest).announcement_is_prerelease && '--prerelease' || '' }}"
+          ANNOUNCEMENT_TITLE: "${{ fromJson(steps.host.outputs.manifest).announcement_title }}"
+          ANNOUNCEMENT_BODY: "${{ fromJson(steps.host.outputs.manifest).announcement_github_body }}"
+          RELEASE_COMMIT: "${{ github.sha }}"
+        run: |
+          # Write and read notes from a file to avoid quoting breaking things
+          echo "$ANNOUNCEMENT_BODY" > $RUNNER_TEMP/notes.txt
+
+          gh release create "${{ needs.plan.outputs.tag }}" --target "$RELEASE_COMMIT" $PRERELEASE_FLAG --title "$ANNOUNCEMENT_TITLE" --notes-file "$RUNNER_TEMP/notes.txt" artifacts/*
+
+  publish-npm:
+    needs:
+      - plan
+      - host
+    runs-on: "ubuntu-22.04"
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      PLAN: ${{ needs.plan.outputs.val }}
+    if: ${{ !fromJson(needs.plan.outputs.val).announcement_is_prerelease || fromJson(needs.plan.outputs.val).publish_prereleases }}
+    steps:
+      - name: Fetch npm packages
+        uses: actions/download-artifact@v4
+        with:
+          pattern: artifacts-*
+          path: npm/
+          merge-multiple: true
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '20.x'
+          registry-url: 'https://registry.npmjs.org'
+      - run: |
+          for release in $(echo "$PLAN" | jq --compact-output '.releases[] | select([.artifacts[] | endswith("-npm-package.tar.gz")] | any)'); do
+            pkg=$(echo "$release" | jq '.artifacts[] | select(endswith("-npm-package.tar.gz"))' --raw-output)
+            npm publish --access public "./npm/${pkg}"
+          done
+        env:
+          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+
+  announce:
+    needs:
+      - plan
+      - host
+      - publish-npm
+    # use "always() && ..." to allow us to wait for all publish jobs while
+    # still allowing individual publish jobs to skip themselves (for prereleases).
+    # "host" however must run to completion, no skipping allowed!
+    if: ${{ always() && needs.host.result == 'success' && (needs.publish-npm.result == 'skipped' || needs.publish-npm.result == 'success') }}
+    runs-on: "ubuntu-22.04"
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+          submodules: recursive

.github/workflows/test.yml

@@ -0,0 +1,20 @@
+name: Run Tests
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+jobs:
+  tests:
+    name: Run Tests
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+      - name: Install Rust
+        uses: dtolnay/rust-toolchain@stable
+        with:
+          profile: minimal
+      - name: Run Tests
+        run: cargo test --all-features -- --nocapture

.gitignore

@@ -4,3 +4,6 @@
 
 target/
 
+# WASM build artifacts (loaded from disk, not bundled)
+*.wasm
+

CHANGELOG.md

@@ -0,0 +1,110 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.1.2](https://github.com/nearai/ironclaw/compare/v0.1.1...v0.1.2) - 2026-02-12
+
+### Other
+
+- Added Installation instructions for the pre-built binaries
+- Disabled Windows ARM64 builds as auto-updater [provided by cargo-dist] does not support this platform yet and it is not a common platform for us to support
+
+## [0.1.1](https://github.com/nearai/ironclaw/compare/v0.1.0...v0.1.1) - 2026-02-12
+
+### Other
+
+- Renamed the secrets in release-plz.yml to match the configuration
+- Make sure that the binaries release CD it kicking in after release-plz
+
+## [0.1.0](https://github.com/nearai/ironclaw/releases/tag/v0.1.0) - 2026-02-12
+
+### Added
+
+- Add multi-provider LLM support via rig-core adapter ([#36](https://github.com/nearai/ironclaw/pull/36))
+- Sandbox jobs ([#4](https://github.com/nearai/ironclaw/pull/4))
+- Add Google Suite & Telegram WASM tools ([#9](https://github.com/nearai/ironclaw/pull/9))
+- Improve CLI ([#5](https://github.com/nearai/ironclaw/pull/5))
+
+### Fixed
+
+- resolve runtime panic in Linux keychain integration ([#32](https://github.com/nearai/ironclaw/pull/32))
+
+### Other
+
+- Skip release-plz on forks
+- Upgraded release-plz CD pipeline
+- Added CI/CD and release pipelines ([#45](https://github.com/nearai/ironclaw/pull/45))
+- DM pairing + Telegram channel improvements ([#17](https://github.com/nearai/ironclaw/pull/17))
+- Fixes build, adds missing sse event and correct command ([#11](https://github.com/nearai/ironclaw/pull/11))
+- Codex/feature parity pr hook ([#6](https://github.com/nearai/ironclaw/pull/6))
+- Add WebSocket gateway and control plane ([#8](https://github.com/nearai/ironclaw/pull/8))
+- select bundled Telegram channel and auto-install ([#3](https://github.com/nearai/ironclaw/pull/3))
+- Adding skills for reusable work
+- Fix MCP tool calls, approval loop, shutdown, and improve web UI
+- Add auth mode, fix MCP token handling, and parallelize startup loading
+- Merge remote-tracking branch 'origin/main' into ui
+- Adding web UI
+- Rename `setup` CLI command to `onboard` for compatibility
+- Add in-chat extension discovery, auth, and activation system
+- Add Telegram typing indicator via WIT on-status callback
+- Add proactivity features: memory CLI, session pruning, self-repair notifications, slash commands, status diagnostics, context warnings
+- Add hosted MCP server support with OAuth 2.1 and token refresh
+- Add interactive setup wizard and persistent settings
+- Rebrand to IronClaw with security-first mission
+- Fix build_software tool stuck in planning mode loop
+- Enable sandbox by default
+- Fix Telegram Markdown formatting and clarify tool/memory distinctions
+- Simplify Telegram channel config with host-injected tunnel/webhook settings
+- Apply Telegram channel learnings to WhatsApp implementation
+- Merge remote-tracking branch 'origin/main'
+- Docker file for sandbox
+- Replace hardcoded intent patterns with job tools
+- Fix router test to match intentional job creation patterns
+- Add Docker execution sandbox for secure shell command isolation
+- Move setup wizard credentials to database storage
+- Add interactive setup wizard for first-run configuration
+- Add Telegram Bot API channel as WASM module
+- Add OpenClaw feature parity tracking matrix
+- Add Chat Completions API support and expand REPL debugging
+- Implementing channels to be handled in wasm
+- Support non interactive mode and model selection
+- Implement tool approval, fix tool definition refresh, and wire embeddings
+- Tool use
+- Wiring more
+- Add heartbeat integration, planning phase, and auto-repair
+- Login flow
+- Extend support for session management
+- Adding builder capability
+- Load tools at launch
+- Fix multiline message rendering in TUI
+- Parse NEAR AI alternative response format with output field
+- Handle NEAR AI plain text responses
+- Disable mouse capture to allow text selection in TUI
+- Add verbose logging to debug empty NEAR AI responses
+- Improve NEAR AI response parsing for varying response formats
+- Show status/thinking messages in chat window, debug empty responses
+- Add timeout and logging to NEAR AI provider
+- Add status updates to show agent thinking/processing state
+- Add CLI subcommands for WASM tool management
+- Fix TUI shutdown: send /shutdown message and handle in agent loop
+- Remove SimpleCliChannel, add Ctrl+D twice quit, redirect logs to TUI
+- Fix TuiChannel integration and enable in main.rs
+- Integrate Codex patterns: task scheduler, TUI, sessions, compaction
+- Adding LICENSE
+- Add README with IronClaw branding
+- Add WASM sandbox secure API extension
+- Wire database Store into agent loop
+- Implementing WASM runtime
+- Add workspace integration tests
+- Compact memory_tree output format
+- Replace memory_list with memory_tree tool
+- Simplify workspace to path-based storage, remove legacy code
+- Add NEAR AI chat-api as default LLM provider
+- Add CLAUDE.md project documentation
+- Add workspace and memory system (OpenClaw-inspired)
+- Initial implementation of the agent framework

CLAUDE.md

@@ -11,8 +11,13 @@
 - **Always available** - Multi-channel access with proactive background execution
 
 ### Features
-- **Multi-channel input**: TUI (Ratatui), HTTP webhooks, Telegram, WhatsApp, Slack (WASM channels)
+- **Multi-channel input**: TUI (Ratatui), HTTP webhooks, WASM channels (Telegram, Slack), web gateway
 - **Parallel job execution** with state machine and self-repair for stuck jobs
+- **Sandbox execution**: Docker container isolation with orchestrator/worker pattern
+- **Claude Code mode**: Delegate jobs to Claude CLI inside containers
+- **Routines**: Scheduled (cron) and reactive (event, webhook) task execution
+- **Web gateway**: Browser UI with SSE/WebSocket real-time streaming
+- **Extension management**: Install, auth, activate MCP/WASM extensions
 - **Extensible tools**: Built-in tools, WASM sandbox, MCP client, dynamic builder
 - **Persistent memory**: Workspace with hybrid search (FTS + vector via RRF)
 - **Prompt injection defense**: Sanitizer, validator, policy rules, leak detection
@@ -59,7 +64,9 @@ src/
 │   ├── context_monitor.rs # Memory pressure detection
 │   ├── undo.rs         # Turn-based undo/redo with checkpoints
 │   ├── submission.rs   # Submission parsing (undo, redo, compact, clear, etc.)
-│   └── task.rs         # Sub-task execution framework
+│   ├── task.rs         # Sub-task execution framework
+│   ├── routine.rs      # Routine types (Trigger, Action, Guardrails)
+│   └── routine_engine.rs # Routine execution (cron ticker, event matcher)
 │
 ├── channels/           # Multi-channel input
 │   ├── channel.rs      # Channel trait, IncomingMessage, OutgoingResponse
@@ -72,8 +79,33 @@ src/
 │   │   ├── overlay.rs  # Approval overlays
 │   │   └── composer.rs # Message composition
 │   ├── http.rs         # HTTP webhook (axum) with secret validation
-│   ├── slack.rs        # Stub
-│   └── telegram.rs     # Stub
+│   ├── repl.rs         # Simple REPL (for testing)
+│   ├── web/            # Web gateway (browser UI)
+│   │   ├── mod.rs      # Gateway builder, startup
+│   │   ├── server.rs   # Axum router, 40+ API endpoints
+│   │   ├── sse.rs      # SSE broadcast manager
+│   │   ├── ws.rs       # WebSocket gateway + connection tracking
+│   │   ├── types.rs    # Request/response types, SseEvent enum
+│   │   ├── auth.rs     # Bearer token auth middleware
+│   │   ├── log_layer.rs # Tracing layer for log streaming
+│   │   └── static/     # HTML, CSS, JS (single-page app)
+│   └── wasm/           # WASM channel runtime
+│       ├── mod.rs
+│       ├── bundled.rs  # Bundled channel discovery
+│       └── wrapper.rs  # Channel trait wrapper for WASM modules
+│
+├── orchestrator/       # Internal HTTP API for sandbox containers
+│   ├── mod.rs
+│   ├── api.rs          # Axum endpoints (LLM proxy, events, prompts)
+│   ├── auth.rs         # Per-job bearer token store
+│   └── job_manager.rs  # Container lifecycle (create, stop, cleanup)
+│
+├── worker/             # Runs inside Docker containers
+│   ├── mod.rs
+│   ├── runtime.rs      # Worker execution loop (tool calls, LLM)
+│   ├── claude_bridge.rs # Claude Code bridge (spawns claude CLI)
+│   ├── api.rs          # HTTP client to orchestrator
+│   └── proxy_llm.rs    # LlmProvider that proxies through orchestrator
 │
 ├── safety/             # Prompt injection defense
 │   ├── sanitizer.rs    # Pattern detection, content escaping
@@ -96,6 +128,9 @@ src/
 │   │   ├── file.rs     # ReadFile, WriteFile, ListDir, ApplyPatch
 │   │   ├── shell.rs    # Shell command execution
 │   │   ├── memory.rs   # Memory tools (search, write, read, tree)
+│   │   ├── job.rs      # CreateJob, ListJobs, JobStatus, CancelJob
+│   │   ├── routine.rs  # routine_create/list/update/delete/history
+│   │   ├── extension_tools.rs # Extension install/auth/activate/remove
 │   │   └── marketplace.rs, ecommerce.rs, taskrabbit.rs, restaurant.rs (stubs)
 │   ├── builder/        # Dynamic tool building
 │   │   ├── core.rs     # BuildRequirement, SoftwareType, Language
@@ -236,6 +271,30 @@ HEARTBEAT_ENABLED=true
 HEARTBEAT_INTERVAL_SECS=1800            # 30 minutes
 HEARTBEAT_NOTIFY_CHANNEL=tui
 HEARTBEAT_NOTIFY_USER=default
+
+# Web gateway
+GATEWAY_ENABLED=true
+GATEWAY_HOST=127.0.0.1
+GATEWAY_PORT=3001
+GATEWAY_AUTH_TOKEN=changeme           # Required for API access
+GATEWAY_USER_ID=default
+
+# Docker sandbox
+SANDBOX_ENABLED=true
+SANDBOX_IMAGE=ironclaw-worker:latest
+SANDBOX_MEMORY_LIMIT_MB=512
+SANDBOX_TIMEOUT_SECS=1800
+
+# Claude Code mode (runs inside sandbox containers)
+CLAUDE_CODE_ENABLED=false
+CLAUDE_CODE_MODEL=claude-sonnet-4-20250514
+CLAUDE_CODE_MAX_TURNS=50
+CLAUDE_CODE_CONFIG_DIR=/home/worker/.claude
+
+# Routines (scheduled/reactive execution)
+ROUTINES_ENABLED=true
+ROUTINES_CRON_INTERVAL=60            # Tick interval in seconds
+ROUTINES_MAX_CONCURRENT=3
 ```
 
 ### NEAR AI Provider
@@ -297,13 +356,14 @@ Key test patterns:
 
 ## Current Limitations / TODOs
 
-1. **Slack/Telegram channels** - Stubs only, need implementation
-2. **Domain-specific tools** - `marketplace.rs`, `restaurant.rs`, `taskrabbit.rs`, `ecommerce.rs` return placeholder responses; need real API integrations
-3. **Integration tests** - Need testcontainers setup for PostgreSQL
-4. **MCP stdio transport** - Only HTTP transport implemented
-5. **WIT bindgen integration** - Auto-extract tool description/schema from WASM modules (stubbed)
-6. **Capability granting after tool build** - Built tools get empty capabilities; need UX for granting HTTP/secrets access
-7. **Tool versioning workflow** - No version tracking or rollback for dynamically built tools
+1. **Domain-specific tools** - `marketplace.rs`, `restaurant.rs`, `taskrabbit.rs`, `ecommerce.rs` return placeholder responses; need real API integrations
+2. **Integration tests** - Need testcontainers setup for PostgreSQL
+3. **MCP stdio transport** - Only HTTP transport implemented
+4. **WIT bindgen integration** - Auto-extract tool description/schema from WASM modules (stubbed)
+5. **Capability granting after tool build** - Built tools get empty capabilities; need UX for granting HTTP/secrets access
+6. **Tool versioning workflow** - No version tracking or rollback for dynamically built tools
+7. **Webhook trigger endpoint** - Routines webhook trigger not yet exposed in web gateway
+8. **Full channel status view** - Gateway status widget exists, but no per-channel connection dashboard
 
 ### Completed
 
@@ -320,6 +380,13 @@ Key test patterns:
 - ✅ **Tool approval enforcement** - Tools with `requires_approval()` (shell, http, file write/patch, build_software) now gate execution, track auto-approved tools per session
 - ✅ **Tool definition refresh** - Tool definitions refreshed each iteration so newly built tools become visible in same session
 - ✅ **Worker tool call handling** - Uses `respond_with_tools()` to properly execute tool calls when `select_tools()` returns empty
+- ✅ **Gateway control plane** - Web gateway with 40+ API endpoints, SSE/WebSocket
+- ✅ **Web Control UI** - Browser-based dashboard with chat, memory, jobs, logs, extensions, routines
+- ✅ **Slack/Telegram channels** - Implemented as WASM tools
+- ✅ **Docker sandbox** - Orchestrator/worker containers with per-job auth
+- ✅ **Claude Code mode** - Delegate jobs to Claude CLI inside containers
+- ✅ **Routines system** - Cron, event, webhook, and manual triggers with guardrails
+- ✅ **Extension management** - Install, auth, activate MCP/WASM extensions via CLI and web UI
 
 ## Adding a New Tool
 

Cargo.lock

@@ -182,6 +182,12 @@ version = "0.7.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7c02d123df017efcdfbd739ef81735b36c5ba83ec3c59c80a9d7ecc718f92e50"
 
+[[package]]
+name = "as-any"
+version = "0.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b0f477b951e452a0b6b4a10b53ccd569042d1d01729b519e02074a9c0958a063"
+
 [[package]]
 name = "async-broadcast"
 version = "0.7.2"
@@ -307,6 +313,28 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "async-stream"
+version = "0.3.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b5a71a6f37880a80d1d7f19efd781e4b5de42c88f0722cc13bcb6cc2cfe8476"
+dependencies = [
+ "async-stream-impl",
+ "futures-core",
+ "pin-project-lite",
+]
+
+[[package]]
+name = "async-stream-impl"
+version = "0.3.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c7c24de15d275a1ecfd47a380fb4d5ec9bfe0933f309ed5e705b775596a3574d"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "async-task"
 version = "4.7.1"
@@ -820,6 +848,16 @@ dependencies = [
  "crossterm 0.29.0",
 ]
 
+[[package]]
+name = "core-foundation"
+version = "0.9.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91e195e091a93c46f7102ec7818a2aa394e1e1771c3ab4825963fa03e45afb8f"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
 [[package]]
 name = "core-foundation"
 version = "0.10.1"
@@ -996,6 +1034,17 @@ dependencies = [
  "syn 2.0.114",
 ]
 
+[[package]]
+name = "cron"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eee8b2b4516038bc0f1d3c9934bcb4a13dd316e04abbc63c96757a6d75978532"
+dependencies = [
+ "chrono",
+ "nom",
+ "once_cell",
+]
+
 [[package]]
 name = "crossbeam"
 version = "0.8.4"
@@ -1468,6 +1517,17 @@ dependencies = [
  "pin-project-lite",
 ]
 
+[[package]]
+name = "eventsource-stream"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "74fef4569247a5f429d9156b9d0a2599914385dd189c539334c625d8099d90ab"
+dependencies = [
+ "futures-core",
+ "nom",
+ "pin-project-lite",
+]
+
 [[package]]
 name = "fallible-iterator"
 version = "0.2.0"
@@ -1526,6 +1586,21 @@ version = "0.1.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2"
 
+[[package]]
+name = "foreign-types"
+version = "0.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f6f339eb8adc052cd2ca78910fda869aefa38d22d5cb648e6485e4d3fc06f3b1"
+dependencies = [
+ "foreign-types-shared",
+]
+
+[[package]]
+name = "foreign-types-shared"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "00b0228411908ca8685dba7fc2cdd70ec9990a6e753e89b6ac91a84c40fbaf4b"
+
 [[package]]
 name = "form_urlencoded"
 version = "1.2.2"
@@ -1546,6 +1621,16 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "fs4"
+version = "0.6.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2eeb4ed9e12f43b7fa0baae3f9cdda28352770132ef2e09a23760c29cae8bd47"
+dependencies = [
+ "rustix 0.38.44",
+ "windows-sys 0.48.0",
+]
+
 [[package]]
 name = "funty"
 version = "2.0.0"
@@ -1636,6 +1721,12 @@ version = "0.3.31"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f90f7dce0722e95104fcb095585910c0977252f286e354b5e3bd38902cd99988"
 
+[[package]]
+name = "futures-timer"
+version = "3.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f288b0a4f20f9a56b5d1da57e2227c661b7b16168e2f72365f57b63326e29b24"
+
 [[package]]
 name = "futures-util"
 version = "0.3.31"
@@ -1734,6 +1825,12 @@ dependencies = [
  "stable_deref_trait",
 ]
 
+[[package]]
+name = "glob"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0cc23270f6e1808e30a928bdc84dea0b9b4136a8bc82338574f23baf47bbd280"
+
 [[package]]
 name = "h2"
 version = "0.4.13"
@@ -1933,6 +2030,22 @@ dependencies = [
  "webpki-roots",
 ]
 
+[[package]]
+name = "hyper-tls"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "70206fc6890eaca9fde8a0bf71caa2ddfc9fe045ac9e5c70df101a7dbde866e0"
+dependencies = [
+ "bytes",
+ "http-body-util",
+ "hyper",
+ "hyper-util",
+ "native-tls",
+ "tokio",
+ "tokio-native-tls",
+ "tower-service",
+]
+
 [[package]]
 name = "hyper-util"
 version = "0.1.20"
@@ -1951,9 +2064,11 @@ dependencies = [
  "percent-encoding",
  "pin-project-lite",
  "socket2",
+ "system-configuration",
  "tokio",
  "tower-service",
  "tracing",
+ "windows-registry",
 ]
 
 [[package]]
@@ -2176,7 +2291,7 @@ dependencies = [
 
 [[package]]
 name = "ironclaw"
-version = "0.1.0"
+version = "0.1.2"
 dependencies = [
  "aes-gcm",
  "aho-corasick",
@@ -2189,15 +2304,18 @@ dependencies = [
  "bytes",
  "chrono",
  "clap",
+ "cron",
  "crossterm 0.28.1",
  "deadpool-postgres",
  "dirs 6.0.0",
  "dotenvy",
+ "fs4",
  "futures",
  "hkdf",
  "http-body-util",
  "hyper",
  "hyper-util",
+ "mime_guess",
  "open",
  "pgvector",
  "postgres-types",
@@ -2206,12 +2324,13 @@ dependencies = [
  "refinery",
  "regex",
  "reqwest",
+ "rig-core",
  "rust_decimal",
  "rust_decimal_macros",
  "rustyline",
  "secrecy",
  "secret-service",
- "security-framework",
+ "security-framework 3.5.1",
  "serde",
  "serde_json",
  "sha2",
@@ -2495,6 +2614,16 @@ version = "0.3.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6877bb514081ee2a7ff5ef9de3281f14a4dd4bceac4c09388074a6b5df8a139a"
 
+[[package]]
+name = "mime_guess"
+version = "2.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f7c44f8e672c00fe5308fa235f821cb4198414e1c77935c1ab6948d3fd78550e"
+dependencies = [
+ "mime",
+ "unicase",
+]
+
 [[package]]
 name = "minimad"
 version = "0.14.0"
@@ -2504,6 +2633,12 @@ dependencies = [
  "once_cell",
 ]
 
+[[package]]
+name = "minimal-lexical"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "68354c5c6bd36d73ff3feceb05efa59b6acb7626617f4962be322a825e61f79a"
+
 [[package]]
 name = "mio"
 version = "1.1.1"
@@ -2516,6 +2651,32 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "nanoid"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3ffa00dec017b5b1a8b7cf5e2c008bfda1aa7e0697ac1508b491fdf2622fb4d8"
+dependencies = [
+ "rand 0.8.5",
+]
+
+[[package]]
+name = "native-tls"
+version = "0.2.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "87de3442987e9dbec73158d5c715e7ad9072fda936bb03d19d7fa10e00520f0e"
+dependencies = [
+ "libc",
+ "log",
+ "openssl",
+ "openssl-probe 0.1.6",
+ "openssl-sys",
+ "schannel",
+ "security-framework 2.11.1",
+ "security-framework-sys",
+ "tempfile",
+]
+
 [[package]]
 name = "nibble_vec"
 version = "0.1.0"
@@ -2550,6 +2711,16 @@ dependencies = [
  "libc",
 ]
 
+[[package]]
+name = "nom"
+version = "7.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d273983c5a657a70a3e8f2a01329822f3b8c8172b73826411a55751e404a0a4a"
+dependencies = [
+ "memchr",
+ "minimal-lexical",
+]
+
 [[package]]
 name = "nu-ansi-term"
 version = "0.50.3"
@@ -2698,18 +2869,71 @@ dependencies = [
  "pathdiff",
 ]
 
+[[package]]
+name = "openssl"
+version = "0.10.75"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "08838db121398ad17ab8531ce9de97b244589089e290a384c900cb9ff7434328"
+dependencies = [
+ "bitflags 2.10.0",
+ "cfg-if",
+ "foreign-types",
+ "libc",
+ "once_cell",
+ "openssl-macros",
+ "openssl-sys",
+]
+
+[[package]]
+name = "openssl-macros"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a948666b637a0f465e8564c73e89d4dde00d72d4d473cc972f390fc3dcee7d9c"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.114",
+]
+
+[[package]]
+name = "openssl-probe"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d05e27ee213611ffe7d6348b942e8f942b37114c00cc03cec254295a4a17852e"
+
 [[package]]
 name = "openssl-probe"
 version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7c87def4c32ab89d880effc9e097653c8da5d6ef28e6b539d313baaacfbafcbe"
 
+[[package]]
+name = "openssl-sys"
+version = "0.9.111"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "82cab2d520aa75e3c58898289429321eb788c3106963d0dc886ec7a5f4adc321"
+dependencies = [
+ "cc",
+ "libc",
+ "pkg-config",
+ "vcpkg",
+]
+
 [[package]]
 name = "option-ext"
 version = "0.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "04744f49eae99ab78e0d5c0b603ab218f515ea8cfe5a456d7629ad883a3b6e7d"
 
+[[package]]
+name = "ordered-float"
+version = "5.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7f4779c6901a562440c3786d08192c6fbda7c1c2060edd10006b05ee35d10f2d"
+dependencies = [
+ "num-traits",
+]
+
 [[package]]
 name = "ordered-stream"
 version = "0.2.0"
@@ -2821,6 +3045,26 @@ dependencies = [
  "siphasher",
 ]
 
+[[package]]
+name = "pin-project"
+version = "1.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "677f1add503faace112b9f1373e43e9e054bfdd22ff1a63c1bc485eaec6a6a8a"
+dependencies = [
+ "pin-project-internal",
+]
+
+[[package]]
+name = "pin-project-internal"
+version = "1.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6e918e4ff8c4549eb882f14b3a4bc8c8bc93de829416eacf579f1207a8fbf861"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "pin-project-lite"
 version = "0.2.16"
@@ -3353,16 +3597,22 @@ checksum = "eddd3ca559203180a307f12d114c268abf583f59b03cb906fd0b3ff8646c1147"
 dependencies = [
  "base64 0.22.1",
  "bytes",
+ "encoding_rs",
  "futures-core",
  "futures-util",
+ "h2",
  "http",
  "http-body",
  "http-body-util",
  "hyper",
  "hyper-rustls",
+ "hyper-tls",
  "hyper-util",
  "js-sys",
  "log",
+ "mime",
+ "mime_guess",
+ "native-tls",
  "percent-encoding",
  "pin-project-lite",
  "quinn",
@@ -3373,6 +3623,7 @@ dependencies = [
  "serde_urlencoded",
  "sync_wrapper",
  "tokio",
+ "tokio-native-tls",
  "tokio-rustls",
  "tokio-util",
  "tower",
@@ -3386,6 +3637,38 @@ dependencies = [
  "webpki-roots",
 ]
 
+[[package]]
+name = "rig-core"
+version = "0.30.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a8f7a3f0c7c00eaced15a68ee16e1bd6bb709ff598d11b9aedac8b628217dc09"
+dependencies = [
+ "as-any",
+ "async-stream",
+ "base64 0.22.1",
+ "bytes",
+ "eventsource-stream",
+ "fastrand",
+ "futures",
+ "futures-timer",
+ "glob",
+ "http",
+ "mime",
+ "mime_guess",
+ "nanoid",
+ "ordered-float",
+ "pin-project-lite",
+ "reqwest",
+ "schemars 1.2.1",
+ "serde",
+ "serde_json",
+ "thiserror 2.0.18",
+ "tokio",
+ "tracing",
+ "tracing-futures",
+ "url",
+]
+
 [[package]]
 name = "ring"
 version = "0.17.14"
@@ -3533,10 +3816,10 @@ version = "0.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "612460d5f7bea540c490b2b6395d8e34a953e52b491accd6c86c8164c5932a63"
 dependencies = [
- "openssl-probe",
+ "openssl-probe 0.2.1",
  "rustls-pki-types",
  "schannel",
- "security-framework",
+ "security-framework 3.5.1",
 ]
 
 [[package]]
@@ -3653,10 +3936,23 @@ checksum = "a2b42f36aa1cd011945615b92222f6bf73c599a102a300334cd7f8dbeec726cc"
 dependencies = [
  "dyn-clone",
  "ref-cast",
+ "schemars_derive",
  "serde",
  "serde_json",
 ]
 
+[[package]]
+name = "schemars_derive"
+version = "1.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7d115b50f4aaeea07e79c1912f645c7513d81715d0420f8bc77a18c6260b307f"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "serde_derive_internals",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "scopeguard"
 version = "1.2.0"
@@ -3698,6 +3994,19 @@ dependencies = [
  "zbus",
 ]
 
+[[package]]
+name = "security-framework"
+version = "2.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "897b2245f0b511c87893af39b033e5ca9cce68824c4d7e7630b5a1d339658d02"
+dependencies = [
+ "bitflags 2.10.0",
+ "core-foundation 0.9.4",
+ "core-foundation-sys",
+ "libc",
+ "security-framework-sys",
+]
+
 [[package]]
 name = "security-framework"
 version = "3.5.1"
@@ -3705,7 +4014,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b3297343eaf830f66ede390ea39da1d462b6b0c1b000f420d0a83f898bbbe6ef"
 dependencies = [
  "bitflags 2.10.0",
- "core-foundation",
+ "core-foundation 0.10.1",
  "core-foundation-sys",
  "libc",
  "security-framework-sys",
@@ -3761,6 +4070,17 @@ dependencies = [
  "syn 2.0.114",
 ]
 
+[[package]]
+name = "serde_derive_internals"
+version = "0.29.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "18d26a20a969b9e3fdf2fc2d9f21eda6c40e2de84c9408bb5d3b05d499aae711"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "serde_json"
 version = "1.0.149"
@@ -4074,6 +4394,27 @@ dependencies = [
  "syn 2.0.114",
 ]
 
+[[package]]
+name = "system-configuration"
+version = "0.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a13f3d0daba03132c0aa9767f98351b3488edc2c100cda2d2ec2b04f3d8d3c8b"
+dependencies = [
+ "bitflags 2.10.0",
+ "core-foundation 0.9.4",
+ "system-configuration-sys",
+]
+
+[[package]]
+name = "system-configuration-sys"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e1d1b10ced5ca923a1fcb8d03e96b8d3268065d724548c0211415ff6ac6bac4"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
 [[package]]
 name = "system-interface"
 version = "0.27.3"
@@ -4312,6 +4653,16 @@ dependencies = [
  "syn 2.0.114",
 ]
 
+[[package]]
+name = "tokio-native-tls"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bbae76ab933c85776efabc971569dd6119c580d8f5d448769dec1764bf796ef2"
+dependencies = [
+ "native-tls",
+ "tokio",
+]
+
 [[package]]
 name = "tokio-postgres"
 version = "0.7.16"
@@ -4574,6 +4925,18 @@ dependencies = [
  "valuable",
 ]
 
+[[package]]
+name = "tracing-futures"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "97d095ae15e245a057c8e8451bab9b3ee1e1f68e9ba2b4fbc18d0ac5237835f2"
+dependencies = [
+ "futures",
+ "futures-task",
+ "pin-project",
+ "tracing",
+]
+
 [[package]]
 name = "tracing-log"
 version = "0.2.0"
@@ -4673,6 +5036,12 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "unicase"
+version = "2.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dbc4bc3a9f746d862c45cb89d705aa10f187bb96c76001afab07a0d35ce60142"
+
 [[package]]
 name = "unicode-bidi"
 version = "0.3.18"
@@ -4795,6 +5164,12 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ba73ea9cf16a25df0c8caa16c51acb937d5712a8429db78a3ee29d5dcacd3a65"
 
+[[package]]
+name = "vcpkg"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "accd4ea62f7bb7a82fe23066fb0957d48ef677f6eeb8215f372f52e48bb32426"
+
 [[package]]
 name = "version_check"
 version = "0.9.5"
@@ -5469,6 +5844,17 @@ version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f0805222e57f7521d6a62e36fa9163bc891acd422f971defe97d64e70d0a4fe5"
 
+[[package]]
+name = "windows-registry"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "02752bf7fbdcce7f2a27a742f798510f3e5ad88dbe84871e5168e2120c3d5720"
+dependencies = [
+ "windows-link",
+ "windows-result",
+ "windows-strings",
+]
+
 [[package]]
 name = "windows-result"
 version = "0.4.1"

Cargo.toml

@@ -1,10 +1,19 @@
 [package]
 name = "ironclaw"
-version = "0.1.0"
+version = "0.1.2"
 edition = "2024"
 rust-version = "1.85"
 description = "Secure personal AI assistant that protects your data and expands its capabilities on the fly"
+authors = ["NEAR AI <[email protected]>"]
 license = "MIT OR Apache-2.0"
+homepage = "https://github.com/nearai/ironclaw"
+repository = "https://github.com/nearai/ironclaw"
+
+[package.metadata.wix]
+upgrade-guid = "D0156E61-BA37-451E-8AB9-1A2ECCCFA48F"
+path-guid = "F90B6EA6-87F7-499B-BB19-CF55DE1EB339"
+license = false
+eula = false
 
 [dependencies]
 # Async runtime
@@ -58,12 +67,16 @@ axum = { version = "0.8", features = ["ws"] }
 tower = "0.5"
 tower-http = { version = "0.6", features = ["trace", "cors"] }
 
+# Cron scheduling for routines
+cron = "0.13"
+
 # Safety/sanitization
 regex = "1"
 aho-corasick = "1"
 
 # Filesystem paths
 dirs = "6"
+fs4 = "0.6"
 
 # Secrecy for sensitive values
 secrecy = { version = "0.10", features = ["serde"] }
@@ -90,6 +103,9 @@ sha2 = "0.10"
 blake3 = "1"
 rand = "0.8"
 
+# Multi-provider LLM support
+rig-core = "0.30"
+
 # Docker sandbox
 bollard = "0.18"
 
@@ -99,6 +115,7 @@ hyper-util = { version = "0.1", features = ["server", "tokio", "http1", "http2"]
 http-body-util = "0.1"
 bytes = "1"
 base64 = "0.22.1"
+mime_guess = "2.0.5"
 
 # macOS keychain
 [target.'cfg(target_os = "macos")'.dependencies]
@@ -119,3 +136,44 @@ tempfile = "3"
 [features]
 default = []
 integration = []
+
+# The profile that 'cargo dist' will build with
+[profile.dist]
+inherits = "release"
+lto = "thin"
+
+# Config for 'dist'
+[workspace.metadata.dist]
+# The preferred dist version to use in CI (Cargo.toml SemVer syntax)
+cargo-dist-version = "0.30.3"
+# CI backends to support
+ci = "github"
+# The installers to generate for each app
+installers = ["shell", "powershell", "npm", "msi"]
+# Publish jobs to run in CI
+publish-jobs = ["npm"]
+# Target platforms to build apps for (Rust target-triple syntax)
+targets = [
+    "aarch64-apple-darwin",
+    "aarch64-unknown-linux-gnu",
+    "x86_64-apple-darwin",
+    "x86_64-unknown-linux-gnu",
+    "x86_64-pc-windows-msvc",
+]
+# The archive format to use for windows builds (defaults .zip)
+windows-archive = ".tar.gz"
+# The archive format to use for non-windows builds (defaults .tar.xz)
+unix-archive = ".tar.gz"
+# Which actions to run on pull requests
+pr-run-mode = "upload"
+# Path that installers should place binaries in
+install-path = "CARGO_HOME"
+# Whether to install an updater program
+install-updater = true
+
+[workspace.metadata.dist.github-custom-runners]
+aarch64-unknown-linux-gnu = "ubuntu-24.04-arm"
+x86_64-unknown-linux-gnu = "ubuntu-22.04"
+x86_64-pc-windows-msvc = "windows-2022"
+x86_64-apple-darwin = "macos-15-intel"
+aarch64-apple-darwin = "macos-14"

Dockerfile.worker

@@ -0,0 +1,63 @@
+# Multi-stage Dockerfile for the IronClaw worker container.
+#
+# This image runs the ironclaw binary in worker mode inside Docker containers.
+# The orchestrator creates instances of this image for sandboxed job execution.
+#
+# Build:
+#   docker build -f Dockerfile.worker -t ironclaw-worker .
+#
+# The image includes common development tools so workers can build software,
+# run tests, and execute shell commands.
+
+FROM rust:1.85-bookworm AS builder
+
+WORKDIR /build
+COPY . .
+
+# Build only the ironclaw binary (release mode)
+RUN cargo build --release --bin ironclaw
+
+# ---
+
+FROM debian:bookworm-slim
+
+# Install common development tools
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    ca-certificates \
+    curl \
+    git \
+    build-essential \
+    pkg-config \
+    libssl-dev \
+    nodejs \
+    npm \
+    python3 \
+    python3-pip \
+    python3-venv \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install Rust toolchain for the sandbox user
+ENV RUSTUP_HOME=/usr/local/rustup \
+    CARGO_HOME=/usr/local/cargo \
+    PATH=/usr/local/cargo/bin:$PATH
+RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --default-toolchain 1.85.0 \
+    && chmod -R a+r /usr/local/rustup /usr/local/cargo
+
+# Install Claude Code CLI (for claude-bridge mode)
+RUN npm install -g @anthropic-ai/claude-code@latest
+
+# Copy the binary
+COPY --from=builder /build/target/release/ironclaw /usr/local/bin/ironclaw
+
+# Create non-root user (UID 1000 matches the orchestrator's container config)
+RUN useradd -m -u 1000 -s /bin/bash sandbox \
+    && mkdir -p /workspace \
+    && chown sandbox:sandbox /workspace \
+    && mkdir -p /home/sandbox/.claude \
+    && chown sandbox:sandbox /home/sandbox/.claude
+
+USER sandbox
+WORKDIR /workspace
+
+# The orchestrator passes the full command via Docker cmd.
+ENTRYPOINT ["ironclaw"]

FEATURE_PARITY.md

@@ -16,8 +16,8 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 
 | Feature | OpenClaw | IronClaw | Notes |
 |---------|----------|----------|-------|
-| Hub-and-spoke architecture | ✅ | 🚧 | IronClaw has channels but no central gateway |
-| WebSocket control plane | ✅ | ❌ | Gateway with ws://127.0.0.1:18789 |
+| Hub-and-spoke architecture | ✅ | ✅ | Web gateway as central hub |
+| WebSocket control plane | ✅ | ✅ | Gateway with WebSocket + SSE |
 | Single-user system | ✅ | ✅ | |
 | Multi-agent routing | ✅ | ❌ | Workspace isolation per-agent |
 | Session-based messaging | ✅ | ✅ | Per-sender sessions |
@@ -31,9 +31,9 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 
 | Feature | OpenClaw | IronClaw | Notes |
 |---------|----------|----------|-------|
-| Gateway control plane | ✅ | ❌ | Central WebSocket server |
-| HTTP endpoints for Control UI | ✅ | ❌ | Web dashboard |
-| Channel connection lifecycle | ✅ | 🚧 | ChannelManager handles streams |
+| Gateway control plane | ✅ | ✅ | Web gateway with 40+ API endpoints |
+| HTTP endpoints for Control UI | ✅ | ✅ | Web dashboard with chat, memory, jobs, logs, extensions |
+| Channel connection lifecycle | ✅ | ✅ | ChannelManager + WebSocket tracker |
 | Session management/routing | ✅ | ✅ | SessionManager exists |
 | Configuration hot-reload | ✅ | ❌ | |
 | Network modes (loopback/LAN/remote) | ✅ | 🚧 | HTTP only |
@@ -43,7 +43,7 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 | launchd/systemd integration | ✅ | ❌ | |
 | Bonjour/mDNS discovery | ✅ | ❌ | |
 | Tailscale integration | ✅ | ❌ | |
-| Health check endpoints | ✅ | ❌ | |
+| Health check endpoints | ✅ | ✅ | /api/health + /api/gateway/status |
 | `doctor` diagnostics | ✅ | ❌ | |
 
 ### Owner: _Unassigned_
@@ -59,14 +59,14 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 | REPL (simple) | ✅ | ✅ | - | For testing |
 | WASM channels | ❌ | ✅ | - | IronClaw innovation |
 | WhatsApp | ✅ | ❌ | P1 | Baileys (Web) |
-| Telegram | ✅ | ❌ | P1 | grammY (Bot API) |
+| Telegram | ✅ | ✅ | - | WASM channel(MTProto), DM pairing, caption, /start, bot_username |
 | Discord | ✅ | ❌ | P2 | discord.js |
 | Signal | ✅ | ❌ | P2 | signal-cli |
-| Slack | ✅ | 🚧 | P1 | Stub exists, needs implementation |
+| Slack | ✅ | ✅ | - | WASM tool |
 | iMessage | ✅ | ❌ | P3 | BlueBubbles recommended |
 | Feishu/Lark | ✅ | ❌ | P3 | |
 | LINE | ✅ | ❌ | P3 | |
-| WebChat | ✅ | ❌ | P2 | Browser-based chat |
+| WebChat | ✅ | ✅ | - | Web gateway chat |
 | Matrix | ✅ | ❌ | P3 | E2EE support |
 | Mattermost | ✅ | ❌ | P3 | |
 | Google Chat | ✅ | ❌ | P3 | |
@@ -79,13 +79,13 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 
 | Feature | OpenClaw | IronClaw | Notes |
 |---------|----------|----------|-------|
-| DM pairing codes | ✅ | ❌ | Verification for unknown senders |
-| Allowlist/blocklist | ✅ | ❌ | Per-channel access control |
+| DM pairing codes | ✅ | ✅ | `ironclaw pairing list/approve`, host APIs |
+| Allowlist/blocklist | ✅ | 🚧 | allow_from + pairing store |
 | Self-message bypass | ✅ | ❌ | Own messages skip pairing |
-| Mention-based activation | ✅ | ❌ | Configurable patterns |
+| Mention-based activation | ✅ | ✅ | bot_username + respond_to_all_group_messages |
 | Per-group tool policies | ✅ | ❌ | Allow/deny specific tools |
 | Thread isolation | ✅ | ✅ | Separate sessions per thread |
-| Per-channel media limits | ✅ | ❌ | |
+| Per-channel media limits | ✅ | 🚧 | Caption support for media; no size limits |
 | Typing indicators | ✅ | 🚧 | TUI shows status |
 
 ### Owner: _Unassigned_
@@ -99,17 +99,17 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 | `run` (agent) | ✅ | ✅ | - | Default command |
 | `tool install/list/remove` | ✅ | ✅ | - | WASM tools |
 | `gateway start/stop` | ✅ | ❌ | P2 | |
-| `onboard` (wizard) | ✅ | ❌ | P2 | Interactive setup |
+| `onboard` (wizard) | ✅ | ✅ | - | Interactive setup |
 | `tui` | ✅ | ✅ | - | Ratatui TUI |
-| `config` | ✅ | ❌ | P2 | Read/write config |
+| `config` | ✅ | ✅ | - | Read/write config |
 | `channels` | ✅ | ❌ | P2 | Channel management |
 | `models` | ✅ | 🚧 | - | Model selector in TUI |
-| `status` | ✅ | ❌ | P2 | System status |
+| `status` | ✅ | ✅ | - | System status |
 | `agents` | ✅ | ❌ | P3 | Multi-agent management |
 | `sessions` | ✅ | ❌ | P3 | Session listing |
-| `memory` | ✅ | ❌ | P2 | Memory search CLI |
+| `memory` | ✅ | ✅ | - | Memory search CLI |
 | `skills` | ✅ | ❌ | P3 | Agent skills |
-| `pairing` | ✅ | ❌ | P3 | Node pairing |
+| `pairing` | ✅ | ✅ | - | list/approve for channel DM pairing |
 | `nodes` | ✅ | ❌ | P3 | Device management |
 | `plugins` | ✅ | ❌ | P3 | Plugin management |
 | `hooks` | ✅ | ❌ | P2 | Lifecycle hooks |
@@ -132,7 +132,7 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 | Feature | OpenClaw | IronClaw | Notes |
 |---------|----------|----------|-------|
 | Pi agent runtime | ✅ | ➖ | IronClaw uses custom runtime |
-| RPC-based execution | ✅ | 🚧 | Worker isolation |
+| RPC-based execution | ✅ | ✅ | Orchestrator/worker pattern |
 | Multi-provider failover | ✅ | ❌ | Provider fallback chains |
 | Per-sender sessions | ✅ | ✅ | |
 | Global sessions | ✅ | ❌ | Optional shared context |
@@ -303,13 +303,13 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 
 | Feature | OpenClaw | IronClaw | Priority | Notes |
 |---------|----------|----------|----------|-------|
-| Control UI Dashboard | ✅ | ❌ | P2 | Web status/config |
-| Channel status view | ✅ | ❌ | P2 | |
+| Control UI Dashboard | ✅ | ✅ | - | Web gateway with chat, memory, jobs, logs, extensions |
+| Channel status view | ✅ | 🚧 | P2 | Gateway status widget, full channel view pending |
 | Agent management | ✅ | ❌ | P3 | |
 | Model selection | ✅ | ✅ | - | TUI only |
 | Config editing | ✅ | ❌ | P3 | |
-| Debug/logs viewer | ✅ | ❌ | P3 | |
-| WebChat interface | ✅ | ❌ | P2 | Browser chat |
+| Debug/logs viewer | ✅ | ✅ | - | Real-time log streaming with level/target filters |
+| WebChat interface | ✅ | ✅ | - | Web gateway chat with SSE/WebSocket |
 | Canvas system (A2UI) | ✅ | ❌ | P3 | Agent-driven UI |
 
 ### Owner: _Unassigned_
@@ -320,13 +320,13 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 
 | Feature | OpenClaw | IronClaw | Priority | Notes |
 |---------|----------|----------|----------|-------|
-| Cron jobs | ✅ | ❌ | P2 | Schedule-based tasks |
-| Timezone support | ✅ | ❌ | P2 | |
-| One-shot/recurring jobs | ✅ | ❌ | P2 | |
+| Cron jobs | ✅ | ✅ | - | Routines with cron trigger |
+| Timezone support | ✅ | ✅ | - | Via cron expressions |
+| One-shot/recurring jobs | ✅ | ✅ | - | Manual + cron triggers |
 | `beforeInbound` hook | ✅ | ❌ | P2 | |
 | `beforeOutbound` hook | ✅ | ❌ | P2 | |
 | `beforeToolCall` hook | ✅ | ❌ | P2 | |
-| `onMessage` hook | ✅ | ❌ | P2 | |
+| `onMessage` hook | ✅ | ✅ | - | Routines with event trigger |
 | `onSessionStart` hook | ✅ | ❌ | P2 | |
 | `onSessionEnd` hook | ✅ | ❌ | P2 | |
 | `transcribeAudio` hook | ✅ | ❌ | P3 | |
@@ -346,18 +346,18 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 
 | Feature | OpenClaw | IronClaw | Notes |
 |---------|----------|----------|-------|
-| Gateway token auth | ✅ | 🚧 | HTTP webhook secret |
+| Gateway token auth | ✅ | ✅ | Bearer token auth on web gateway |
 | Device pairing | ✅ | ❌ | |
 | Tailscale identity | ✅ | ❌ | |
 | OAuth flows | ✅ | 🚧 | NEAR AI OAuth |
-| DM pairing verification | ✅ | ❌ | |
-| Allowlist/blocklist | ✅ | ❌ | |
+| DM pairing verification | ✅ | ✅ | ironclaw pairing approve, host APIs |
+| Allowlist/blocklist | ✅ | 🚧 | allow_from + pairing store |
 | Per-group tool policies | ✅ | ❌ | |
 | Exec approvals | ✅ | ✅ | TUI overlay |
 | TLS 1.3 minimum | ✅ | ✅ | reqwest rustls |
 | SSRF protection | ✅ | ✅ | WASM allowlist |
 | Loopback-first | ✅ | 🚧 | HTTP binds 0.0.0.0 |
-| Docker sandbox | ✅ | ❌ | Uses WASM sandbox |
+| Docker sandbox | ✅ | ✅ | Orchestrator/worker containers |
 | WASM sandbox | ❌ | ✅ | IronClaw innovation |
 | Tool policies | ✅ | ✅ | |
 | Elevated mode | ✅ | ❌ | |
@@ -397,6 +397,7 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 ### P0 - Core (Already Done)
 - ✅ TUI channel with approval overlays
 - ✅ HTTP webhook channel
+- ✅ DM pairing (ironclaw pairing list/approve, host APIs)
 - ✅ WASM tool sandbox
 - ✅ Workspace/memory with hybrid search
 - ✅ Prompt injection defense
@@ -404,23 +405,32 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 - ✅ Session management
 - ✅ Context compaction
 - ✅ Model selection
+- ✅ Gateway control plane + WebSocket
+- ✅ Web Control UI (chat, memory, jobs, logs, extensions, routines)
+- ✅ WebChat channel (web gateway)
+- ✅ Slack channel (WASM tool)
+- ✅ Telegram channel (WASM tool, MTProto)
+- ✅ Docker sandbox (orchestrator/worker)
+- ✅ Cron job scheduling (routines)
+- ✅ CLI subcommands (onboard, config, status, memory)
+- ✅ Gateway token auth
 
 ### P1 - High Priority
 - ❌ Slack channel (real implementation)
-- ❌ Telegram channel
+- ✅ Telegram channel (WASM, DM pairing, caption, /start)
 - ❌ WhatsApp channel
 - ❌ Multi-provider failover
-- ❌ Gateway control plane + WebSocket
 - ❌ Hooks system (beforeInbound, beforeToolCall, etc.)
 
 ### P2 - Medium Priority
 - ❌ Cron job scheduling
 - ❌ Web Control UI
 - ❌ WebChat channel
-- ❌ Media handling (images, PDFs)
+- 🚧 Media handling (caption support; no image/PDF processing)
 - ❌ CLI subcommands (config, status, memory, doctor)
 - ❌ Ollama/local model support
 - ❌ Configuration hot-reload
+- ❌ Webhook trigger endpoint in web gateway
 
 ### P3 - Lower Priority
 - ❌ Discord channel

README.md

@@ -43,7 +43,10 @@ IronClaw is the AI assistant you can actually trust with your personal and profe
 
 ### Always Available
 
-- **Multi-channel** - REPL, HTTP webhooks, and extensible WASM channels (Telegram, Slack, and more)
+- **Multi-channel** - REPL, HTTP webhooks, WASM channels (Telegram, Slack), and web gateway
+- **Docker Sandbox** - Isolated container execution with per-job tokens and orchestrator/worker pattern
+- **Web Gateway** - Browser UI with real-time SSE/WebSocket streaming
+- **Routines** - Cron schedules, event triggers, webhook handlers for background automation
 - **Heartbeat System** - Proactive background execution for monitoring and maintenance tasks
 - **Parallel Jobs** - Handle multiple requests concurrently with isolated contexts
 - **Self-repair** - Automatic detection and recovery of stuck operations
@@ -65,10 +68,57 @@ IronClaw is the AI assistant you can actually trust with your personal and profe
 ### Prerequisites
 
 - Rust 1.85+
-- PostgreSQL 15+ with pgvector extension
+- PostgreSQL 15+ with [pgvector](https://github.com/pgvector/pgvector) extension
 - NEAR AI account (authentication handled via setup wizard)
 
-### Build
+## Download or Build
+
+Visit [Releases page](https://github.com/nearai/ironclaw/releases/) to see the latest updates.
+
+<details>
+  <summary>Install via Windows Installer (Windows)</summary>
+
+Download the [Windows Installer](https://github.com/nearai/ironclaw/releases/latest/download/ironclaw-x86_64-pc-windows-msvc.msi) and run it.
+
+</details>
+
+<details>
+  <summary>Install via powershell script (Windows)</summary>
+
+```sh
+irm https://github.com/nearai/ironclaw/releases/latest/download/ironclaw-installer.ps1 | iex
+```
+
+</details>
+
+<details>
+  <summary>Install via shell script (macOS, Linux, Windows/WSL)</summary>
+
+```sh
+curl --proto '=https' --tlsv1.2 -LsSf https://github.com/nearai/ironclaw/releases/latest/download/ironclaw-installer.sh | sh
+```
+</details>
+
+<details>
+  <summary>Run via npx (Node.js on Windows, Linux, macOS)</summary>
+
+```sh
+npx ironclaw
+```
+</details>
+
+<details>
+  <summary>Use in package.json scripts (Node.js on Windows, Linux, macOS)</summary>
+
+```sh
+npm install ironclaw
+```
+</details>
+
+<details>
+  <summary>Compile the source code (Cargo on Windows, Linux, macOS)</summary>
+
+Install it with `cargo`, just make sure you have [Rust](https://rustup.rs) installed on your computer.
 
 ```bash
 # Clone the repository
@@ -82,6 +132,10 @@ cargo build --release
 cargo test
 ```
 
+For **full release** (after modifying channel sources), run `./scripts/build-all.sh` to rebuild channels first.
+
+</details>
+
 ### Database Setup
 
 ```bash
@@ -97,7 +151,7 @@ psql ironclaw -c "CREATE EXTENSION IF NOT EXISTS vector;"
 Run the setup wizard to configure IronClaw:
 
 ```bash
-ironclaw setup
+ironclaw onboard
 ```
 
 The wizard handles database connection, NEAR AI authentication (via browser OAuth),
@@ -143,37 +197,42 @@ External content passes through multiple security layers:
 ## Architecture
 
 ```
-┌─────────────────────────────────────────────────────────────────┐
-│                         Channels                                 │
-│  ┌──────┐  ┌──────┐  ┌──────────────┐                           │
-│  │ REPL │  │ HTTP │  │ WASM Channels│                           │
-│  └──┬───┘  └──┬───┘  └──────┬───────┘                           │
-│     └─────────┴─────────────┘                                    │
-│                         │                                        │
-│                    ┌────▼────┐                                  │
-│                    │  Router │  Intent classification           │
-│                    └────┬────┘                                  │
-│                         │                                        │
-│              ┌──────────▼──────────┐                            │
-│              │     Scheduler       │  Parallel job management   │
-│              └──────────┬──────────┘                            │
-│                         │                                        │
-│         ┌───────────────┼───────────────┐                       │
-│         ▼               ▼               ▼                       │
-│    ┌─────────┐    ┌─────────┐    ┌─────────┐                   │
-│    │ Worker  │    │ Worker  │    │ Worker  │  LLM reasoning    │
-│    └────┬────┘    └────┬────┘    └────┬────┘                   │
-│         └───────────────┼───────────────┘                       │
-│                         │                                        │
-│              ┌──────────▼──────────┐                            │
-│              │   Tool Registry     │                            │
-│              │  ┌───────────────┐  │                            │
-│              │  │ Built-in      │  │                            │
-│              │  │ MCP           │  │                            │
-│              │  │ WASM Sandbox  │  │                            │
-│              │  └───────────────┘  │                            │
-│              └─────────────────────┘                            │
-└─────────────────────────────────────────────────────────────────┘
+┌────────────────────────────────────────────────────────────────────┐
+│                          Channels                                  │
+│  ┌──────┐  ┌──────┐  ┌─────────────┐  ┌─────────────┐            │
+│  │ REPL │  │ HTTP │  │WASM Channels│  │ Web Gateway │            │
+│  └──┬───┘  └──┬───┘  └──────┬──────┘  │ (SSE + WS) │            │
+│     │         │              │         └──────┬──────┘            │
+│     └─────────┴──────────────┴────────────────┘                   │
+│                              │                                     │
+│                    ┌─────────▼─────────┐                          │
+│                    │    Agent Loop     │  Intent routing           │
+│                    └────┬─────────┬────┘                          │
+│                         │         │                                │
+│              ┌──────────▼───┐  ┌──▼──────────────┐               │
+│              │  Scheduler   │  │ Routines Engine  │               │
+│              │(parallel jobs)│  │(cron, event, wh) │               │
+│              └──────┬───────┘  └────────┬─────────┘               │
+│                     │                   │                          │
+│       ┌─────────────┼───────────────────┘                         │
+│       │             │                                              │
+│   ┌───▼────┐   ┌────▼────────────────┐                           │
+│   │ Local  │   │    Orchestrator     │                           │
+│   │Workers │   │  ┌───────────────┐  │                           │
+│   │(in-proc)│   │  │ Docker Sandbox│  │                           │
+│   └───┬────┘   │  │   Containers  │  │                           │
+│       │        │  │ ┌───────────┐ │  │                           │
+│       │        │  │ │Worker / CC│ │  │                           │
+│       │        │  │ └───────────┘ │  │                           │
+│       │        │  └───────────────┘  │                           │
+│       │        └─────────┬───────────┘                           │
+│       └──────────────────┤                                        │
+│                          │                                        │
+│              ┌───────────▼──────────┐                             │
+│              │    Tool Registry     │                             │
+│              │  Built-in, MCP, WASM │                             │
+│              └──────────────────────┘                             │
+└────────────────────────────────────────────────────────────────────┘
 ```
 
 ### Core Components
@@ -184,6 +243,9 @@ External content passes through multiple security layers:
 | **Router** | Classifies user intent (command, query, task) |
 | **Scheduler** | Manages parallel job execution with priorities |
 | **Worker** | Executes jobs with LLM reasoning and tool calls |
+| **Orchestrator** | Container lifecycle, LLM proxying, per-job auth |
+| **Web Gateway** | Browser UI with chat, memory, jobs, logs, extensions, routines |
+| **Routines Engine** | Scheduled (cron) and reactive (event, webhook) background tasks |
 | **Workspace** | Persistent memory with hybrid search |
 | **Safety Layer** | Prompt injection defense and content sanitization |
 
@@ -191,7 +253,7 @@ External content passes through multiple security layers:
 
 ```bash
 # First-time setup (configures database, auth, etc.)
-ironclaw setup
+ironclaw onboard
 
 # Start interactive REPL
 cargo run
@@ -210,12 +272,16 @@ cargo fmt
 cargo clippy --all --benches --tests --examples --all-features
 
 # Run tests
+createdb ironclaw_test
 cargo test
 
 # Run specific test
 cargo test test_name
 ```
 
+- **Telegram channel**: See [docs/TELEGRAM_SETUP.md](docs/TELEGRAM_SETUP.md) for setup and DM pairing.
+- **Changing channel sources**: Run `./channels-src/telegram/build.sh` before `cargo build` so the updated WASM is bundled.
+
 ## OpenClaw Heritage
 
 IronClaw is a Rust reimplementation inspired by [OpenClaw](https://github.com/openclaw/openclaw). See [FEATURE_PARITY.md](FEATURE_PARITY.md) for the complete tracking matrix.

build.rs

@@ -0,0 +1,106 @@
+//! Build script: compile Telegram channel WASM from source.
+//!
+//! Do not commit compiled WASM binaries — they are a supply chain risk.
+//! This script builds telegram.wasm from channels-src/telegram before the main crate compiles.
+//!
+//! Reproducible build:
+//!   cargo build --release
+//! (build.rs invokes the channel build automatically)
+//!
+//! Prerequisites: rustup target add wasm32-wasip2, cargo install wasm-tools
+
+use std::env;
+use std::path::PathBuf;
+use std::process::Command;
+
+fn main() {
+    let manifest_dir = env::var("CARGO_MANIFEST_DIR").unwrap();
+    let root = PathBuf::from(&manifest_dir);
+    let channel_dir = root.join("channels-src/telegram");
+    let wasm_out = channel_dir.join("telegram.wasm");
+
+    // Rerun when channel source or build script changes
+    println!("cargo:rerun-if-changed=channels-src/telegram/src");
+    println!("cargo:rerun-if-changed=channels-src/telegram/Cargo.toml");
+    println!("cargo:rerun-if-changed=wit/channel.wit");
+
+    if !channel_dir.is_dir() {
+        return;
+    }
+
+    // Build WASM module
+    let status = match Command::new("cargo")
+        .args([
+            "build",
+            "--release",
+            "--target",
+            "wasm32-wasip2",
+            "--manifest-path",
+            channel_dir.join("Cargo.toml").to_str().unwrap(),
+        ])
+        .current_dir(&root)
+        .status()
+    {
+        Ok(s) => s,
+        Err(_) => {
+            eprintln!(
+                "cargo:warning=Telegram channel build failed. Run: ./channels-src/telegram/build.sh"
+            );
+            return;
+        }
+    };
+
+    if !status.success() {
+        eprintln!(
+            "cargo:warning=Telegram channel build failed. Run: ./channels-src/telegram/build.sh"
+        );
+        return;
+    }
+
+    let raw_wasm = channel_dir.join("target/wasm32-wasip2/release/telegram_channel.wasm");
+    if !raw_wasm.exists() {
+        eprintln!(
+            "cargo:warning=Telegram WASM output not found at {:?}",
+            raw_wasm
+        );
+        return;
+    }
+
+    // Convert to component and strip (wasm-tools)
+    let component_ok = Command::new("wasm-tools")
+        .args([
+            "component",
+            "new",
+            raw_wasm.to_str().unwrap(),
+            "-o",
+            wasm_out.to_str().unwrap(),
+        ])
+        .current_dir(&root)
+        .status()
+        .map(|s| s.success())
+        .unwrap_or(false);
+
+    if !component_ok {
+        // Fallback: copy raw module if wasm-tools unavailable
+        if std::fs::copy(&raw_wasm, &wasm_out).is_err() {
+            eprintln!("cargo:warning=wasm-tools not found. Run: cargo install wasm-tools");
+        }
+    } else {
+        // Strip debug info (use temp file to avoid clobbering)
+        let stripped = wasm_out.with_extension("wasm.stripped");
+        let strip_ok = Command::new("wasm-tools")
+            .args([
+                "strip",
+                wasm_out.to_str().unwrap(),
+                "-o",
+                stripped.to_str().unwrap(),
+            ])
+            .current_dir(&root)
+            .status()
+            .map(|s| s.success())
+            .unwrap_or(false);
+        if strip_ok {
+            let _ = std::fs::rename(&stripped, &wasm_out);
+        }
+    }
+}

channels-src/slack/build.sh

@@ -30,7 +30,7 @@ if [ -f "$WASM_PATH" ]; then
     wasm-tools strip slack.wasm -o slack.wasm
 
     echo "Built: slack.wasm ($(du -h slack.wasm | cut -f1))"
-    echo "Copy slack.wasm and slack.capabilities.json to ~/.near-agent/channels/"
+    echo "Copy slack.wasm and slack.capabilities.json to ~/.ironclaw/channels/"
 else
     echo "Error: WASM output not found at $WASM_PATH"
     exit 1

channels-src/slack/src/lib.rs

@@ -108,7 +108,10 @@ struct SlackPostMessageResponse {
 #[derive(Debug, Deserialize)]
 struct SlackConfig {
     /// Name of secret containing signing secret (for verification by host).
+    /// Parsed from config for forward compatibility; not yet used in WASM
+    /// (host handles signature verification).
     #[serde(default = "default_signing_secret_name")]
+    #[allow(dead_code)]
     signing_secret_name: String,
 }
 
@@ -175,11 +178,7 @@ impl Guest for SlackChannel {
             // Actual event callback
             "event_callback" => {
                 if let Some(event) = event_wrapper.event {
-                    handle_slack_event(
-                        event,
-                        event_wrapper.team_id,
-                        event_wrapper.event_id,
-                    );
+                    handle_slack_event(event, event_wrapper.team_id, event_wrapper.event_id);
                 }
                 // Always respond 200 quickly to Slack (they have a 3s timeout)
                 json_response(200, serde_json::json!({"ok": true}))
@@ -230,6 +229,7 @@ impl Guest for SlackChannel {
             "https://slack.com/api/chat.postMessage",
             &headers.to_string(),
             Some(&payload_bytes),
+            None,
         );
 
         match result {
@@ -243,14 +243,15 @@ impl Guest for SlackChannel {
 
                 // Parse Slack response
                 let slack_response: SlackPostMessageResponse =
-                    serde_json::from_slice(&http_response.body).map_err(|e| {
-                        format!("Failed to parse Slack response: {}", e)
-                    })?;
+                    serde_json::from_slice(&http_response.body)
+                        .map_err(|e| format!("Failed to parse Slack response: {}", e))?;
 
                 if !slack_response.ok {
                     return Err(format!(
                         "Slack API error: {}",
-                        slack_response.error.unwrap_or_else(|| "unknown".to_string())
+                        slack_response
+                            .error
+                            .unwrap_or_else(|| "unknown".to_string())
                     ));
                 }
 
@@ -277,17 +278,16 @@ impl Guest for SlackChannel {
 }
 
 /// Handle a Slack event and emit message if applicable.
-fn handle_slack_event(
-    event: SlackEvent,
-    team_id: Option<String>,
-    _event_id: Option<String>,
-) {
+fn handle_slack_event(event: SlackEvent, team_id: Option<String>, _event_id: Option<String>) {
     match event.event_type.as_str() {
         // Direct mention of the bot
         "app_mention" => {
-            if let (Some(user), Some(channel), Some(text), Some(ts)) =
-                (event.user, event.channel.clone(), event.text, event.ts.clone())
-            {
+            if let (Some(user), Some(channel), Some(text), Some(ts)) = (
+                event.user,
+                event.channel.clone(),
+                event.text,
+                event.ts.clone(),
+            ) {
                 emit_message(user, text, channel, event.thread_ts.or(Some(ts)), team_id);
             }
         }
@@ -299,9 +299,12 @@ fn handle_slack_event(
                 return;
             }
 
-            if let (Some(user), Some(channel), Some(text), Some(ts)) =
-                (event.user, event.channel.clone(), event.text, event.ts.clone())
-            {
+            if let (Some(user), Some(channel), Some(text), Some(ts)) = (
+                event.user,
+                event.channel.clone(),
+                event.text,
+                event.ts.clone(),
+            ) {
                 // Only process DMs (channel IDs starting with D)
                 if channel.starts_with('D') {
                     emit_message(user, text, channel, event.thread_ts.or(Some(ts)), team_id);
@@ -335,8 +338,7 @@ fn emit_message(
         team_id,
     };
 
-    let metadata_json =
-        serde_json::to_string(&metadata).unwrap_or_else(|_| "{}".to_string());
+    let metadata_json = serde_json::to_string(&metadata).unwrap_or_else(|_| "{}".to_string());
 
     // Strip @ mentions of the bot from the text for cleaner messages
     let cleaned_text = strip_bot_mention(&text);

channels-src/telegram/build.sh

@@ -32,8 +32,8 @@ if [ -f "$WASM_PATH" ]; then
     echo "Built: telegram.wasm ($(du -h telegram.wasm | cut -f1))"
     echo ""
     echo "To install:"
-    echo "  mkdir -p ~/.near-agent/channels"
-    echo "  cp telegram.wasm telegram.capabilities.json ~/.near-agent/channels/"
+    echo "  mkdir -p ~/.ironclaw/channels"
+    echo "  cp telegram.wasm telegram.capabilities.json ~/.ironclaw/channels/"
     echo ""
     echo "Then add your bot token to secrets:"
     echo "  # Set TELEGRAM_BOT_TOKEN in your environment or secrets store"

channels-src/telegram/src/lib.rs

@@ -72,6 +72,10 @@ struct TelegramMessage {
     /// Message text.
     text: Option<String>,
 
+    /// Caption for media (photo, video, document, etc.).
+    #[serde(default)]
+    caption: Option<String>,
+
     /// Original message if this is a reply.
     reply_to_message: Option<Box<TelegramMessage>>,
 
@@ -160,6 +164,21 @@ const POLLING_STATE_PATH: &str = "state/last_update_id";
 /// Workspace path for persisting owner_id across WASM callbacks.
 const OWNER_ID_PATH: &str = "state/owner_id";
 
+/// Workspace path for persisting dm_policy across WASM callbacks.
+const DM_POLICY_PATH: &str = "state/dm_policy";
+
+/// Workspace path for persisting allow_from (JSON array) across WASM callbacks.
+const ALLOW_FROM_PATH: &str = "state/allow_from";
+
+/// Channel name for pairing store (used by pairing host APIs).
+const CHANNEL_NAME: &str = "telegram";
+
+/// Workspace path for persisting bot_username for mention detection in groups.
+const BOT_USERNAME_PATH: &str = "state/bot_username";
+
+/// Workspace path for persisting respond_to_all_group_messages flag.
+const RESPOND_TO_ALL_GROUP_PATH: &str = "state/respond_to_all_group_messages";
+
 // ============================================================================
 // Channel Metadata
 // ============================================================================
@@ -196,6 +215,14 @@ struct TelegramConfig {
     #[serde(default)]
     owner_id: Option<i64>,
 
+    /// DM policy: "pairing" (default), "allowlist", or "open".
+    #[serde(default)]
+    dm_policy: Option<String>,
+
+    /// Allowed sender IDs/usernames from config (merged with pairing-approved store).
+    #[serde(default)]
+    allow_from: Option<Vec<String>>,
+
     /// Whether to respond to all group messages (not just mentions).
     #[serde(default)]
     respond_to_all_group_messages: bool,
@@ -257,6 +284,28 @@ impl Guest for TelegramChannel {
             );
         }
 
+        // Persist dm_policy and allow_from for DM pairing in handle_message
+        let dm_policy = config
+            .dm_policy
+            .as_deref()
+            .unwrap_or("pairing")
+            .to_string();
+        let _ = channel_host::workspace_write(DM_POLICY_PATH, &dm_policy);
+
+        let allow_from_json = serde_json::to_string(&config.allow_from.unwrap_or_default())
+            .unwrap_or_else(|_| "[]".to_string());
+        let _ = channel_host::workspace_write(ALLOW_FROM_PATH, &allow_from_json);
+
+        // Persist bot_username and respond_to_all_group_messages for group handling
+        let _ = channel_host::workspace_write(
+            BOT_USERNAME_PATH,
+            &config.bot_username.unwrap_or_default(),
+        );
+        let _ = channel_host::workspace_write(
+            RESPOND_TO_ALL_GROUP_PATH,
+            &config.respond_to_all_group_messages.to_string(),
+        );
+
         // Mode is determined by whether the host injected a tunnel_url
         // If tunnel is configured, use webhooks. Otherwise, use polling.
         let webhook_mode = config.tunnel_url.is_some();
@@ -388,7 +437,9 @@ impl Guest for TelegramChannel {
 
         let headers = serde_json::json!({});
 
-        let result = channel_host::http_request("GET", &url, &headers.to_string(), None);
+        // 35s HTTP timeout outlives Telegram's 30s server-side long-poll
+        let result =
+            channel_host::http_request("GET", &url, &headers.to_string(), None, Some(35_000));
 
         match result {
             Ok(response) => {
@@ -461,72 +512,52 @@ impl Guest for TelegramChannel {
     }
 
     fn on_respond(response: AgentResponse) -> Result<(), String> {
-        // Parse metadata to get chat info
         let metadata: TelegramMessageMetadata = serde_json::from_str(&response.metadata_json)
             .map_err(|e| format!("Failed to parse metadata: {}", e))?;
 
-        // Build sendMessage payload
-        let mut payload = serde_json::json!({
-            "chat_id": metadata.chat_id,
-            "text": response.content,
-            "parse_mode": "Markdown",
-        });
-
-        // Reply to the original message for context
-        payload["reply_to_message_id"] = serde_json::Value::Number(metadata.message_id.into());
-
-        let payload_bytes = serde_json::to_vec(&payload)
-            .map_err(|e| format!("Failed to serialize payload: {}", e))?;
-
-        // Make HTTP request to Telegram API
-        // The bot token is injected into the URL by the host
-        let headers = serde_json::json!({
-            "Content-Type": "application/json"
-        });
-
-        let result = channel_host::http_request(
-            "POST",
-            "https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendMessage",
-            &headers.to_string(),
-            Some(&payload_bytes),
+        // Try sending with Markdown first; fall back to plain text if Telegram
+        // can't parse the entities (e.g. model leaked <tool_call> with underscores).
+        let result = send_message(
+            metadata.chat_id,
+            &response.content,
+            metadata.message_id,
+            Some("Markdown"),
         );
 
         match result {
-            Ok(http_response) => {
-                if http_response.status != 200 {
-                    let body_str = String::from_utf8_lossy(&http_response.body);
-                    return Err(format!(
-                        "Telegram API returned status {}: {}",
-                        http_response.status, body_str
-                    ));
-                }
-
-                // Parse Telegram response
-                let api_response: TelegramApiResponse<SentMessage> =
-                    serde_json::from_slice(&http_response.body)
-                        .map_err(|e| format!("Failed to parse Telegram response: {}", e))?;
-
-                if !api_response.ok {
-                    return Err(format!(
-                        "Telegram API error: {}",
-                        api_response
-                            .description
-                            .unwrap_or_else(|| "unknown".to_string())
-                    ));
-                }
-
+            Ok(msg_id) => {
                 channel_host::log(
                     channel_host::LogLevel::Debug,
                     &format!(
                         "Sent message to chat {}: message_id={}",
-                        metadata.chat_id,
-                        api_response.result.map(|r| r.message_id).unwrap_or(0)
+                        metadata.chat_id, msg_id
                     ),
                 );
-
                 Ok(())
             }
-            Err(e) => Err(format!("HTTP request failed: {}", e)),
+            Err(SendError::ParseEntities(detail)) => {
+                channel_host::log(
+                    channel_host::LogLevel::Warn,
+                    &format!("Markdown parse failed ({}), retrying as plain text", detail),
+                );
+                let msg_id = send_message(
+                    metadata.chat_id,
+                    &response.content,
+                    metadata.message_id,
+                    None,
+                )
+                .map_err(|e| format!("Plain-text retry also failed: {}", e))?;
+
+                channel_host::log(
+                    channel_host::LogLevel::Debug,
+                    &format!(
+                        "Sent plain-text message to chat {}: message_id={}",
+                        metadata.chat_id, msg_id
+                    ),
+                );
+                Ok(())
+            }
+            Err(e) => Err(e.to_string()),
         }
     }
 
@@ -568,6 +599,7 @@ impl Guest for TelegramChannel {
             "https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendChatAction",
             &headers.to_string(),
             Some(&payload_bytes),
+            None,
         );
 
         if let Err(e) = result {
@@ -586,6 +618,101 @@ impl Guest for TelegramChannel {
     }
 }
 
+// ============================================================================
+// Send Message Helper
+// ============================================================================
+
+/// Errors from send_message, split so callers can match on parse-entity failures.
+enum SendError {
+    /// Telegram returned 400 with "can't parse entities" (Markdown issue).
+    ParseEntities(String),
+    /// Any other failure.
+    Other(String),
+}
+
+impl std::fmt::Display for SendError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            SendError::ParseEntities(detail) => write!(f, "parse entities error: {}", detail),
+            SendError::Other(msg) => write!(f, "{}", msg),
+        }
+    }
+}
+
+/// Send a message via the Telegram Bot API.
+///
+/// Returns the sent message_id on success. When `parse_mode` is set and
+/// Telegram returns a 400 "can't parse entities" error, returns
+/// `SendError::ParseEntities` so the caller can retry without formatting.
+fn send_message(
+    chat_id: i64,
+    text: &str,
+    reply_to_message_id: i64,
+    parse_mode: Option<&str>,
+) -> Result<i64, SendError> {
+    let mut payload = serde_json::json!({
+        "chat_id": chat_id,
+        "text": text,
+        "reply_to_message_id": reply_to_message_id,
+    });
+
+    if let Some(mode) = parse_mode {
+        payload["parse_mode"] = serde_json::Value::String(mode.to_string());
+    }
+
+    let payload_bytes = serde_json::to_vec(&payload)
+        .map_err(|e| SendError::Other(format!("Failed to serialize payload: {}", e)))?;
+
+    let headers = serde_json::json!({ "Content-Type": "application/json" });
+
+    let result = channel_host::http_request(
+        "POST",
+        "https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendMessage",
+        &headers.to_string(),
+        Some(&payload_bytes),
+        None,
+    );
+
+    match result {
+        Ok(http_response) => {
+            if http_response.status == 400 {
+                let body_str = String::from_utf8_lossy(&http_response.body);
+                if body_str.contains("can't parse entities") {
+                    return Err(SendError::ParseEntities(body_str.to_string()));
+                }
+                return Err(SendError::Other(format!(
+                    "Telegram API returned 400: {}",
+                    body_str
+                )));
+            }
+
+            if http_response.status != 200 {
+                let body_str = String::from_utf8_lossy(&http_response.body);
+                return Err(SendError::Other(format!(
+                    "Telegram API returned status {}: {}",
+                    http_response.status, body_str
+                )));
+            }
+
+            let api_response: TelegramApiResponse<SentMessage> =
+                serde_json::from_slice(&http_response.body)
+                    .map_err(|e| SendError::Other(format!("Failed to parse response: {}", e)))?;
+
+            if !api_response.ok {
+                return Err(SendError::Other(format!(
+                    "Telegram API error: {}",
+                    api_response
+                        .description
+                        .unwrap_or_else(|| "unknown".to_string())
+                )));
+            }
+
+            Ok(api_response.result.map(|r| r.message_id).unwrap_or(0))
+        }
+        Err(e) => Err(SendError::Other(format!("HTTP request failed: {}", e))),
+    }
+}
+
 // ============================================================================
 // Webhook Management
 // ============================================================================
@@ -604,6 +731,7 @@ fn delete_webhook() -> Result<(), String> {
         "https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/deleteWebhook",
         &headers.to_string(),
         None,
+        None,
     );
 
     match result {
@@ -666,6 +794,7 @@ fn register_webhook(tunnel_url: &str, webhook_secret: Option<&str>) -> Result<()
         "https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/setWebhook",
         &headers.to_string(),
         Some(&body_bytes),
+        None,
     );
 
     match result {
@@ -700,6 +829,47 @@ fn register_webhook(tunnel_url: &str, webhook_secret: Option<&str>) -> Result<()
     }
 }
 
+// ============================================================================
+// Pairing Reply
+// ============================================================================
+
+/// Send a pairing code message to a chat. Used when an unknown user DMs the bot.
+fn send_pairing_reply(chat_id: i64, code: &str) -> Result<(), String> {
+    let payload = serde_json::json!({
+        "chat_id": chat_id,
+        "text": format!(
+            "To pair with this bot, run: `ironclaw pairing approve telegram {}`",
+            code
+        ),
+        "parse_mode": "Markdown",
+    });
+
+    let payload_bytes = serde_json::to_vec(&payload)
+        .map_err(|e| format!("Failed to serialize payload: {}", e))?;
+
+    let headers = serde_json::json!({
+        "Content-Type": "application/json"
+    });
+
+    let result = channel_host::http_request(
+        "POST",
+        "https://api.telegram.org/bot{TELEGRAM_BOT_TOKEN}/sendMessage",
+        &headers.to_string(),
+        Some(&payload_bytes),
+    );
+
+    match result {
+        Ok(response) => {
+            if response.status != 200 {
+                let body_str = String::from_utf8_lossy(&response.body);
+                return Err(format!("HTTP {}: {}", response.status, body_str));
+            }
+            Ok(())
+        }
+        Err(e) => Err(format!("HTTP request failed: {}", e)),
+    }
+}
+
 // ============================================================================
 // Update Handling
 // ============================================================================
@@ -719,11 +889,16 @@ fn handle_update(update: TelegramUpdate) {
 
 /// Process a single message.
 fn handle_message(message: TelegramMessage) {
-    // Skip messages without text
-    let text = match message.text {
-        Some(t) if !t.is_empty() => t,
-        _ => return,
-    };
+    // Use text or caption (for media messages)
+    let content = message
+        .text
+        .filter(|t| !t.is_empty())
+        .or_else(|| message.caption.filter(|c| !c.is_empty()))
+        .unwrap_or_default();
+
+    if content.is_empty() {
+        return;
+    }
 
     // Skip messages without a sender (channel posts)
     let from = match message.from {
@@ -736,41 +911,111 @@ fn handle_message(message: TelegramMessage) {
         return;
     }
 
-    // Owner validation: silently drop messages from non-owner users
-    if let Some(owner_id_str) = channel_host::workspace_read(OWNER_ID_PATH) {
-        if !owner_id_str.is_empty() {
-            if let Ok(owner_id) = owner_id_str.parse::<i64>() {
-                if from.id != owner_id {
-                    channel_host::log(
-                        channel_host::LogLevel::Debug,
-                        &format!(
-                            "Dropping message from non-owner user {} (owner: {})",
-                            from.id, owner_id
-                        ),
-                    );
-                    return;
+    let is_private = message.chat.chat_type == "private";
+
+    // Owner validation: when owner_id is set, only that user can message
+    let owner_configured = channel_host::workspace_read(OWNER_ID_PATH)
+        .map(|s| !s.is_empty())
+        .unwrap_or(false);
+
+    if owner_configured {
+        if let Ok(owner_id) = channel_host::workspace_read(OWNER_ID_PATH)
+            .unwrap()
+            .parse::<i64>()
+        {
+            if from.id != owner_id {
+                channel_host::log(
+                    channel_host::LogLevel::Debug,
+                    &format!(
+                        "Dropping message from non-owner user {} (owner: {})",
+                        from.id, owner_id
+                    ),
+                );
+                return;
+            }
+        }
+    } else if is_private {
+        // No owner_id: apply dm_policy for private chats
+        let dm_policy = channel_host::workspace_read(DM_POLICY_PATH)
+            .unwrap_or_else(|| "pairing".to_string());
+
+        if dm_policy != "open" {
+            // Build effective allow list: config allow_from + pairing store
+            let mut allowed: Vec<String> = channel_host::workspace_read(ALLOW_FROM_PATH)
+                .and_then(|s| serde_json::from_str(&s).ok())
+                .unwrap_or_default();
+
+            if let Ok(store_allowed) = channel_host::pairing_read_allow_from(CHANNEL_NAME) {
+                allowed.extend(store_allowed);
+            }
+
+            let id_str = from.id.to_string();
+            let username_opt = from.username.as_deref();
+            let is_allowed = allowed.contains(&"*".to_string())
+                || allowed.contains(&id_str)
+                || username_opt.map_or(false, |u| allowed.contains(&u.to_string()));
+
+            if !is_allowed {
+                if dm_policy == "pairing" {
+                    // Upsert pairing request and send reply
+                    let meta = serde_json::json!({
+                        "chat_id": message.chat.id,
+                        "user_id": from.id,
+                        "username": username_opt,
+                    })
+                    .to_string();
+
+                    match channel_host::pairing_upsert_request(CHANNEL_NAME, &id_str, &meta) {
+                        Ok(result) => {
+                            channel_host::log(
+                                channel_host::LogLevel::Info,
+                                &format!(
+                                    "Pairing request for user {} (chat {}): code {}",
+                                    from.id, message.chat.id, result.code
+                                ),
+                            );
+                            if result.created {
+                                let _ = send_pairing_reply(message.chat.id, &result.code);
+                            }
+                        }
+                        Err(e) => {
+                            channel_host::log(
+                                channel_host::LogLevel::Error,
+                                &format!("Pairing upsert failed: {}", e),
+                            );
+                        }
+                    }
                 }
+                return;
             }
         }
     }
 
-    let is_private = message.chat.chat_type == "private";
-
-    // For group chats, check if the bot was mentioned
-    // TODO: Read bot_username from config and check mentions
-    // For now, process all messages in private chats and groups
+    // For group chats, only respond if bot was mentioned or respond_to_all is enabled
     if !is_private {
-        // In groups, only respond if there's a bot mention or command
-        // This is a simplified check - proper implementation would use entities
-        let has_command = text.starts_with('/');
-        let has_mention = text.contains('@');
+        let respond_to_all = channel_host::workspace_read(RESPOND_TO_ALL_GROUP_PATH)
+            .as_deref()
+            .unwrap_or("false")
+            == "true";
 
-        if !has_command && !has_mention {
-            channel_host::log(
-                channel_host::LogLevel::Debug,
-                &format!("Ignoring group message without mention: {}", text),
-            );
-            return;
+        if !respond_to_all {
+            let has_command = content.starts_with('/');
+            let bot_username = channel_host::workspace_read(BOT_USERNAME_PATH)
+                .unwrap_or_default();
+            let has_bot_mention = if bot_username.is_empty() {
+                content.contains('@')
+            } else {
+                let mention = format!("@{}", bot_username);
+                content.to_lowercase().contains(&mention.to_lowercase())
+            };
+
+            if !has_command && !has_bot_mention {
+                channel_host::log(
+                    channel_host::LogLevel::Debug,
+                    &format!("Ignoring group message without mention: {}", content),
+                );
+                return;
+            }
         }
     }
 
@@ -792,17 +1037,30 @@ fn handle_message(message: TelegramMessage) {
     let metadata_json = serde_json::to_string(&metadata).unwrap_or_else(|_| "{}".to_string());
 
     // Clean the message text (strip bot mentions and commands)
-    let cleaned_text = clean_message_text(&text);
+    let bot_username = channel_host::workspace_read(BOT_USERNAME_PATH).unwrap_or_default();
+    let cleaned_text = clean_message_text(
+        &content,
+        if bot_username.is_empty() {
+            None
+        } else {
+            Some(bot_username.as_str())
+        },
+    );
 
-    if cleaned_text.is_empty() {
+    // For /start with no args, emit placeholder so agent can respond with welcome
+    let content_to_emit = if cleaned_text.is_empty() && content.trim().starts_with('/') {
+        "[User started the bot]".to_string()
+    } else if cleaned_text.is_empty() {
         return;
-    }
+    } else {
+        cleaned_text
+    };
 
     // Emit the message to the agent
     channel_host::emit_message(&EmittedMessage {
         user_id: from.id.to_string(),
         user_name: Some(user_name),
-        content: cleaned_text,
+        content: content_to_emit,
         thread_id: None, // Telegram doesn't have threads in the same way
         metadata_json,
     });
@@ -817,7 +1075,8 @@ fn handle_message(message: TelegramMessage) {
 }
 
 /// Clean message text by removing bot commands and @mentions at the start.
-fn clean_message_text(text: &str) -> String {
+/// When bot_username is set, only strips that specific mention; otherwise strips any leading @mention.
+fn clean_message_text(text: &str, bot_username: Option<&str>) -> String {
     let mut result = text.trim().to_string();
 
     // Remove leading /command
@@ -832,11 +1091,30 @@ fn clean_message_text(text: &str) -> String {
 
     // Remove leading @mention
     if result.starts_with('@') {
-        if let Some(space_idx) = result.find(' ') {
-            result = result[space_idx..].trim_start().to_string();
+        if let Some(bot) = bot_username {
+            let mention = format!("@{}", bot);
+            let mention_lower = mention.to_lowercase();
+            let result_lower = result.to_lowercase();
+            if result_lower.starts_with(&mention_lower) {
+                let rest = result[mention.len()..].trim_start();
+                if rest.is_empty() {
+                    return String::new();
+                }
+                result = rest.to_string();
+            } else if let Some(space_idx) = result.find(' ') {
+                // Different leading @mention - only strip if it's the bot
+                let first_word = &result[..space_idx];
+                if first_word.eq_ignore_ascii_case(&mention) {
+                    result = result[space_idx..].trim_start().to_string();
+                }
+            }
         } else {
-            // Just a mention with no text
-            return String::new();
+            // No bot_username: strip any leading @mention
+            if let Some(space_idx) = result.find(' ') {
+                result = result[space_idx..].trim_start().to_string();
+            } else {
+                return String::new();
+            }
         }
     }
 
@@ -872,12 +1150,22 @@ mod tests {
 
     #[test]
     fn test_clean_message_text() {
-        assert_eq!(clean_message_text("/start hello"), "hello");
-        assert_eq!(clean_message_text("@bot hello world"), "hello world");
-        assert_eq!(clean_message_text("/start"), "");
-        assert_eq!(clean_message_text("@botname"), "");
-        assert_eq!(clean_message_text("just text"), "just text");
-        assert_eq!(clean_message_text("  spaced  "), "spaced");
+        // Without bot_username: strips any leading @mention
+        assert_eq!(clean_message_text("/start hello", None), "hello");
+        assert_eq!(clean_message_text("@bot hello world", None), "hello world");
+        assert_eq!(clean_message_text("/start", None), "");
+        assert_eq!(clean_message_text("@botname", None), "");
+        assert_eq!(clean_message_text("just text", None), "just text");
+        assert_eq!(clean_message_text("  spaced  ", None), "spaced");
+
+        // With bot_username: only strips @MyBot, not @alice
+        assert_eq!(clean_message_text("@MyBot hello", Some("MyBot")), "hello");
+        assert_eq!(clean_message_text("@mybot hi", Some("MyBot")), "hi");
+        assert_eq!(
+            clean_message_text("@alice hello", Some("MyBot")),
+            "@alice hello"
+        );
+        assert_eq!(clean_message_text("@MyBot", Some("MyBot")), "");
     }
 
     #[test]
@@ -945,4 +1233,17 @@ mod tests {
         assert_eq!(from.id, 789);
         assert_eq!(from.first_name, "John");
     }
+
+    #[test]
+    fn test_parse_message_with_caption() {
+        let json = r#"{
+            "message_id": 1,
+            "from": {"id": 1, "is_bot": false, "first_name": "A"},
+            "chat": {"id": 1, "type": "private"},
+            "caption": "What's in this image?"
+        }"#;
+        let msg: TelegramMessage = serde_json::from_str(json).unwrap();
+        assert_eq!(msg.text, None);
+        assert_eq!(msg.caption.as_deref(), Some("What's in this image?"));
+    }
 }

channels-src/telegram/telegram.capabilities.json

@@ -1,43 +1 @@
-{
-  "type": "channel",
-  "name": "telegram",
-  "description": "Telegram Bot API channel for receiving and responding to Telegram messages",
-  "capabilities": {
-    "http": {
-      "allowlist": [
-        { "host": "api.telegram.org", "path_prefix": "/bot" }
-      ],
-      "credentials": {
-        "telegram_bot": {
-          "secret_name": "telegram_bot_token",
-          "location": { "type": "url_path", "placeholder": "{TELEGRAM_BOT_TOKEN}" },
-          "host_patterns": ["api.telegram.org"]
-        }
-      },
-      "rate_limit": {
-        "requests_per_minute": 30,
-        "requests_per_hour": 1000
-      }
-    },
-    "secrets": {
-      "allowed_names": ["telegram_*"]
-    },
-    "channel": {
-      "allowed_paths": ["/webhook/telegram"],
-      "allow_polling": true,
-      "min_poll_interval_ms": 30000,
-      "workspace_prefix": "channels/telegram/",
-      "emit_rate_limit": {
-        "messages_per_minute": 100,
-        "messages_per_hour": 5000
-      }
-    }
-  },
-  "config": {
-    "bot_username": null,
-    "owner_id": null,
-    "respond_to_all_group_messages": false,
-    "polling_enabled": false,
-    "poll_interval_ms": 30000
-  }
-}
+{"type":"channel","name":"telegram","description":"Telegram Bot API channel for receiving and responding to Telegram messages","capabilities":{"http":{"allowlist":[{"host":"api.telegram.org","path_prefix":"/bot"}],"credentials":{"telegram_bot":{"secret_name":"telegram_bot_token","location":{"type":"url_path","placeholder":"{TELEGRAM_BOT_TOKEN}"},"host_patterns":["api.telegram.org"]}},"rate_limit":{"requests_per_minute":30,"requests_per_hour":1000}},"secrets":{"allowed_names":["telegram_*"]},"channel":{"allowed_paths":["/webhook/telegram"],"allow_polling":true,"min_poll_interval_ms":30000,"workspace_prefix":"channels/telegram/","emit_rate_limit":{"messages_per_minute":100,"messages_per_hour":5000}}},"config":{"bot_username":null,"owner_id":null,"respond_to_all_group_messages":false,"polling_enabled":false,"poll_interval_ms":30000,"dm_policy":"pairing","allow_from":[]}}

channels-src/whatsapp/src/lib.rs

@@ -361,6 +361,7 @@ impl Guest for WhatsAppChannel {
             &api_url,
             &headers.to_string(),
             Some(&payload_bytes),
+            None,
         );
 
         match result {

docs/BUILDING_CHANNELS.md

@@ -246,13 +246,46 @@ Create `my-channel.capabilities.json`:
 
 ## Building and Deploying
 
+### Supply Chain Security: No Committed Binaries
+
+**Do not commit compiled WASM binaries.** They are a supply chain risk — the binary in a PR may not match the source. IronClaw builds channels from source:
+
+- `cargo build` automatically builds `telegram.wasm` via `build.rs`
+- The built binary is in `.gitignore` and is not committed
+- CI should run `cargo build` (or `./scripts/build-all.sh`) to produce releases
+
+**Reproducible build:**
+```bash
+cargo build --release
+```
+
+Prerequisites: `rustup target add wasm32-wasip2`, `cargo install wasm-tools` (optional; fallback copies raw WASM if unavailable).
+
+### Telegram Channel (Manual Build)
+
+```bash
+# Add WASM target if needed
+rustup target add wasm32-wasip2
+
+# Build Telegram channel
+./channels-src/telegram/build.sh
+
+# Install (or use ironclaw onboard to install bundled channel)
+mkdir -p ~/.ironclaw/channels
+cp channels-src/telegram/telegram.wasm channels-src/telegram/telegram.capabilities.json ~/.ironclaw/channels/
+```
+
+**Note**: The main IronClaw binary bundles `telegram.wasm` via `include_bytes!`. When modifying the Telegram channel source, run `./channels-src/telegram/build.sh` **before** building the main crate, so the updated WASM is included.
+
+### Other Channels
+
 ```bash
 # Build the WASM component
-cd channels/my-channel
-cargo component build --release
+cd channels-src/my-channel
+cargo build --release --target wasm32-wasip2
 
 # Deploy to ~/.ironclaw/channels/
-cp target/wasm32-wasip1/release/my_channel.wasm ~/.ironclaw/channels/my-channel.wasm
+cp target/wasm32-wasip2/release/my_channel.wasm ~/.ironclaw/channels/my-channel.wasm
 cp my-channel.capabilities.json ~/.ironclaw/channels/
 ```
 

docs/TELEGRAM_SETUP.md

@@ -0,0 +1,135 @@
+# Telegram Channel Setup
+
+This guide covers configuring the Telegram channel for IronClaw, including DM pairing for access control.
+
+## Overview
+
+The Telegram channel lets you interact with IronClaw via Telegram DMs and groups. It supports:
+
+- **Webhook mode** (recommended): Instant delivery via tunnel
+- **Polling mode**: No tunnel required; ~30s delay
+- **DM pairing**: Approve unknown users before they can message the agent
+- **Group mentions**: `@YourBot` or `/command` to trigger in groups
+
+## Prerequisites
+
+- IronClaw installed and configured (`ironclaw onboard`)
+- A Telegram bot token from [@BotFather](https://t.me/BotFather)
+
+## Quick Start
+
+### 1. Create a Bot
+
+1. Message [@BotFather](https://t.me/BotFather) on Telegram
+2. Send `/newbot` and follow the prompts
+3. Copy the bot token (e.g., `123456789:ABCdefGHIjklMNOpqrsTUVwxyz`)
+
+### 2. Configure via Setup Wizard
+
+```bash
+ironclaw onboard
+```
+
+When prompted, enable the Telegram channel and paste your bot token. The wizard will:
+
+- Validate the token
+- Optionally configure a webhook secret
+- Set up tunnel (if you want webhook mode)
+
+### 3. (Optional) Configure Tunnel for Webhooks
+
+For instant message delivery, expose your agent via a tunnel:
+
+```bash
+# ngrok
+ngrok http 8080
+
+# Cloudflare
+cloudflared tunnel --url http://localhost:8080
+```
+
+Set the tunnel URL in settings or via `TUNNEL_URL` env var. Without a tunnel, the channel uses polling (~30s delay).
+
+## DM Pairing
+
+When an unknown user DMs your bot, they receive a pairing code. You must approve them before they can message the agent.
+
+### Flow
+
+1. Unknown user sends a message to your bot
+2. Bot replies: `To pair with this bot, run: ironclaw pairing approve telegram ABC12345`
+3. You run: `ironclaw pairing approve telegram ABC12345`
+4. User is added to the allow list; future messages are delivered
+
+### Commands
+
+```bash
+# List pending pairing requests
+ironclaw pairing list telegram
+
+# List as JSON
+ironclaw pairing list telegram --json
+
+# Approve a user by code
+ironclaw pairing approve telegram ABC12345
+```
+
+### Configuration
+
+Edit `~/.ironclaw/channels/telegram.capabilities.json` (or the config injected by the host):
+
+| Option | Values | Default | Description |
+|--------|--------|---------|-------------|
+| `dm_policy` | `open`, `allowlist`, `pairing` | `pairing` | `open` = allow all; `allowlist` = config + approved only; `pairing` = allowlist + send pairing reply to unknown |
+| `allow_from` | `["user_id", "username", "*"]` | `[]` | Pre-approved IDs/usernames. `*` allows everyone. |
+| `owner_id` | Telegram user ID | `null` | When set, only this user can message (overrides dm_policy) |
+| `bot_username` | Bot username (no @) | `null` | Used for mention detection in groups; when set, only strips this mention from messages |
+| `respond_to_all_group_messages` | `true`/`false` | `false` | When true, respond to all group messages; when false, only @mentions and /commands |
+
+## Manual Installation
+
+If the channel isn't installed via the wizard:
+
+```bash
+# Build the Telegram channel (requires wasm32-wasip2 target)
+rustup target add wasm32-wasip2
+./channels-src/telegram/build.sh
+
+# Install
+mkdir -p ~/.ironclaw/channels
+cp channels-src/telegram/telegram.wasm channels-src/telegram/telegram.capabilities.json ~/.ironclaw/channels/
+```
+
+## Secrets
+
+The channel expects a secret named `telegram_bot_token`. Configure via:
+
+- **Setup wizard**: Saves to encrypted secrets store
+- **Environment**: `TELEGRAM_BOT_TOKEN=your_token`
+- **Secrets store**: `ironclaw` CLI (if available)
+
+## Webhook Secret (Optional)
+
+For webhook validation, set `telegram_webhook_secret` in secrets. Telegram will send `X-Telegram-Bot-Api-Secret-Token` with each request; the host validates it before forwarding.
+
+## Troubleshooting
+
+### Messages not delivered
+
+- **Polling mode**: Check logs for `getUpdates` errors. Ensure the bot token is valid.
+- **Webhook mode**: Verify tunnel is running and `TUNNEL_URL` is correct. Telegram requires HTTPS.
+
+### Pairing code not received
+
+- Verify the channel can send messages (HTTP allowlist includes `api.telegram.org`)
+- Check `dm_policy` is `pairing` (not `allowlist` which blocks without reply)
+
+### Group mentions not working
+
+- Set `bot_username` in config to your bot's username (e.g., `MyIronClawBot`)
+- Ensure the message contains `@YourBot` or starts with `/`
+
+### "Connection refused" when starting
+
+- For webhook mode: Start your tunnel before `ironclaw run`
+- For polling only: No tunnel needed; ignore tunnel-related warnings

examples/test_heartbeat.rs

@@ -0,0 +1,122 @@
+//! Standalone heartbeat test.
+//!
+//! Exercises the heartbeat system in isolation: connects to the real
+//! database, reads the real HEARTBEAT.md, calls the real LLM, and prints
+//! every step so you can see exactly where it breaks.
+//!
+//! Usage:
+//!   cargo run --example test_heartbeat
+
+use std::sync::Arc;
+
+use ironclaw::{
+    agent::HeartbeatRunner,
+    config::Config,
+    history::Store,
+    llm::{SessionConfig, create_llm_provider, create_session_manager},
+    workspace::Workspace,
+};
+
+#[tokio::main]
+async fn main() -> anyhow::Result<()> {
+    // Load .env and set up logging
+    let _ = dotenvy::dotenv();
+    tracing_subscriber::fmt()
+        .with_env_filter("ironclaw=debug")
+        .init();
+
+    println!("=== Heartbeat Integration Test ===\n");
+
+    // 1. Load config
+    let config = Config::from_env()
+        .await
+        .map_err(|e| anyhow::anyhow!("Config: {}", e))?;
+    println!("[1/6] Config loaded");
+    println!("  heartbeat.enabled = {}", config.heartbeat.enabled);
+    println!(
+        "  heartbeat.interval_secs = {}",
+        config.heartbeat.interval_secs
+    );
+    println!(
+        "  heartbeat.notify_channel = {:?}",
+        config.heartbeat.notify_channel
+    );
+    println!(
+        "  heartbeat.notify_user = {:?}",
+        config.heartbeat.notify_user
+    );
+
+    // 2. Connect to database
+    let store = Store::new(&config.database).await?;
+    store.run_migrations().await?;
+    println!("[2/6] Database connected");
+
+    // 3. Create workspace
+    let workspace = Arc::new(Workspace::new("default", store.pool()));
+    println!("[3/6] Workspace created");
+
+    // 4. Read HEARTBEAT.md
+    let checklist = workspace.heartbeat_checklist().await;
+    match &checklist {
+        Ok(Some(content)) => {
+            let preview: String = content.chars().take(200).collect();
+            println!("[4/6] HEARTBEAT.md found ({} chars)", content.len());
+            println!("  Preview: {}...", preview);
+        }
+        Ok(None) => {
+            println!("[4/6] HEARTBEAT.md is None (no file, no seed fallback)");
+            println!("  Heartbeat will return Skipped.");
+        }
+        Err(e) => {
+            println!("[4/6] HEARTBEAT.md read error: {}", e);
+        }
+    }
+
+    // Check if the checklist would be considered "effectively empty"
+    if let Ok(Some(_)) = checklist {
+        println!("  (Will verify via runner below)");
+    }
+
+    // 5. Create LLM provider
+    let session = create_session_manager(SessionConfig {
+        auth_base_url: config.llm.nearai.auth_base_url.clone(),
+        session_path: config.llm.nearai.session_path.clone(),
+        ..Default::default()
+    })
+    .await;
+    let llm = create_llm_provider(&config.llm, session)?;
+    println!("[5/6] LLM provider created (model: {})", llm.model_name());
+
+    // 6. Run heartbeat check
+    println!("[6/6] Running check_heartbeat()...\n");
+
+    let hb_config = ironclaw::agent::HeartbeatConfig::default();
+    let runner = HeartbeatRunner::new(hb_config, workspace, llm);
+
+    let result = runner.check_heartbeat().await;
+
+    println!("=== Result ===\n");
+    match &result {
+        ironclaw::agent::HeartbeatResult::Ok => {
+            println!("HeartbeatResult::Ok");
+            println!("  LLM responded HEARTBEAT_OK, nothing needs attention.");
+        }
+        ironclaw::agent::HeartbeatResult::NeedsAttention(msg) => {
+            println!("HeartbeatResult::NeedsAttention");
+            println!("  Message:\n{}", msg);
+        }
+        ironclaw::agent::HeartbeatResult::Skipped => {
+            println!("HeartbeatResult::Skipped");
+            println!("  No checklist found, or checklist was effectively empty.");
+            println!("  This means the HEARTBEAT.md either:");
+            println!("    - Does not exist in the workspace database");
+            println!("    - Contains only headers, comments, and empty checkboxes");
+        }
+        ironclaw::agent::HeartbeatResult::Failed(err) => {
+            println!("HeartbeatResult::Failed");
+            println!("  Error: {}", err);
+        }
+    }
+
+    Ok(())
+}

migrations/V4__sandbox_columns.sql

@@ -0,0 +1,10 @@
+-- Add project_dir and user_id columns for sandbox job tracking.
+-- user_id was previously hardcoded to "default" in the Rust layer;
+-- now it's persisted so we can filter per-user.
+
+ALTER TABLE agent_jobs ADD COLUMN IF NOT EXISTS project_dir TEXT;
+ALTER TABLE agent_jobs ADD COLUMN IF NOT EXISTS user_id TEXT NOT NULL DEFAULT 'default';
+
+CREATE INDEX IF NOT EXISTS idx_agent_jobs_source ON agent_jobs(source);
+CREATE INDEX IF NOT EXISTS idx_agent_jobs_user ON agent_jobs(user_id);
+CREATE INDEX IF NOT EXISTS idx_agent_jobs_created ON agent_jobs(created_at DESC);

migrations/V5__claude_code.sql

@@ -0,0 +1,14 @@
+-- Track which mode a sandbox job uses (worker vs claude_code).
+ALTER TABLE agent_jobs ADD COLUMN IF NOT EXISTS job_mode TEXT NOT NULL DEFAULT 'worker';
+
+-- Persist Claude Code streaming events so they survive restarts and can be
+-- loaded when the frontend opens a job detail view after the fact.
+CREATE TABLE IF NOT EXISTS claude_code_events (
+    id BIGSERIAL PRIMARY KEY,
+    job_id UUID NOT NULL REFERENCES agent_jobs(id),
+    event_type TEXT NOT NULL,
+    data JSONB NOT NULL,
+    created_at TIMESTAMPTZ NOT NULL DEFAULT now()
+);
+
+CREATE INDEX IF NOT EXISTS idx_cc_events_job ON claude_code_events(job_id, id);

migrations/V6__routines.sql

@@ -0,0 +1,73 @@
+-- Routines: scheduled and reactive job system.
+--
+-- A routine is a named, persistent, user-owned task with a trigger and an action.
+-- Triggers fire independently (cron, event, webhook, manual) so only the
+-- relevant routine's prompt hits the LLM, not the whole checklist.
+
+CREATE TABLE routines (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    name TEXT NOT NULL,
+    description TEXT NOT NULL DEFAULT '',
+    user_id TEXT NOT NULL,
+    enabled BOOLEAN NOT NULL DEFAULT true,
+
+    -- Trigger definition
+    trigger_type TEXT NOT NULL,          -- 'cron', 'event', 'webhook', 'manual'
+    trigger_config JSONB NOT NULL,       -- type-specific config (schedule, pattern, etc.)
+
+    -- Action definition
+    action_type TEXT NOT NULL,           -- 'lightweight', 'full_job'
+    action_config JSONB NOT NULL,        -- prompt, context_paths, max_tokens / title, max_iterations
+
+    -- Guardrails
+    cooldown_secs INTEGER NOT NULL DEFAULT 300,
+    max_concurrent INTEGER NOT NULL DEFAULT 1,
+    dedup_window_secs INTEGER,           -- NULL = no dedup
+
+    -- Notification preferences
+    notify_channel TEXT,                 -- NULL = use default
+    notify_user TEXT NOT NULL DEFAULT 'default',
+    notify_on_success BOOLEAN NOT NULL DEFAULT false,
+    notify_on_failure BOOLEAN NOT NULL DEFAULT true,
+    notify_on_attention BOOLEAN NOT NULL DEFAULT true,
+
+    -- Runtime state (updated by engine)
+    state JSONB NOT NULL DEFAULT '{}',
+    last_run_at TIMESTAMPTZ,
+    next_fire_at TIMESTAMPTZ,            -- pre-computed for cron triggers
+    run_count BIGINT NOT NULL DEFAULT 0,
+    consecutive_failures INTEGER NOT NULL DEFAULT 0,
+
+    created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
+    updated_at TIMESTAMPTZ NOT NULL DEFAULT now(),
+
+    UNIQUE (user_id, name)
+);
+
+-- Fast lookup: "which cron routines need to fire right now?"
+CREATE INDEX idx_routines_next_fire
+    ON routines (next_fire_at)
+    WHERE enabled AND next_fire_at IS NOT NULL;
+
+-- Fast lookup: event triggers for a user
+CREATE INDEX idx_routines_event_triggers
+    ON routines (user_id)
+    WHERE enabled AND trigger_type = 'event';
+
+-- Audit log of individual routine executions.
+CREATE TABLE routine_runs (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    routine_id UUID NOT NULL REFERENCES routines(id) ON DELETE CASCADE,
+    trigger_type TEXT NOT NULL,
+    trigger_detail TEXT,                  -- e.g. matched message preview, cron expression
+    started_at TIMESTAMPTZ NOT NULL DEFAULT now(),
+    completed_at TIMESTAMPTZ,
+    status TEXT NOT NULL DEFAULT 'running',  -- running, ok, attention, failed
+    result_summary TEXT,
+    tokens_used INTEGER,
+    job_id UUID REFERENCES agent_jobs(id),   -- non-NULL for full_job runs
+    created_at TIMESTAMPTZ NOT NULL DEFAULT now()
+);
+
+CREATE INDEX idx_routine_runs_routine ON routine_runs (routine_id);
+CREATE INDEX idx_routine_runs_status ON routine_runs (status) WHERE status = 'running';

migrations/V7__rename_events.sql

@@ -0,0 +1,3 @@
+-- Rename claude_code_events to job_events (generic for all sandbox job types).
+ALTER TABLE claude_code_events RENAME TO job_events;
+ALTER INDEX idx_cc_events_job RENAME TO idx_job_events_job;

migrations/V8__settings.sql

@@ -0,0 +1,16 @@
+-- Settings table: key-value store for all user configuration.
+--
+-- Replaces ~/.ironclaw/settings.json, session.json, and mcp-servers.json.
+-- Keys use dotted paths matching the existing Settings.get()/set() convention
+-- (e.g., "agent.name", "sandbox.enabled", "mcp_servers").
+-- One row per setting so individual values can be updated atomically.
+
+CREATE TABLE IF NOT EXISTS settings (
+    user_id    TEXT        NOT NULL,
+    key        TEXT        NOT NULL,
+    value      JSONB       NOT NULL,
+    updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    PRIMARY KEY (user_id, key)
+);
+
+CREATE INDEX IF NOT EXISTS idx_settings_user ON settings (user_id);

release-plz.toml

@@ -0,0 +1,2 @@
+[workspace]
+git_release_enable = false

scripts/build-all.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+# Build IronClaw and all bundled channels.
+#
+# Run this before release or when channel sources have changed.
+# The main binary bundles telegram.wasm via include_bytes!; it must exist.
+
+set -euo pipefail
+
+cd "$(dirname "$0")/.."
+
+echo "Building bundled channels..."
+if [ -d "channels-src/telegram" ]; then
+    ./channels-src/telegram/build.sh
+fi
+
+echo ""
+echo "Building IronClaw..."
+cargo build --release
+
+echo ""
+echo "Done. Binary: target/release/ironclaw"

src/agent/agent_loop.rs

@@ -9,13 +9,14 @@ use uuid::Uuid;
 use crate::agent::compaction::ContextCompactor;
 use crate::agent::context_monitor::ContextMonitor;
 use crate::agent::heartbeat::spawn_heartbeat;
+use crate::agent::routine_engine::{RoutineEngine, spawn_cron_ticker};
 use crate::agent::self_repair::{DefaultSelfRepair, RepairResult, SelfRepair};
 use crate::agent::session::{PendingApproval, Session, ThreadState};
 use crate::agent::session_manager::SessionManager;
 use crate::agent::submission::{Submission, SubmissionParser, SubmissionResult};
 use crate::agent::{HeartbeatConfig as AgentHeartbeatConfig, MessageIntent, Router, Scheduler};
 use crate::channels::{ChannelManager, IncomingMessage, OutgoingResponse, StatusUpdate};
-use crate::config::{AgentConfig, HeartbeatConfig};
+use crate::config::{AgentConfig, HeartbeatConfig, RoutineConfig};
 use crate::context::ContextManager;
 use crate::context::JobContext;
 use crate::error::Error;
@@ -77,6 +78,7 @@ pub struct Agent {
     session_manager: Arc<SessionManager>,
     context_monitor: ContextMonitor,
     heartbeat_config: Option<HeartbeatConfig>,
+    routine_config: Option<RoutineConfig>,
 }
 
 impl Agent {
@@ -89,6 +91,7 @@ impl Agent {
         deps: AgentDeps,
         channels: ChannelManager,
         heartbeat_config: Option<HeartbeatConfig>,
+        routine_config: Option<RoutineConfig>,
         context_manager: Option<Arc<ContextManager>>,
         session_manager: Option<Arc<SessionManager>>,
     ) -> Self {
@@ -116,6 +119,7 @@ impl Agent {
             session_manager,
             context_monitor: ContextMonitor::new(),
             heartbeat_config,
+            routine_config,
         }
     }
 
@@ -256,53 +260,28 @@ impl Agent {
                     let channels = self.channels.clone();
                     tokio::spawn(async move {
                         while let Some(response) = notify_rx.recv().await {
-                            // Route notification to configured channel/user, or broadcast to all
-                            match (&notify_channel, &notify_user) {
-                                (Some(channel), Some(user)) => {
-                                    // Send to specific channel and user
-                                    if let Err(e) =
-                                        channels.broadcast(channel, user, response.clone()).await
-                                    {
+                            let user = notify_user.as_deref().unwrap_or("default");
+
+                            // Try the configured channel first, fall back to
+                            // broadcasting on all channels.
+                            let targeted_ok = if let Some(ref channel) = notify_channel {
+                                channels
+                                    .broadcast(channel, user, response.clone())
+                                    .await
+                                    .is_ok()
+                            } else {
+                                false
+                            };
+
+                            if !targeted_ok {
+                                let results = channels.broadcast_all(user, response).await;
+                                for (ch, result) in results {
+                                    if let Err(e) = result {
                                         tracing::warn!(
-                                            "Failed to send heartbeat to {}/{}: {}",
-                                            channel,
-                                            user,
+                                            "Failed to broadcast heartbeat to {}: {}",
+                                            ch,
                                             e
                                         );
-                                    } else {
-                                        tracing::debug!(
-                                            "Heartbeat notification sent to {}/{}",
-                                            channel,
-                                            user
-                                        );
-                                    }
-                                }
-                                (None, Some(user)) => {
-                                    // Broadcast to all channels for this user
-                                    let results = channels.broadcast_all(user, response).await;
-                                    for (ch, result) in results {
-                                        if let Err(e) = result {
-                                            tracing::warn!(
-                                                "Failed to broadcast heartbeat to {}: {}",
-                                                ch,
-                                                e
-                                            );
-                                        }
-                                    }
-                                }
-                                _ => {
-                                    // No explicit target, broadcast to all channels
-                                    // for the default user so notifications actually
-                                    // reach someone instead of vanishing into logs.
-                                    let results = channels.broadcast_all("default", response).await;
-                                    for (ch, result) in results {
-                                        if let Err(e) = result {
-                                            tracing::warn!(
-                                                "Failed to broadcast heartbeat to {}: {}",
-                                                ch,
-                                                e
-                                            );
-                                        }
                                     }
                                 }
                             }
@@ -330,6 +309,85 @@ impl Agent {
             None
         };
 
+        // Spawn routine engine if enabled
+        let routine_handle = if let Some(ref rt_config) = self.routine_config {
+            if rt_config.enabled {
+                if let (Some(store), Some(workspace)) = (self.store(), self.workspace()) {
+                    // Set up notification channel (same pattern as heartbeat)
+                    let (notify_tx, mut notify_rx) =
+                        tokio::sync::mpsc::channel::<OutgoingResponse>(32);
+
+                    let engine = Arc::new(RoutineEngine::new(
+                        rt_config.clone(),
+                        Arc::clone(store),
+                        self.llm().clone(),
+                        Arc::clone(workspace),
+                        notify_tx,
+                    ));
+
+                    // Register routine tools
+                    self.deps
+                        .tools
+                        .register_routine_tools(Arc::clone(store), Arc::clone(&engine));
+
+                    // Load initial event cache
+                    engine.refresh_event_cache().await;
+
+                    // Spawn notification forwarder
+                    let channels = self.channels.clone();
+                    tokio::spawn(async move {
+                        while let Some(response) = notify_rx.recv().await {
+                            let user = response
+                                .metadata
+                                .get("notify_user")
+                                .and_then(|v| v.as_str())
+                                .unwrap_or("default")
+                                .to_string();
+                            let results = channels.broadcast_all(&user, response).await;
+                            for (ch, result) in results {
+                                if let Err(e) = result {
+                                    tracing::warn!(
+                                        "Failed to broadcast routine notification to {}: {}",
+                                        ch,
+                                        e
+                                    );
+                                }
+                            }
+                        }
+                    });
+
+                    // Spawn cron ticker
+                    let cron_interval =
+                        std::time::Duration::from_secs(rt_config.cron_check_interval_secs);
+                    let cron_handle = spawn_cron_ticker(Arc::clone(&engine), cron_interval);
+
+                    // Store engine reference for event trigger checking
+                    // Safety: we're in run() which takes self, no other reference exists
+                    let engine_ref = Arc::clone(&engine);
+                    // SAFETY: self is consumed by run(), we can smuggle the engine in
+                    // via a local to use in the message loop below.
+
+                    tracing::info!(
+                        "Routines enabled: cron ticker every {}s, max {} concurrent",
+                        rt_config.cron_check_interval_secs,
+                        rt_config.max_concurrent_routines
+                    );
+
+                    Some((cron_handle, engine_ref))
+                } else {
+                    tracing::warn!("Routines enabled but store/workspace not available");
+                    None
+                }
+            } else {
+                None
+            }
+        } else {
+            None
+        };
+
+        // Extract engine ref for use in message loop
+        let routine_engine_for_loop = routine_handle.as_ref().map(|(_, e)| Arc::clone(e));
+
         // Main message loop
         tracing::info!("Agent {} ready and listening", self.config.name);
 
@@ -374,6 +432,14 @@ impl Agent {
                         .await;
                 }
             }
+
+            // Check event triggers (cheap in-memory regex, fires async if matched)
+            if let Some(ref engine) = routine_engine_for_loop {
+                let fired = engine.check_event_triggers(&message).await;
+                if fired > 0 {
+                    tracing::debug!("Fired {} event-triggered routines", fired);
+                }
+            }
         }
 
         // Cleanup
@@ -383,6 +449,9 @@ impl Agent {
         if let Some(handle) = heartbeat_handle {
             handle.abort();
         }
+        if let Some((cron_handle, _)) = routine_handle {
+            cron_handle.abort();
+        }
         self.scheduler.stop_all().await;
         self.channels.shutdown_all().await?;
 
@@ -393,6 +462,11 @@ impl Agent {
         // Parse submission type first
         let submission = SubmissionParser::parse(&message.content);
 
+        // Hydrate thread from DB if it's a historical thread not in memory
+        if let Some(ref external_thread_id) = message.thread_id {
+            self.maybe_hydrate_thread(message, external_thread_id).await;
+        }
+
         // Resolve session and thread
         let (session, thread_id) = self
             .session_manager
@@ -444,6 +518,9 @@ impl Agent {
                 self.process_user_input(message, session, thread_id, &content)
                     .await
             }
+            Submission::SystemCommand { command, args } => {
+                self.handle_system_command(&command, &args).await
+            }
             Submission::Undo => self.process_undo(session, thread_id).await,
             Submission::Redo => self.process_redo(session, thread_id).await,
             Submission::Interrupt => self.process_interrupt(session, thread_id).await,
@@ -515,6 +592,107 @@ impl Agent {
         }
     }
 
+    /// Hydrate a historical thread from DB into memory if not already present.
+    ///
+    /// Called before `resolve_thread` so that the session manager finds the
+    /// thread on lookup instead of creating a new one.
+    ///
+    /// Creates an in-memory thread with the exact UUID the frontend sent,
+    /// even when the conversation has zero messages (e.g. a brand-new
+    /// assistant thread). Without this, `resolve_thread` would mint a
+    /// fresh UUID and all messages would land in the wrong conversation.
+    async fn maybe_hydrate_thread(&self, message: &IncomingMessage, external_thread_id: &str) {
+        // Only hydrate UUID-shaped thread IDs (web gateway uses UUIDs)
+        let thread_uuid = match Uuid::parse_str(external_thread_id) {
+            Ok(id) => id,
+            Err(_) => return,
+        };
+
+        // Check if already in memory
+        let session = self
+            .session_manager
+            .get_or_create_session(&message.user_id)
+            .await;
+        {
+            let sess = session.lock().await;
+            if sess.threads.contains_key(&thread_uuid) {
+                return;
+            }
+        }
+
+        // Load history from DB (may be empty for a newly created thread).
+        let mut chat_messages: Vec<ChatMessage> = Vec::new();
+        let msg_count;
+
+        if let Some(store) = self.store() {
+            let db_messages = store
+                .list_conversation_messages(thread_uuid)
+                .await
+                .unwrap_or_default();
+            msg_count = db_messages.len();
+            chat_messages = db_messages
+                .iter()
+                .filter_map(|m| match m.role.as_str() {
+                    "user" => Some(ChatMessage::user(&m.content)),
+                    "assistant" => Some(ChatMessage::assistant(&m.content)),
+                    _ => None,
+                })
+                .collect();
+        } else {
+            msg_count = 0;
+        }
+
+        // Create thread with the historical ID and restore messages
+        let session_id = {
+            let sess = session.lock().await;
+            sess.id
+        };
+
+        let mut thread = crate::agent::session::Thread::with_id(thread_uuid, session_id);
+        if !chat_messages.is_empty() {
+            thread.restore_from_messages(chat_messages);
+        }
+
+        // Restore response chain from conversation metadata
+        if let Some(store) = self.store() {
+            if let Ok(Some(metadata)) = store.get_conversation_metadata(thread_uuid).await {
+                if let Some(rid) = metadata
+                    .get("last_response_id")
+                    .and_then(|v| v.as_str())
+                    .map(String::from)
+                {
+                    thread.last_response_id = Some(rid.clone());
+                    self.llm()
+                        .seed_response_chain(&thread_uuid.to_string(), rid);
+                    tracing::debug!("Restored response chain for thread {}", thread_uuid);
+                }
+            }
+        }
+
+        // Insert into session and register with session manager
+        {
+            let mut sess = session.lock().await;
+            sess.threads.insert(thread_uuid, thread);
+            sess.active_thread = Some(thread_uuid);
+            sess.last_active_at = chrono::Utc::now();
+        }
+
+        self.session_manager
+            .register_thread(
+                &message.user_id,
+                &message.channel,
+                thread_uuid,
+                Arc::clone(&session),
+            )
+            .await;
+
+        tracing::debug!(
+            "Hydrated thread {} from DB ({} messages)",
+            thread_uuid,
+            msg_count
+        );
+    }
+
     async fn process_user_input(
         &self,
         message: &IncomingMessage,
@@ -694,6 +872,7 @@ impl Agent {
         match result {
             Ok(AgenticLoopResult::Response(response)) => {
                 thread.complete_turn(&response);
+                self.persist_response_chain(thread);
                 let _ = self
                     .channels
                     .send_status(
@@ -702,6 +881,10 @@ impl Agent {
                         &message.metadata,
                     )
                     .await;
+
+                // Fire-and-forget: persist turn to DB
+                self.persist_turn(thread_id, &message.user_id, content, Some(&response));
+
                 Ok(SubmissionResult::response(response))
             }
             Ok(AgenticLoopResult::NeedApproval { pending }) => {
@@ -728,11 +911,95 @@ impl Agent {
             }
             Err(e) => {
                 thread.fail_turn(e.to_string());
+
+                // Persist the user message even on failure
+                self.persist_turn(thread_id, &message.user_id, content, None);
+
                 Ok(SubmissionResult::error(e.to_string()))
             }
         }
     }
 
+    /// Fire-and-forget: persist a turn (user message + optional assistant response) to the DB.
+    fn persist_turn(
+        &self,
+        thread_id: Uuid,
+        user_id: &str,
+        user_input: &str,
+        response: Option<&str>,
+    ) {
+        let store = match self.store() {
+            Some(s) => Arc::clone(s),
+            None => return,
+        };
+
+        let user_id = user_id.to_string();
+        let user_input = user_input.to_string();
+        let response = response.map(String::from);
+
+        tokio::spawn(async move {
+            if let Err(e) = store
+                .ensure_conversation(thread_id, "gateway", &user_id, None)
+                .await
+            {
+                tracing::warn!("Failed to ensure conversation {}: {}", thread_id, e);
+                return;
+            }
+
+            if let Err(e) = store
+                .add_conversation_message(thread_id, "user", &user_input)
+                .await
+            {
+                tracing::warn!("Failed to persist user message: {}", e);
+                return;
+            }
+
+            if let Some(ref resp) = response {
+                if let Err(e) = store
+                    .add_conversation_message(thread_id, "assistant", resp)
+                    .await
+                {
+                    tracing::warn!("Failed to persist assistant message: {}", e);
+                }
+            }
+        });
+    }
+
+    /// Sync the provider's response chain ID to the thread and DB metadata.
+    ///
+    /// Call after a successful agentic loop to persist the latest
+    /// `previous_response_id` so chaining survives restarts.
+    fn persist_response_chain(&self, thread: &mut crate::agent::session::Thread) {
+        let tid = thread.id.to_string();
+        let response_id = match self.llm().get_response_chain_id(&tid) {
+            Some(rid) => rid,
+            None => return,
+        };
+
+        // Update in-memory thread
+        thread.last_response_id = Some(response_id.clone());
+
+        // Fire-and-forget DB write
+        let store = match self.store() {
+            Some(s) => Arc::clone(s),
+            None => return,
+        };
+        let thread_id = thread.id;
+        tokio::spawn(async move {
+            let val = serde_json::json!(response_id);
+            if let Err(e) = store
+                .update_conversation_metadata_field(thread_id, "last_response_id", &val)
+                .await
+            {
+                tracing::warn!(
+                    "Failed to persist response chain for thread {}: {}",
+                    thread_id,
+                    e
+                );
+            }
+        });
+    }
+
     /// Run the agentic loop: call LLM, execute tools, repeat until text response.
     ///
     /// Returns `AgenticLoopResult::Response` on completion, or
@@ -782,7 +1049,7 @@ impl Agent {
             iteration += 1;
             if iteration > MAX_TOOL_ITERATIONS {
                 return Err(crate::error::LlmError::InvalidResponse {
-                    provider: "nearai".to_string(),
+                    provider: "agent".to_string(),
                     reason: format!("Exceeded maximum tool iterations ({})", MAX_TOOL_ITERATIONS),
                 }
                 .into());
@@ -808,7 +1075,12 @@ impl Agent {
             // Call LLM with current context
             let context = ReasoningContext::new()
                 .with_messages(context_messages.clone())
-                .with_tools(tool_defs);
+                .with_tools(tool_defs)
+                .with_metadata({
+                    let mut m = std::collections::HashMap::new();
+                    m.insert("thread_id".to_string(), thread_id.to_string());
+                    m
+                });
 
             let result = reasoning.respond_with_tools(&context).await?;
 
@@ -832,13 +1104,16 @@ impl Agent {
                     // Tools have been executed or we've tried multiple times, return response
                     return Ok(AgenticLoopResult::Response(text));
                 }
-                RespondResult::ToolCalls(tool_calls) => {
+                RespondResult::ToolCalls {
+                    tool_calls,
+                    content,
+                } => {
                     tools_executed = true;
 
                     // Add the assistant message with tool_calls to context.
-                    // OpenAI-compatible APIs require this before tool-result messages.
+                    // OpenAI protocol requires this before tool-result messages.
                     context_messages.push(ChatMessage::assistant_with_tool_calls(
-                        "",
+                        content,
                         tool_calls.clone(),
                     ));
 
@@ -960,10 +1235,26 @@ impl Agent {
                         if let Some((ext_name, instructions)) =
                             detect_auth_awaiting(&tc.name, &tool_result)
                         {
-                            let mut sess = session.lock().await;
-                            if let Some(thread) = sess.threads.get_mut(&thread_id) {
-                                thread.enter_auth_mode(ext_name);
+                            let auth_data = parse_auth_result(&tool_result);
+                            {
+                                let mut sess = session.lock().await;
+                                if let Some(thread) = sess.threads.get_mut(&thread_id) {
+                                    thread.enter_auth_mode(ext_name.clone());
+                                }
                             }
+                            let _ = self
+                                .channels
+                                .send_status(
+                                    &message.channel,
+                                    StatusUpdate::AuthRequired {
+                                        extension_name: ext_name,
+                                        instructions: Some(instructions.clone()),
+                                        auth_url: auth_data.auth_url,
+                                        setup_url: auth_data.setup_url,
+                                    },
+                                    &message.metadata,
+                                )
+                                .await;
                             return Ok(AgenticLoopResult::Response(instructions));
                         }
 
@@ -1024,19 +1315,59 @@ impl Agent {
             .into());
         }
 
-        // Execute with timeout
-        let result = tokio::time::timeout(std::time::Duration::from_secs(60), async {
+        tracing::debug!(
+            tool = %tool_name,
+            params = %params,
+            "Tool call started"
+        );
+
+        // Execute with per-tool timeout
+        let timeout = tool.execution_timeout();
+        let start = std::time::Instant::now();
+        let result = tokio::time::timeout(timeout, async {
             tool.execute(params.clone(), job_ctx).await
         })
-        .await
-        .map_err(|_| crate::error::ToolError::Timeout {
-            name: tool_name.to_string(),
-            timeout: std::time::Duration::from_secs(60),
-        })?
-        .map_err(|e| crate::error::ToolError::ExecutionFailed {
-            name: tool_name.to_string(),
-            reason: e.to_string(),
-        })?;
+        .await;
+        let elapsed = start.elapsed();
+
+        match &result {
+            Ok(Ok(output)) => {
+                let result_str = serde_json::to_string(&output.result)
+                    .unwrap_or_else(|_| "<serialize error>".to_string());
+                tracing::debug!(
+                    tool = %tool_name,
+                    elapsed_ms = elapsed.as_millis() as u64,
+                    result = %result_str,
+                    "Tool call succeeded"
+                );
+            }
+            Ok(Err(e)) => {
+                tracing::debug!(
+                    tool = %tool_name,
+                    elapsed_ms = elapsed.as_millis() as u64,
+                    error = %e,
+                    "Tool call failed"
+                );
+            }
+            Err(_) => {
+                tracing::debug!(
+                    tool = %tool_name,
+                    elapsed_ms = elapsed.as_millis() as u64,
+                    timeout_secs = timeout.as_secs(),
+                    "Tool call timed out"
+                );
+            }
+        }
+
+        let result = result
+            .map_err(|_| crate::error::ToolError::Timeout {
+                name: tool_name.to_string(),
+                timeout,
+            })?
+            .map_err(|e| crate::error::ToolError::ExecutionFailed {
+                name: tool_name.to_string(),
+                reason: e.to_string(),
+            })?;
 
         // Convert result to string
         serde_json::to_string_pretty(&result.result).map_err(|e| {
@@ -1380,10 +1711,11 @@ impl Agent {
             if let Some((ext_name, instructions)) =
                 detect_auth_awaiting(&pending.tool_name, &tool_result)
             {
+                let auth_data = parse_auth_result(&tool_result);
                 {
                     let mut sess = session.lock().await;
                     if let Some(thread) = sess.threads.get_mut(&thread_id) {
-                        thread.enter_auth_mode(ext_name);
+                        thread.enter_auth_mode(ext_name.clone());
                         thread.complete_turn(&instructions);
                     }
                 }
@@ -1391,7 +1723,12 @@ impl Agent {
                     .channels
                     .send_status(
                         &message.channel,
-                        StatusUpdate::Status("Awaiting token".into()),
+                        StatusUpdate::AuthRequired {
+                            extension_name: ext_name,
+                            instructions: Some(instructions.clone()),
+                            auth_url: auth_data.auth_url,
+                            setup_url: auth_data.setup_url,
+                        },
                         &message.metadata,
                     )
                     .await;
@@ -1434,6 +1771,7 @@ impl Agent {
             match result {
                 Ok(AgenticLoopResult::Response(response)) => {
                     thread.complete_turn(&response);
+                    self.persist_response_chain(thread);
                     let _ = self
                         .channels
                         .send_status(
@@ -1532,16 +1870,6 @@ impl Agent {
                     pending.extension_name
                 );
 
-                // Notify via channel status so the response doesn't echo the token
-                let _ = self
-                    .channels
-                    .send_status(
-                        &message.channel,
-                        StatusUpdate::Status("Authenticated, loading tools...".into()),
-                        &message.metadata,
-                    )
-                    .await;
-
                 // Auto-activate so tools are available immediately after auth
                 match ext_mgr.activate(&pending.extension_name).await {
                     Ok(activate_result) => {
@@ -1551,10 +1879,23 @@ impl Agent {
                         } else {
                             format!("\n\nTools: {}", activate_result.tools_loaded.join(", "))
                         };
-                        Ok(Some(format!(
+                        let msg = format!(
                             "{} authenticated and activated ({} tools loaded).{}",
                             pending.extension_name, tool_count, tool_list
-                        )))
+                        );
+                        let _ = self
+                            .channels
+                            .send_status(
+                                &message.channel,
+                                StatusUpdate::AuthCompleted {
+                                    extension_name: pending.extension_name.clone(),
+                                    success: true,
+                                    message: msg.clone(),
+                                },
+                                &message.metadata,
+                            )
+                            .await;
+                        Ok(Some(msg))
                     }
                     Err(e) => {
                         tracing::warn!(
@@ -1562,16 +1903,29 @@ impl Agent {
                             pending.extension_name,
                             e
                         );
-                        Ok(Some(format!(
+                        let msg = format!(
                             "{} authenticated successfully, but activation failed: {}. \
                              Try activating manually.",
                             pending.extension_name, e
-                        )))
+                        );
+                        let _ = self
+                            .channels
+                            .send_status(
+                                &message.channel,
+                                StatusUpdate::AuthCompleted {
+                                    extension_name: pending.extension_name.clone(),
+                                    success: true,
+                                    message: msg.clone(),
+                                },
+                                &message.metadata,
+                            )
+                            .await;
+                        Ok(Some(msg))
                     }
                 }
             }
             Ok(result) => {
-                // Unexpected state, re-enter auth mode
+                // Invalid token, re-enter auth mode
                 {
                     let mut sess = session.lock().await;
                     if let Some(thread) = sess.threads.get_mut(&thread_id) {
@@ -1580,13 +1934,43 @@ impl Agent {
                 }
                 let msg = result
                     .instructions
+                    .clone()
                     .unwrap_or_else(|| "Invalid token. Please try again.".to_string());
+                // Re-emit AuthRequired so web UI re-shows the card
+                let _ = self
+                    .channels
+                    .send_status(
+                        &message.channel,
+                        StatusUpdate::AuthRequired {
+                            extension_name: pending.extension_name.clone(),
+                            instructions: Some(msg.clone()),
+                            auth_url: result.auth_url,
+                            setup_url: result.setup_url,
+                        },
+                        &message.metadata,
+                    )
+                    .await;
+                Ok(Some(msg))
+            }
+            Err(e) => {
+                let msg = format!(
+                    "Authentication failed for {}: {}",
+                    pending.extension_name, e
+                );
+                let _ = self
+                    .channels
+                    .send_status(
+                        &message.channel,
+                        StatusUpdate::AuthCompleted {
+                            extension_name: pending.extension_name.clone(),
+                            success: false,
+                            message: msg.clone(),
+                        },
+                        &message.metadata,
+                    )
+                    .await;
                 Ok(Some(msg))
             }
-            Err(e) => Ok(Some(format!(
-                "Authentication failed for {}: {}",
-                pending.extension_name, e
-            ))),
         }
     }
 
@@ -1937,40 +2321,49 @@ impl Agent {
         }
     }
 
-    async fn handle_command(
+    /// Handle system commands that bypass thread-state checks entirely.
+    async fn handle_system_command(
         &self,
         command: &str,
-        _args: &[String],
-    ) -> Result<Option<String>, Error> {
+        args: &[String],
+    ) -> Result<SubmissionResult, Error> {
         match command {
-            "help" => Ok(Some(
-                r#"Commands:
-  /job <desc>     - Create a job
-  /status [id]    - Check job status
-  /cancel <id>    - Cancel a job
-  /list           - List all jobs
-  /help <job_id>  - Help a stuck job
+            "help" => Ok(SubmissionResult::response(concat!(
+                "System:\n",
+                "  /help             Show this help\n",
+                "  /model [name]     Show or switch the active model\n",
+                "  /version          Show version info\n",
+                "  /tools            List available tools\n",
+                "  /debug            Toggle debug mode\n",
+                "  /ping             Connectivity check\n",
+                "\n",
+                "Jobs:\n",
+                "  /job <desc>       Create a new job\n",
+                "  /status [id]      Check job status\n",
+                "  /cancel <id>      Cancel a job\n",
+                "  /list             List all jobs\n",
+                "\n",
+                "Session:\n",
+                "  /undo             Undo last turn\n",
+                "  /redo             Redo undone turn\n",
+                "  /compact          Compress context window\n",
+                "  /clear            Clear current thread\n",
+                "  /interrupt        Stop current operation\n",
+                "  /new              New conversation thread\n",
+                "  /thread <id>      Switch to thread\n",
+                "  /resume <id>      Resume from checkpoint\n",
+                "\n",
+                "Agent:\n",
+                "  /heartbeat        Run heartbeat check\n",
+                "  /summarize        Summarize current thread\n",
+                "  /suggest          Suggest next steps\n",
+                "\n",
+                "  /quit             Exit",
+            ))),
 
-  /undo           - Undo last turn
-  /redo           - Redo undone turn
-  /compact        - Compress context
-  /clear          - Clear thread
-  /interrupt      - Stop current turn
-  /thread new     - New thread
-  /thread <id>    - Switch thread
-  /resume <id>    - Resume checkpoint
+            "ping" => Ok(SubmissionResult::response("pong!")),
 
-  /heartbeat      - Run heartbeat check now
-  /summarize      - Summarize current thread
-  /suggest        - Suggest next steps
-
-  /quit           - Exit"#
-                    .to_string(),
-            )),
-
-            "ping" => Ok(Some("pong!".to_string())),
-
-            "version" => Ok(Some(format!(
+            "version" => Ok(SubmissionResult::response(format!(
                 "{} v{}",
                 env!("CARGO_PKG_NAME"),
                 env!("CARGO_PKG_VERSION")
@@ -1978,12 +2371,113 @@ impl Agent {
 
             "tools" => {
                 let tools = self.tools().list().await;
-                Ok(Some(format!("Available tools: {}", tools.join(", "))))
+                Ok(SubmissionResult::response(format!(
+                    "Available tools: {}",
+                    tools.join(", ")
+                )))
             }
 
-            _ => Ok(Some(format!("Unknown command: {}. Try /help", command))),
+            "debug" => {
+                // Debug toggle is handled client-side in the REPL.
+                // For non-REPL channels, just acknowledge.
+                Ok(SubmissionResult::ok_with_message(
+                    "Debug toggle is handled by your client.",
+                ))
+            }
+
+            "model" => {
+                if args.is_empty() {
+                    // Show current model
+                    let name = self.llm().active_model_name();
+                    Ok(SubmissionResult::response(format!(
+                        "Active model: {}",
+                        name
+                    )))
+                } else {
+                    let requested = &args[0];
+
+                    // Validate the model exists
+                    match self.llm().list_models().await {
+                        Ok(models) if !models.is_empty() => {
+                            if !models.iter().any(|m| m == requested) {
+                                return Ok(SubmissionResult::error(format!(
+                                    "Unknown model: {}. Available models:\n  {}",
+                                    requested,
+                                    models.join("\n  ")
+                                )));
+                            }
+                        }
+                        Ok(_) => {
+                            // Empty model list, can't validate but try anyway
+                        }
+                        Err(e) => {
+                            tracing::warn!("Could not fetch model list for validation: {}", e);
+                            // Proceed anyway, the provider will error on the next call if invalid
+                        }
+                    }
+
+                    match self.llm().set_model(requested) {
+                        Ok(()) => Ok(SubmissionResult::response(format!(
+                            "Switched model to: {}",
+                            requested
+                        ))),
+                        Err(e) => Ok(SubmissionResult::error(format!(
+                            "Failed to switch model: {}",
+                            e
+                        ))),
+                    }
+                }
+            }
+
+            _ => Ok(SubmissionResult::error(format!(
+                "Unknown command: {}. Try /help",
+                command
+            ))),
         }
     }
+
+    /// Handle legacy command routing from the Router (job commands that go through
+    /// process_user_input -> router -> handle_job_or_command -> here).
+    async fn handle_command(
+        &self,
+        command: &str,
+        args: &[String],
+    ) -> Result<Option<String>, Error> {
+        // System commands are now handled directly via Submission::SystemCommand,
+        // but the router may still send us unknown /commands.
+        match self.handle_system_command(command, args).await? {
+            SubmissionResult::Response { content } => Ok(Some(content)),
+            SubmissionResult::Ok { message } => Ok(message),
+            SubmissionResult::Error { message } => Ok(Some(format!("Error: {}", message))),
+            _ => Ok(None),
+        }
+    }
+}
+
+/// Parsed auth result fields for emitting StatusUpdate::AuthRequired.
+struct ParsedAuthData {
+    auth_url: Option<String>,
+    setup_url: Option<String>,
+}
+
+/// Extract auth_url and setup_url from a tool_auth result JSON string.
+fn parse_auth_result(result: &Result<String, Error>) -> ParsedAuthData {
+    let parsed = result
+        .as_ref()
+        .ok()
+        .and_then(|s| serde_json::from_str::<serde_json::Value>(s).ok());
+    ParsedAuthData {
+        auth_url: parsed
+            .as_ref()
+            .and_then(|v| v.get("auth_url"))
+            .and_then(|v| v.as_str())
+            .map(|s| s.to_string()),
+        setup_url: parsed
+            .as_ref()
+            .and_then(|v| v.get("setup_url"))
+            .and_then(|v| v.as_str())
+            .map(|s| s.to_string()),
+    }
 }
 
 /// Check if a tool_auth result indicates the extension is awaiting a token.
@@ -1994,7 +2488,7 @@ fn detect_auth_awaiting(
     tool_name: &str,
     result: &Result<String, Error>,
 ) -> Option<(String, String)> {
-    if tool_name != "tool_auth" {
+    if tool_name != "tool_auth" && tool_name != "tool_activate" {
         return None;
     }
     let output = result.as_ref().ok()?;
@@ -2078,4 +2572,34 @@ mod tests {
         let (_, instructions) = detect_auth_awaiting("tool_auth", &result).unwrap();
         assert_eq!(instructions, "Please provide your API token/key.");
     }
+
+    #[test]
+    fn test_detect_auth_awaiting_tool_activate() {
+        let result: Result<String, Error> = Ok(serde_json::json!({
+            "name": "slack",
+            "kind": "McpServer",
+            "awaiting_token": true,
+            "status": "awaiting_token",
+            "instructions": "Provide your Slack Bot token."
+        })
+        .to_string());
+
+        let detected = detect_auth_awaiting("tool_activate", &result);
+        assert!(detected.is_some());
+        let (name, instructions) = detected.unwrap();
+        assert_eq!(name, "slack");
+        assert!(instructions.contains("Slack Bot"));
+    }
+
+    #[test]
+    fn test_detect_auth_awaiting_tool_activate_not_awaiting() {
+        let result: Result<String, Error> = Ok(serde_json::json!({
+            "name": "slack",
+            "tools_loaded": ["slack_post_message"],
+            "message": "Activated"
+        })
+        .to_string());
+
+        assert!(detect_auth_awaiting("tool_activate", &result).is_none());
+    }
 }

src/agent/heartbeat.rs

@@ -29,7 +29,7 @@ use std::time::Duration;
 use tokio::sync::mpsc;
 
 use crate::channels::OutgoingResponse;
-use crate::llm::{ChatMessage, CompletionRequest, LlmProvider};
+use crate::llm::{ChatMessage, CompletionRequest, FinishReason, LlmProvider};
 use crate::workspace::Workspace;
 
 /// Configuration for the heartbeat runner.
@@ -217,9 +217,26 @@ impl HeartbeatRunner {
             ]
         };
 
+        // Use the model's context_length to set max_tokens. The API returns
+        // the total context window; we cap output at half of that (the rest is
+        // the prompt) with a floor of 4096.
+        let max_tokens = match self.llm.model_metadata().await {
+            Ok(meta) => {
+                let from_api = meta.context_length.map(|ctx| ctx / 2).unwrap_or(4096);
+                from_api.max(4096)
+            }
+            Err(e) => {
+                tracing::warn!(
+                    "Could not fetch model metadata, using default max_tokens: {}",
+                    e
+                );
+                4096
+            }
+        };
+
         let request = CompletionRequest::new(messages)
-            .with_max_tokens(1024)
-            .with_temperature(0.3); // Lower temperature for more focused responses
+            .with_max_tokens(max_tokens)
+            .with_temperature(0.3);
 
         let response = match self.llm.complete(request).await {
             Ok(r) => r,
@@ -228,6 +245,20 @@ impl HeartbeatRunner {
 
         let content = response.content.trim();
 
+        // Guard against empty content. Reasoning models (e.g. GLM-4.7) may
+        // burn all output tokens on chain-of-thought and return content: null.
+        if content.is_empty() {
+            return if response.finish_reason == FinishReason::Length {
+                HeartbeatResult::Failed(
+                    "LLM response was truncated (finish_reason=length) with no content. \
+                     The model may have exhausted its token budget on reasoning."
+                        .to_string(),
+                )
+            } else {
+                HeartbeatResult::Failed("LLM returned empty content.".to_string())
+            };
+        }
+
         // Check if nothing needs attention
         if content == "HEARTBEAT_OK" || content.contains("HEARTBEAT_OK") {
             return HeartbeatResult::Ok;

src/agent/mod.rs

@@ -6,6 +6,7 @@
 //! - Tool invocation with safety
 //! - Self-repair for stuck jobs
 //! - Proactive heartbeat execution
+//! - Routine-based scheduled and reactive jobs
 //! - Turn-based session management with undo
 //! - Context compaction for long conversations
 
@@ -14,6 +15,8 @@ pub mod compaction;
 pub mod context_monitor;
 mod heartbeat;
 mod router;
+pub mod routine;
+pub mod routine_engine;
 mod scheduler;
 mod self_repair;
 pub mod session;
@@ -28,6 +31,8 @@ pub use compaction::{CompactionResult, ContextCompactor};
 pub use context_monitor::{CompactionStrategy, ContextBreakdown, ContextMonitor};
 pub use heartbeat::{HeartbeatConfig, HeartbeatResult, HeartbeatRunner, spawn_heartbeat};
 pub use router::{MessageIntent, Router};
+pub use routine::{Routine, RoutineAction, RoutineRun, Trigger};
+pub use routine_engine::RoutineEngine;
 pub use scheduler::Scheduler;
 pub use self_repair::{BrokenTool, RepairResult, RepairTask, SelfRepair, StuckJob};
 pub use session::{PendingApproval, PendingAuth, Session, Thread, ThreadState, Turn, TurnState};

src/agent/routine.rs

@@ -0,0 +1,509 @@
+//! Core types for the routines system.
+//!
+//! A routine is a named, persistent, user-owned task with a trigger and an action.
+//! Each routine fires independently when its trigger condition is met, with only
+//! that routine's prompt and context sent to the LLM.
+//!
+//! ```text
+//! ┌──────────┐     ┌─────────┐     ┌──────────────────┐
+//! │  Trigger  │────▶│ Engine  │────▶│  Execution Mode  │
+//! │ cron/event│     │guardrail│     │lightweight│full_job│
+//! │ webhook   │     │ check   │     └──────────────────┘
+//! │ manual    │     └─────────┘              │
+//! └──────────┘                               ▼
+//!                                     ┌──────────────┐
+//!                                     │  Notify user │
+//!                                     │  if needed   │
+//!                                     └──────────────┘
+//! ```
+
+use std::collections::hash_map::DefaultHasher;
+use std::hash::{Hash, Hasher};
+use std::str::FromStr;
+use std::time::Duration;
+
+use chrono::{DateTime, Utc};
+use serde::{Deserialize, Serialize};
+use uuid::Uuid;
+
+/// A routine is a named, persistent, user-owned task with a trigger and an action.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Routine {
+    pub id: Uuid,
+    pub name: String,
+    pub description: String,
+    pub user_id: String,
+    pub enabled: bool,
+    pub trigger: Trigger,
+    pub action: RoutineAction,
+    pub guardrails: RoutineGuardrails,
+    pub notify: NotifyConfig,
+
+    // Runtime state (DB-managed)
+    pub last_run_at: Option<DateTime<Utc>>,
+    pub next_fire_at: Option<DateTime<Utc>>,
+    pub run_count: u64,
+    pub consecutive_failures: u32,
+    pub state: serde_json::Value,
+
+    pub created_at: DateTime<Utc>,
+    pub updated_at: DateTime<Utc>,
+}
+
+/// When a routine should fire.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(tag = "type", rename_all = "snake_case")]
+pub enum Trigger {
+    /// Fire on a cron schedule (e.g. "0 9 * * MON-FRI" or "every 2h").
+    Cron { schedule: String },
+    /// Fire when a channel message matches a pattern.
+    Event {
+        /// Optional channel filter (e.g. "telegram", "slack").
+        channel: Option<String>,
+        /// Regex pattern to match against message content.
+        pattern: String,
+    },
+    /// Fire on incoming webhook POST to /hooks/routine/{id}.
+    Webhook {
+        /// Optional webhook path suffix (defaults to routine id).
+        path: Option<String>,
+        /// Optional shared secret for HMAC validation.
+        secret: Option<String>,
+    },
+    /// Only fires via tool call or CLI.
+    Manual,
+}
+
+impl Trigger {
+    /// The string tag stored in the DB trigger_type column.
+    pub fn type_tag(&self) -> &'static str {
+        match self {
+            Trigger::Cron { .. } => "cron",
+            Trigger::Event { .. } => "event",
+            Trigger::Webhook { .. } => "webhook",
+            Trigger::Manual => "manual",
+        }
+    }
+
+    /// Parse a trigger from its DB representation.
+    pub fn from_db(trigger_type: &str, config: serde_json::Value) -> Result<Self, String> {
+        match trigger_type {
+            "cron" => {
+                let schedule = config
+                    .get("schedule")
+                    .and_then(|v| v.as_str())
+                    .ok_or("cron trigger missing 'schedule'")?
+                    .to_string();
+                Ok(Trigger::Cron { schedule })
+            }
+            "event" => {
+                let pattern = config
+                    .get("pattern")
+                    .and_then(|v| v.as_str())
+                    .ok_or("event trigger missing 'pattern'")?
+                    .to_string();
+                let channel = config
+                    .get("channel")
+                    .and_then(|v| v.as_str())
+                    .map(String::from);
+                Ok(Trigger::Event { channel, pattern })
+            }
+            "webhook" => {
+                let path = config
+                    .get("path")
+                    .and_then(|v| v.as_str())
+                    .map(String::from);
+                let secret = config
+                    .get("secret")
+                    .and_then(|v| v.as_str())
+                    .map(String::from);
+                Ok(Trigger::Webhook { path, secret })
+            }
+            "manual" => Ok(Trigger::Manual),
+            other => Err(format!("unknown trigger type: {other}")),
+        }
+    }
+
+    /// Serialize trigger-specific config to JSON for DB storage.
+    pub fn to_config_json(&self) -> serde_json::Value {
+        match self {
+            Trigger::Cron { schedule } => serde_json::json!({ "schedule": schedule }),
+            Trigger::Event { channel, pattern } => serde_json::json!({
+                "pattern": pattern,
+                "channel": channel,
+            }),
+            Trigger::Webhook { path, secret } => serde_json::json!({
+                "path": path,
+                "secret": secret,
+            }),
+            Trigger::Manual => serde_json::json!({}),
+        }
+    }
+}
+
+/// What happens when a routine fires.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(tag = "type", rename_all = "snake_case")]
+pub enum RoutineAction {
+    /// Single LLM call, no tools. Cheap and fast.
+    Lightweight {
+        /// The prompt sent to the LLM.
+        prompt: String,
+        /// Workspace paths to load as context (e.g. ["context/priorities.md"]).
+        #[serde(default)]
+        context_paths: Vec<String>,
+        /// Max output tokens (default: 4096).
+        #[serde(default = "default_max_tokens")]
+        max_tokens: u32,
+    },
+    /// Full multi-turn worker job with tool access.
+    FullJob {
+        /// Job title for the scheduler.
+        title: String,
+        /// Job description / initial prompt.
+        description: String,
+        /// Max reasoning iterations (default: 10).
+        #[serde(default = "default_max_iterations")]
+        max_iterations: u32,
+    },
+}
+
+fn default_max_tokens() -> u32 {
+    4096
+}
+
+fn default_max_iterations() -> u32 {
+    10
+}
+
+impl RoutineAction {
+    /// The string tag stored in the DB action_type column.
+    pub fn type_tag(&self) -> &'static str {
+        match self {
+            RoutineAction::Lightweight { .. } => "lightweight",
+            RoutineAction::FullJob { .. } => "full_job",
+        }
+    }
+
+    /// Parse an action from its DB representation.
+    pub fn from_db(action_type: &str, config: serde_json::Value) -> Result<Self, String> {
+        match action_type {
+            "lightweight" => {
+                let prompt = config
+                    .get("prompt")
+                    .and_then(|v| v.as_str())
+                    .ok_or("lightweight action missing 'prompt'")?
+                    .to_string();
+                let context_paths = config
+                    .get("context_paths")
+                    .and_then(|v| v.as_array())
+                    .map(|arr| {
+                        arr.iter()
+                            .filter_map(|v| v.as_str().map(String::from))
+                            .collect()
+                    })
+                    .unwrap_or_default();
+                let max_tokens = config
+                    .get("max_tokens")
+                    .and_then(|v| v.as_u64())
+                    .unwrap_or(default_max_tokens() as u64) as u32;
+                Ok(RoutineAction::Lightweight {
+                    prompt,
+                    context_paths,
+                    max_tokens,
+                })
+            }
+            "full_job" => {
+                let title = config
+                    .get("title")
+                    .and_then(|v| v.as_str())
+                    .ok_or("full_job action missing 'title'")?
+                    .to_string();
+                let description = config
+                    .get("description")
+                    .and_then(|v| v.as_str())
+                    .ok_or("full_job action missing 'description'")?
+                    .to_string();
+                let max_iterations = config
+                    .get("max_iterations")
+                    .and_then(|v| v.as_u64())
+                    .unwrap_or(default_max_iterations() as u64)
+                    as u32;
+                Ok(RoutineAction::FullJob {
+                    title,
+                    description,
+                    max_iterations,
+                })
+            }
+            other => Err(format!("unknown action type: {other}")),
+        }
+    }
+
+    /// Serialize action config to JSON for DB storage.
+    pub fn to_config_json(&self) -> serde_json::Value {
+        match self {
+            RoutineAction::Lightweight {
+                prompt,
+                context_paths,
+                max_tokens,
+            } => serde_json::json!({
+                "prompt": prompt,
+                "context_paths": context_paths,
+                "max_tokens": max_tokens,
+            }),
+            RoutineAction::FullJob {
+                title,
+                description,
+                max_iterations,
+            } => serde_json::json!({
+                "title": title,
+                "description": description,
+                "max_iterations": max_iterations,
+            }),
+        }
+    }
+}
+
+/// Guardrails to prevent runaway execution.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RoutineGuardrails {
+    /// Minimum time between fires.
+    pub cooldown: Duration,
+    /// Max simultaneous runs of this routine.
+    pub max_concurrent: u32,
+    /// Window for content-hash dedup (event triggers). None = no dedup.
+    pub dedup_window: Option<Duration>,
+}
+
+impl Default for RoutineGuardrails {
+    fn default() -> Self {
+        Self {
+            cooldown: Duration::from_secs(300),
+            max_concurrent: 1,
+            dedup_window: None,
+        }
+    }
+}
+
+/// Notification preferences for a routine.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct NotifyConfig {
+    /// Channel to notify on (None = default/broadcast all).
+    pub channel: Option<String>,
+    /// User to notify.
+    pub user: String,
+    /// Notify when routine produces actionable output.
+    pub on_attention: bool,
+    /// Notify when routine errors.
+    pub on_failure: bool,
+    /// Notify when routine runs with no findings.
+    pub on_success: bool,
+}
+
+impl Default for NotifyConfig {
+    fn default() -> Self {
+        Self {
+            channel: None,
+            user: "default".to_string(),
+            on_attention: true,
+            on_failure: true,
+            on_success: false,
+        }
+    }
+}
+
+/// Status of a routine run.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "snake_case")]
+pub enum RunStatus {
+    Running,
+    Ok,
+    Attention,
+    Failed,
+}
+
+impl std::fmt::Display for RunStatus {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            RunStatus::Running => write!(f, "running"),
+            RunStatus::Ok => write!(f, "ok"),
+            RunStatus::Attention => write!(f, "attention"),
+            RunStatus::Failed => write!(f, "failed"),
+        }
+    }
+}
+
+impl FromStr for RunStatus {
+    type Err = String;
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        match s {
+            "running" => Ok(RunStatus::Running),
+            "ok" => Ok(RunStatus::Ok),
+            "attention" => Ok(RunStatus::Attention),
+            "failed" => Ok(RunStatus::Failed),
+            other => Err(format!("unknown run status: {other}")),
+        }
+    }
+}
+
+/// A single execution of a routine.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RoutineRun {
+    pub id: Uuid,
+    pub routine_id: Uuid,
+    pub trigger_type: String,
+    pub trigger_detail: Option<String>,
+    pub started_at: DateTime<Utc>,
+    pub completed_at: Option<DateTime<Utc>>,
+    pub status: RunStatus,
+    pub result_summary: Option<String>,
+    pub tokens_used: Option<i32>,
+    pub job_id: Option<Uuid>,
+    pub created_at: DateTime<Utc>,
+}
+
+/// Compute a content hash for event dedup.
+pub fn content_hash(content: &str) -> u64 {
+    let mut hasher = DefaultHasher::new();
+    content.hash(&mut hasher);
+    hasher.finish()
+}
+
+/// Parse a cron expression and compute the next fire time from now.
+pub fn next_cron_fire(schedule: &str) -> Result<Option<DateTime<Utc>>, String> {
+    let cron_schedule =
+        cron::Schedule::from_str(schedule).map_err(|e| format!("invalid cron: {e}"))?;
+    Ok(cron_schedule.upcoming(Utc).next())
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::agent::routine::{
+        RoutineAction, RoutineGuardrails, RunStatus, Trigger, content_hash, next_cron_fire,
+    };
+
+    #[test]
+    fn test_trigger_roundtrip() {
+        let trigger = Trigger::Cron {
+            schedule: "0 9 * * MON-FRI".to_string(),
+        };
+        let json = trigger.to_config_json();
+        let parsed = Trigger::from_db("cron", json).expect("parse cron");
+        assert!(matches!(parsed, Trigger::Cron { schedule } if schedule == "0 9 * * MON-FRI"));
+    }
+
+    #[test]
+    fn test_event_trigger_roundtrip() {
+        let trigger = Trigger::Event {
+            channel: Some("telegram".to_string()),
+            pattern: r"deploy\s+\w+".to_string(),
+        };
+        let json = trigger.to_config_json();
+        let parsed = Trigger::from_db("event", json).expect("parse event");
+        assert!(matches!(parsed, Trigger::Event { channel, pattern }
+            if channel == Some("telegram".to_string()) && pattern == r"deploy\s+\w+"));
+    }
+
+    #[test]
+    fn test_action_lightweight_roundtrip() {
+        let action = RoutineAction::Lightweight {
+            prompt: "Check PRs".to_string(),
+            context_paths: vec!["context/priorities.md".to_string()],
+            max_tokens: 2048,
+        };
+        let json = action.to_config_json();
+        let parsed = RoutineAction::from_db("lightweight", json).expect("parse lightweight");
+        assert!(
+            matches!(parsed, RoutineAction::Lightweight { prompt, context_paths, max_tokens }
+            if prompt == "Check PRs" && context_paths.len() == 1 && max_tokens == 2048)
+        );
+    }
+
+    #[test]
+    fn test_action_full_job_roundtrip() {
+        let action = RoutineAction::FullJob {
+            title: "Deploy review".to_string(),
+            description: "Review and deploy pending changes".to_string(),
+            max_iterations: 5,
+        };
+        let json = action.to_config_json();
+        let parsed = RoutineAction::from_db("full_job", json).expect("parse full_job");
+        assert!(
+            matches!(parsed, RoutineAction::FullJob { title, max_iterations, .. }
+            if title == "Deploy review" && max_iterations == 5)
+        );
+    }
+
+    #[test]
+    fn test_run_status_display_parse() {
+        for status in [
+            RunStatus::Running,
+            RunStatus::Ok,
+            RunStatus::Attention,
+            RunStatus::Failed,
+        ] {
+            let s = status.to_string();
+            let parsed: RunStatus = s.parse().expect("parse status");
+            assert_eq!(parsed, status);
+        }
+    }
+
+    #[test]
+    fn test_content_hash_deterministic() {
+        let h1 = content_hash("deploy production");
+        let h2 = content_hash("deploy production");
+        assert_eq!(h1, h2);
+
+        let h3 = content_hash("deploy staging");
+        assert_ne!(h1, h3);
+    }
+
+    #[test]
+    fn test_next_cron_fire_valid() {
+        // Every minute should always have a next fire
+        let next = next_cron_fire("* * * * * *").expect("valid cron");
+        assert!(next.is_some());
+    }
+
+    #[test]
+    fn test_next_cron_fire_invalid() {
+        let result = next_cron_fire("not a cron");
+        assert!(result.is_err());
+    }
+
+    #[test]
+    fn test_guardrails_default() {
+        let g = RoutineGuardrails::default();
+        assert_eq!(g.cooldown.as_secs(), 300);
+        assert_eq!(g.max_concurrent, 1);
+        assert!(g.dedup_window.is_none());
+    }
+
+    #[test]
+    fn test_trigger_type_tag() {
+        assert_eq!(
+            Trigger::Cron {
+                schedule: String::new()
+            }
+            .type_tag(),
+            "cron"
+        );
+        assert_eq!(
+            Trigger::Event {
+                channel: None,
+                pattern: String::new()
+            }
+            .type_tag(),
+            "event"
+        );
+        assert_eq!(
+            Trigger::Webhook {
+                path: None,
+                secret: None
+            }
+            .type_tag(),
+            "webhook"
+        );
+        assert_eq!(Trigger::Manual.type_tag(), "manual");
+    }
+}

src/agent/routine_engine.rs

@@ -0,0 +1,606 @@
+//! Routine execution engine.
+//!
+//! Handles loading routines, checking triggers, enforcing guardrails,
+//! and executing both lightweight (single LLM call) and full-job routines.
+//!
+//! The engine runs two independent loops:
+//! - A **cron ticker** that polls the DB every N seconds for due cron routines
+//! - An **event matcher** called synchronously from the agent main loop
+//!
+//! Lightweight routines execute inline (single LLM call, no scheduler slot).
+//! Full-job routines are delegated to the existing `Scheduler`.
+
+use std::sync::Arc;
+use std::time::Duration;
+
+use chrono::Utc;
+use regex::Regex;
+use tokio::sync::{RwLock, mpsc};
+use uuid::Uuid;
+
+use crate::agent::routine::{
+    NotifyConfig, Routine, RoutineAction, RoutineRun, RunStatus, Trigger, next_cron_fire,
+};
+use crate::channels::{IncomingMessage, OutgoingResponse};
+use crate::config::RoutineConfig;
+use crate::history::Store;
+use crate::llm::{ChatMessage, CompletionRequest, FinishReason, LlmProvider};
+use crate::workspace::Workspace;
+
+/// The routine execution engine.
+pub struct RoutineEngine {
+    config: RoutineConfig,
+    store: Arc<Store>,
+    llm: Arc<dyn LlmProvider>,
+    workspace: Arc<Workspace>,
+    /// Sender for notifications (routed to channel manager).
+    notify_tx: mpsc::Sender<OutgoingResponse>,
+    /// Currently running routine count (across all routines).
+    running_count: Arc<RwLock<usize>>,
+    /// Compiled event regex cache: routine_id -> compiled regex.
+    event_cache: Arc<RwLock<Vec<(Uuid, Routine, Regex)>>>,
+}
+
+impl RoutineEngine {
+    pub fn new(
+        config: RoutineConfig,
+        store: Arc<Store>,
+        llm: Arc<dyn LlmProvider>,
+        workspace: Arc<Workspace>,
+        notify_tx: mpsc::Sender<OutgoingResponse>,
+    ) -> Self {
+        Self {
+            config,
+            store,
+            llm,
+            workspace,
+            notify_tx,
+            running_count: Arc::new(RwLock::new(0)),
+            event_cache: Arc::new(RwLock::new(Vec::new())),
+        }
+    }
+
+    /// Refresh the in-memory event trigger cache from DB.
+    pub async fn refresh_event_cache(&self) {
+        match self.store.list_event_routines().await {
+            Ok(routines) => {
+                let mut cache = Vec::new();
+                for routine in routines {
+                    if let Trigger::Event { ref pattern, .. } = routine.trigger {
+                        match Regex::new(pattern) {
+                            Ok(re) => cache.push((routine.id, routine.clone(), re)),
+                            Err(e) => {
+                                tracing::warn!(
+                                    routine = %routine.name,
+                                    "Invalid event regex '{}': {}",
+                                    pattern, e
+                                );
+                            }
+                        }
+                    }
+                }
+                let count = cache.len();
+                *self.event_cache.write().await = cache;
+                tracing::debug!("Refreshed event cache: {} routines", count);
+            }
+            Err(e) => {
+                tracing::error!("Failed to refresh event cache: {}", e);
+            }
+        }
+    }
+
+    /// Check incoming message against event triggers. Returns number of routines fired.
+    ///
+    /// Called synchronously from the main loop after handle_message(). The actual
+    /// execution is spawned async so this returns quickly.
+    pub async fn check_event_triggers(&self, message: &IncomingMessage) -> usize {
+        let cache = self.event_cache.read().await;
+        let mut fired = 0;
+
+        for (_, routine, re) in cache.iter() {
+            // Channel filter
+            if let Trigger::Event {
+                channel: Some(ch), ..
+            } = &routine.trigger
+            {
+                if ch != &message.channel {
+                    continue;
+                }
+            }
+
+            // Regex match
+            if !re.is_match(&message.content) {
+                continue;
+            }
+
+            // Cooldown check
+            if !self.check_cooldown(routine) {
+                tracing::debug!(routine = %routine.name, "Skipped: cooldown active");
+                continue;
+            }
+
+            // Concurrent run check
+            if !self.check_concurrent(routine).await {
+                tracing::debug!(routine = %routine.name, "Skipped: max concurrent reached");
+                continue;
+            }
+
+            // Global capacity check
+            if *self.running_count.read().await >= self.config.max_concurrent_routines {
+                tracing::warn!(routine = %routine.name, "Skipped: global max concurrent reached");
+                continue;
+            }
+
+            let detail = truncate(&message.content, 200);
+            self.spawn_fire(routine.clone(), "event", Some(detail));
+            fired += 1;
+        }
+
+        fired
+    }
+
+    /// Check all due cron routines and fire them. Called by the cron ticker.
+    pub async fn check_cron_triggers(&self) {
+        let routines = match self.store.list_due_cron_routines().await {
+            Ok(r) => r,
+            Err(e) => {
+                tracing::error!("Failed to load due cron routines: {}", e);
+                return;
+            }
+        };
+
+        for routine in routines {
+            if *self.running_count.read().await >= self.config.max_concurrent_routines {
+                tracing::warn!("Global max concurrent routines reached, skipping remaining");
+                break;
+            }
+
+            if !self.check_cooldown(&routine) {
+                continue;
+            }
+
+            if !self.check_concurrent(&routine).await {
+                continue;
+            }
+
+            let detail = if let Trigger::Cron { ref schedule } = routine.trigger {
+                Some(schedule.clone())
+            } else {
+                None
+            };
+
+            self.spawn_fire(routine, "cron", detail);
+        }
+    }
+
+    /// Fire a routine manually (from tool call or CLI).
+    pub async fn fire_manual(&self, routine_id: Uuid) -> Result<Uuid, String> {
+        let routine = self
+            .store
+            .get_routine(routine_id)
+            .await
+            .map_err(|e| format!("DB error: {e}"))?
+            .ok_or_else(|| format!("routine {routine_id} not found"))?;
+
+        if !routine.enabled {
+            return Err(format!("routine '{}' is disabled", routine.name));
+        }
+
+        if !self.check_concurrent(&routine).await {
+            return Err(format!(
+                "routine '{}' already at max concurrent runs",
+                routine.name
+            ));
+        }
+
+        let run_id = Uuid::new_v4();
+        let run = RoutineRun {
+            id: run_id,
+            routine_id: routine.id,
+            trigger_type: "manual".to_string(),
+            trigger_detail: None,
+            started_at: Utc::now(),
+            completed_at: None,
+            status: RunStatus::Running,
+            result_summary: None,
+            tokens_used: None,
+            job_id: None,
+            created_at: Utc::now(),
+        };
+
+        if let Err(e) = self.store.create_routine_run(&run).await {
+            return Err(format!("failed to create run record: {e}"));
+        }
+
+        // Execute inline for manual triggers (caller wants to wait)
+        let engine = EngineContext {
+            store: self.store.clone(),
+            llm: self.llm.clone(),
+            workspace: self.workspace.clone(),
+            notify_tx: self.notify_tx.clone(),
+            running_count: self.running_count.clone(),
+            max_lightweight_tokens: self.config.max_lightweight_tokens,
+        };
+
+        tokio::spawn(async move {
+            execute_routine(engine, routine, run).await;
+        });
+
+        Ok(run_id)
+    }
+
+    /// Spawn a fire in a background task.
+    fn spawn_fire(&self, routine: Routine, trigger_type: &str, trigger_detail: Option<String>) {
+        let run = RoutineRun {
+            id: Uuid::new_v4(),
+            routine_id: routine.id,
+            trigger_type: trigger_type.to_string(),
+            trigger_detail,
+            started_at: Utc::now(),
+            completed_at: None,
+            status: RunStatus::Running,
+            result_summary: None,
+            tokens_used: None,
+            job_id: None,
+            created_at: Utc::now(),
+        };
+
+        let engine = EngineContext {
+            store: self.store.clone(),
+            llm: self.llm.clone(),
+            workspace: self.workspace.clone(),
+            notify_tx: self.notify_tx.clone(),
+            running_count: self.running_count.clone(),
+            max_lightweight_tokens: self.config.max_lightweight_tokens,
+        };
+
+        // Record the run in DB, then spawn execution
+        let store = self.store.clone();
+        tokio::spawn(async move {
+            if let Err(e) = store.create_routine_run(&run).await {
+                tracing::error!(routine = %routine.name, "Failed to record run: {}", e);
+                return;
+            }
+            execute_routine(engine, routine, run).await;
+        });
+    }
+
+    fn check_cooldown(&self, routine: &Routine) -> bool {
+        if let Some(last_run) = routine.last_run_at {
+            let elapsed = Utc::now().signed_duration_since(last_run);
+            let cooldown = chrono::Duration::from_std(routine.guardrails.cooldown)
+                .unwrap_or(chrono::Duration::seconds(300));
+            if elapsed < cooldown {
+                return false;
+            }
+        }
+        true
+    }
+
+    async fn check_concurrent(&self, routine: &Routine) -> bool {
+        match self.store.count_running_routine_runs(routine.id).await {
+            Ok(count) => count < routine.guardrails.max_concurrent as i64,
+            Err(e) => {
+                tracing::error!(
+                    routine = %routine.name,
+                    "Failed to check concurrent runs: {}", e
+                );
+                false
+            }
+        }
+    }
+}
+
+/// Shared context passed to the execution function.
+struct EngineContext {
+    store: Arc<Store>,
+    llm: Arc<dyn LlmProvider>,
+    workspace: Arc<Workspace>,
+    notify_tx: mpsc::Sender<OutgoingResponse>,
+    running_count: Arc<RwLock<usize>>,
+    max_lightweight_tokens: u32,
+}
+
+/// Execute a routine run. Handles both lightweight and full_job modes.
+async fn execute_routine(ctx: EngineContext, routine: Routine, run: RoutineRun) {
+    // Increment running count
+    {
+        let mut count = ctx.running_count.write().await;
+        *count += 1;
+    }
+
+    let result = match &routine.action {
+        RoutineAction::Lightweight {
+            prompt,
+            context_paths,
+            max_tokens,
+        } => execute_lightweight(&ctx, &routine, prompt, context_paths, *max_tokens).await,
+        RoutineAction::FullJob { description, .. } => {
+            // Full job mode: for now, execute as lightweight with the description
+            // as prompt. Full scheduler integration will come as a follow-up.
+            tracing::info!(
+                routine = %routine.name,
+                "FullJob mode executing as lightweight (scheduler integration pending)"
+            );
+            execute_lightweight(&ctx, &routine, description, &[], ctx.max_lightweight_tokens).await
+        }
+    };
+
+    // Decrement running count
+    {
+        let mut count = ctx.running_count.write().await;
+        *count = count.saturating_sub(1);
+    }
+
+    // Process result
+    let (status, summary, tokens) = match result {
+        Ok(execution) => execution,
+        Err(e) => {
+            tracing::error!(routine = %routine.name, "Execution failed: {}", e);
+            (RunStatus::Failed, Some(e), None)
+        }
+    };
+
+    // Complete the run record
+    if let Err(e) = ctx
+        .store
+        .complete_routine_run(run.id, status, summary.as_deref(), tokens)
+        .await
+    {
+        tracing::error!(routine = %routine.name, "Failed to complete run record: {}", e);
+    }
+
+    // Update routine runtime state
+    let now = Utc::now();
+    let next_fire = if let Trigger::Cron { ref schedule } = routine.trigger {
+        next_cron_fire(schedule).unwrap_or(None)
+    } else {
+        None
+    };
+
+    let new_failures = if status == RunStatus::Failed {
+        routine.consecutive_failures + 1
+    } else {
+        0
+    };
+
+    if let Err(e) = ctx
+        .store
+        .update_routine_runtime(
+            routine.id,
+            now,
+            next_fire,
+            routine.run_count + 1,
+            new_failures,
+            &routine.state,
+        )
+        .await
+    {
+        tracing::error!(routine = %routine.name, "Failed to update runtime state: {}", e);
+    }
+
+    // Send notifications based on config
+    send_notification(
+        &ctx.notify_tx,
+        &routine.notify,
+        &routine.name,
+        status,
+        summary.as_deref(),
+    )
+    .await;
+}
+
+/// Execute a lightweight routine (single LLM call).
+async fn execute_lightweight(
+    ctx: &EngineContext,
+    routine: &Routine,
+    prompt: &str,
+    context_paths: &[String],
+    max_tokens: u32,
+) -> Result<(RunStatus, Option<String>, Option<i32>), String> {
+    // Load context from workspace
+    let mut context_parts = Vec::new();
+    for path in context_paths {
+        match ctx.workspace.read(path).await {
+            Ok(doc) => {
+                context_parts.push(format!("## {}\n\n{}", path, doc.content));
+            }
+            Err(e) => {
+                tracing::debug!(
+                    routine = %routine.name,
+                    "Failed to read context path {}: {}", path, e
+                );
+            }
+        }
+    }
+
+    // Load routine state from workspace
+    let state_path = format!("routines/{}/state.md", routine.name);
+    let state_content = match ctx.workspace.read(&state_path).await {
+        Ok(doc) => Some(doc.content),
+        Err(_) => None,
+    };
+
+    // Build the prompt
+    let mut full_prompt = String::new();
+    full_prompt.push_str(prompt);
+
+    if !context_parts.is_empty() {
+        full_prompt.push_str("\n\n---\n\n# Context\n\n");
+        full_prompt.push_str(&context_parts.join("\n\n"));
+    }
+
+    if let Some(state) = &state_content {
+        full_prompt.push_str("\n\n---\n\n# Previous State\n\n");
+        full_prompt.push_str(state);
+    }
+
+    full_prompt.push_str(
+        "\n\n---\n\nIf nothing needs attention, reply EXACTLY with: ROUTINE_OK\n\
+         If something needs attention, provide a concise summary.",
+    );
+
+    // Get system prompt
+    let system_prompt = match ctx.workspace.system_prompt().await {
+        Ok(p) => p,
+        Err(e) => {
+            tracing::warn!(routine = %routine.name, "Failed to get system prompt: {}", e);
+            String::new()
+        }
+    };
+
+    let messages = if system_prompt.is_empty() {
+        vec![ChatMessage::user(&full_prompt)]
+    } else {
+        vec![
+            ChatMessage::system(&system_prompt),
+            ChatMessage::user(&full_prompt),
+        ]
+    };
+
+    // Determine max_tokens from model metadata with fallback
+    let effective_max_tokens = match ctx.llm.model_metadata().await {
+        Ok(meta) => {
+            let from_api = meta.context_length.map(|ctx| ctx / 2).unwrap_or(max_tokens);
+            from_api.max(max_tokens)
+        }
+        Err(_) => max_tokens,
+    };
+
+    let request = CompletionRequest::new(messages)
+        .with_max_tokens(effective_max_tokens)
+        .with_temperature(0.3);
+
+    let response = ctx
+        .llm
+        .complete(request)
+        .await
+        .map_err(|e| format!("LLM call failed: {e}"))?;
+
+    let content = response.content.trim();
+    let tokens_used = Some((response.input_tokens + response.output_tokens) as i32);
+
+    // Empty content guard (same as heartbeat)
+    if content.is_empty() {
+        return if response.finish_reason == FinishReason::Length {
+            Err(
+                "LLM response truncated (finish_reason=length) with no content. \
+                 Model may have exhausted token budget on reasoning."
+                    .to_string(),
+            )
+        } else {
+            Err("LLM returned empty content.".to_string())
+        };
+    }
+
+    // Check for the "nothing to do" sentinel
+    if content == "ROUTINE_OK" || content.contains("ROUTINE_OK") {
+        return Ok((RunStatus::Ok, None, tokens_used));
+    }
+
+    Ok((RunStatus::Attention, Some(content.to_string()), tokens_used))
+}
+
+/// Send a notification based on the routine's notify config and run status.
+async fn send_notification(
+    tx: &mpsc::Sender<OutgoingResponse>,
+    notify: &NotifyConfig,
+    routine_name: &str,
+    status: RunStatus,
+    summary: Option<&str>,
+) {
+    let should_notify = match status {
+        RunStatus::Ok => notify.on_success,
+        RunStatus::Attention => notify.on_attention,
+        RunStatus::Failed => notify.on_failure,
+        RunStatus::Running => false,
+    };
+
+    if !should_notify {
+        return;
+    }
+
+    let icon = match status {
+        RunStatus::Ok => "✅",
+        RunStatus::Attention => "🔔",
+        RunStatus::Failed => "❌",
+        RunStatus::Running => "⏳",
+    };
+
+    let message = match summary {
+        Some(s) => format!("{} *Routine '{}'*: {}\n\n{}", icon, routine_name, status, s),
+        None => format!("{} *Routine '{}'*: {}", icon, routine_name, status),
+    };
+
+    let response = OutgoingResponse {
+        content: message,
+        thread_id: None,
+        metadata: serde_json::json!({
+            "source": "routine",
+            "routine_name": routine_name,
+            "status": status.to_string(),
+        }),
+    };
+
+    if let Err(e) = tx.send(response).await {
+        tracing::error!(routine = %routine_name, "Failed to send notification: {}", e);
+    }
+}
+
+/// Spawn the cron ticker background task.
+pub fn spawn_cron_ticker(
+    engine: Arc<RoutineEngine>,
+    interval: Duration,
+) -> tokio::task::JoinHandle<()> {
+    tokio::spawn(async move {
+        let mut ticker = tokio::time::interval(interval);
+        // Skip immediate first tick
+        ticker.tick().await;
+
+        loop {
+            ticker.tick().await;
+            engine.check_cron_triggers().await;
+        }
+    })
+}
+
+fn truncate(s: &str, max: usize) -> String {
+    if s.len() <= max {
+        s.to_string()
+    } else {
+        format!("{}...", &s[..max])
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::agent::routine::{NotifyConfig, RunStatus};
+
+    #[test]
+    fn test_notification_gating() {
+        let config = NotifyConfig {
+            on_success: false,
+            on_failure: true,
+            on_attention: true,
+            ..Default::default()
+        };
+
+        // on_success = false means Ok status should not notify
+        assert!(!config.on_success);
+        assert!(config.on_failure);
+        assert!(config.on_attention);
+    }
+
+    #[test]
+    fn test_run_status_icons() {
+        // Just verify the mapping doesn't panic
+        for status in [
+            RunStatus::Ok,
+            RunStatus::Attention,
+            RunStatus::Failed,
+            RunStatus::Running,
+        ] {
+            let _ = status.to_string();
+        }
+    }
+}

src/agent/scheduler.rs

@@ -373,23 +373,23 @@ impl Scheduler {
             .into());
         }
 
-        // Execute with timeout
-        let result = tokio::time::timeout(Duration::from_secs(60), async {
-            tool.execute(params, &job_ctx).await
-        })
-        .await
-        .map_err(|_| {
-            Error::Tool(crate::error::ToolError::Timeout {
-                name: tool_name.to_string(),
-                timeout: Duration::from_secs(60),
-            })
-        })?
-        .map_err(|e| {
-            Error::Tool(crate::error::ToolError::ExecutionFailed {
-                name: tool_name.to_string(),
-                reason: e.to_string(),
-            })
-        })?;
+        // Execute with per-tool timeout
+        let tool_timeout = tool.execution_timeout();
+        let result =
+            tokio::time::timeout(tool_timeout, async { tool.execute(params, &job_ctx).await })
+                .await
+                .map_err(|_| {
+                    Error::Tool(crate::error::ToolError::Timeout {
+                        name: tool_name.to_string(),
+                        timeout: tool_timeout,
+                    })
+                })?
+                .map_err(|e| {
+                    Error::Tool(crate::error::ToolError::ExecutionFailed {
+                        name: tool_name.to_string(),
+                        reason: e.to_string(),
+                    })
+                })?;
 
         Ok(TaskOutput::new(result.result, start.elapsed()))
     }

src/agent/session.rs

@@ -173,6 +173,10 @@ pub struct Thread {
     /// Pending auth token request (thread is in auth mode).
     #[serde(default)]
     pub pending_auth: Option<PendingAuth>,
+    /// Last NEAR AI response ID for response chaining. Persisted to DB
+    /// metadata so we can resume chaining across restarts.
+    #[serde(default)]
+    pub last_response_id: Option<String>,
 }
 
 impl Thread {
@@ -189,6 +193,24 @@ impl Thread {
             metadata: serde_json::Value::Null,
             pending_approval: None,
             pending_auth: None,
+            last_response_id: None,
+        }
+    }
+
+    /// Create a thread with a specific ID (for DB hydration).
+    pub fn with_id(id: Uuid, session_id: Uuid) -> Self {
+        let now = Utc::now();
+        Self {
+            id,
+            session_id,
+            state: ThreadState::Idle,
+            turns: Vec::new(),
+            created_at: now,
+            updated_at: now,
+            metadata: serde_json::Value::Null,
+            pending_approval: None,
+            pending_auth: None,
+            last_response_id: None,
         }
     }
 
@@ -593,4 +615,386 @@ mod tests {
         let restored: Thread = serde_json::from_str(&json).expect("should deserialize");
         assert!(restored.pending_auth.is_none());
     }
+
+    #[test]
+    fn test_thread_with_id() {
+        let specific_id = Uuid::new_v4();
+        let session_id = Uuid::new_v4();
+        let thread = Thread::with_id(specific_id, session_id);
+
+        assert_eq!(thread.id, specific_id);
+        assert_eq!(thread.session_id, session_id);
+        assert_eq!(thread.state, ThreadState::Idle);
+        assert!(thread.turns.is_empty());
+    }
+
+    #[test]
+    fn test_thread_with_id_restore_messages() {
+        let thread_id = Uuid::new_v4();
+        let session_id = Uuid::new_v4();
+        let mut thread = Thread::with_id(thread_id, session_id);
+
+        let messages = vec![
+            ChatMessage::user("Hello from DB"),
+            ChatMessage::assistant("Restored response"),
+        ];
+        thread.restore_from_messages(messages);
+
+        assert_eq!(thread.id, thread_id);
+        assert_eq!(thread.turns.len(), 1);
+        assert_eq!(thread.turns[0].user_input, "Hello from DB");
+        assert_eq!(
+            thread.turns[0].response,
+            Some("Restored response".to_string())
+        );
+    }
+
+    #[test]
+    fn test_restore_from_messages_empty() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Add a turn first, then restore with empty vec
+        thread.start_turn("hello");
+        thread.complete_turn("hi");
+        assert_eq!(thread.turns.len(), 1);
+
+        thread.restore_from_messages(Vec::new());
+
+        // Should clear all turns and stay idle
+        assert!(thread.turns.is_empty());
+        assert_eq!(thread.state, ThreadState::Idle);
+    }
+
+    #[test]
+    fn test_restore_from_messages_only_assistant_messages() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Only assistant messages (no user messages to anchor turns)
+        let messages = vec![
+            ChatMessage::assistant("I'm here"),
+            ChatMessage::assistant("Still here"),
+        ];
+
+        thread.restore_from_messages(messages);
+
+        // Assistant-only messages have no user turn to attach to, so
+        // they should be skipped entirely.
+        assert!(thread.turns.is_empty());
+    }
+
+    #[test]
+    fn test_restore_from_messages_multiple_user_messages_in_a_row() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Two user messages with no assistant response between them
+        let messages = vec![
+            ChatMessage::user("first"),
+            ChatMessage::user("second"),
+            ChatMessage::assistant("reply to second"),
+        ];
+
+        thread.restore_from_messages(messages);
+
+        // First user message becomes a turn with no response,
+        // second user message pairs with the assistant response.
+        assert_eq!(thread.turns.len(), 2);
+        assert_eq!(thread.turns[0].user_input, "first");
+        assert!(thread.turns[0].response.is_none());
+        assert_eq!(thread.turns[1].user_input, "second");
+        assert_eq!(
+            thread.turns[1].response,
+            Some("reply to second".to_string())
+        );
+    }
+
+    #[test]
+    fn test_thread_switch() {
+        let mut session = Session::new("user-1");
+
+        let t1_id = session.create_thread().id;
+        let t2_id = session.create_thread().id;
+
+        // After creating two threads, active should be the last one
+        assert_eq!(session.active_thread, Some(t2_id));
+
+        // Switch back to the first
+        assert!(session.switch_thread(t1_id));
+        assert_eq!(session.active_thread, Some(t1_id));
+
+        // Switching to a nonexistent thread should fail
+        let fake_id = Uuid::new_v4();
+        assert!(!session.switch_thread(fake_id));
+        // Active thread should remain unchanged
+        assert_eq!(session.active_thread, Some(t1_id));
+    }
+
+    #[test]
+    fn test_get_or_create_thread_idempotent() {
+        let mut session = Session::new("user-1");
+
+        let tid1 = session.get_or_create_thread().id;
+        let tid2 = session.get_or_create_thread().id;
+
+        // Should return the same thread (not create a new one each time)
+        assert_eq!(tid1, tid2);
+        assert_eq!(session.threads.len(), 1);
+    }
+
+    #[test]
+    fn test_truncate_turns() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        for i in 0..5 {
+            thread.start_turn(format!("msg-{}", i));
+            thread.complete_turn(format!("resp-{}", i));
+        }
+        assert_eq!(thread.turns.len(), 5);
+
+        thread.truncate_turns(3);
+        assert_eq!(thread.turns.len(), 3);
+
+        // Should keep the most recent turns
+        assert_eq!(thread.turns[0].user_input, "msg-2");
+        assert_eq!(thread.turns[1].user_input, "msg-3");
+        assert_eq!(thread.turns[2].user_input, "msg-4");
+
+        // Turn numbers should be re-indexed
+        assert_eq!(thread.turns[0].turn_number, 0);
+        assert_eq!(thread.turns[1].turn_number, 1);
+        assert_eq!(thread.turns[2].turn_number, 2);
+    }
+
+    #[test]
+    fn test_truncate_turns_noop_when_fewer() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        thread.start_turn("only one");
+        thread.complete_turn("response");
+
+        thread.truncate_turns(10);
+        assert_eq!(thread.turns.len(), 1);
+        assert_eq!(thread.turns[0].user_input, "only one");
+    }
+
+    #[test]
+    fn test_thread_interrupt_and_resume() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        thread.start_turn("do something");
+        assert_eq!(thread.state, ThreadState::Processing);
+
+        thread.interrupt();
+        assert_eq!(thread.state, ThreadState::Interrupted);
+
+        let last_turn = thread.last_turn().unwrap();
+        assert_eq!(last_turn.state, TurnState::Interrupted);
+        assert!(last_turn.completed_at.is_some());
+
+        thread.resume();
+        assert_eq!(thread.state, ThreadState::Idle);
+    }
+
+    #[test]
+    fn test_resume_only_from_interrupted() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Idle thread: resume should be a no-op
+        assert_eq!(thread.state, ThreadState::Idle);
+        thread.resume();
+        assert_eq!(thread.state, ThreadState::Idle);
+
+        // Processing thread: resume should not change state
+        thread.start_turn("work");
+        assert_eq!(thread.state, ThreadState::Processing);
+        thread.resume();
+        assert_eq!(thread.state, ThreadState::Processing);
+    }
+
+    #[test]
+    fn test_turn_fail() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        thread.start_turn("risky operation");
+        thread.fail_turn("connection timed out");
+
+        assert_eq!(thread.state, ThreadState::Idle);
+
+        let turn = thread.last_turn().unwrap();
+        assert_eq!(turn.state, TurnState::Failed);
+        assert_eq!(turn.error, Some("connection timed out".to_string()));
+        assert!(turn.response.is_none());
+        assert!(turn.completed_at.is_some());
+    }
+
+    #[test]
+    fn test_messages_with_incomplete_last_turn() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        thread.start_turn("first");
+        thread.complete_turn("first reply");
+        thread.start_turn("second (in progress)");
+
+        let messages = thread.messages();
+        // Should have 3 messages: user, assistant, user (no assistant for in-progress)
+        assert_eq!(messages.len(), 3);
+        assert_eq!(messages[0].content, "first");
+        assert_eq!(messages[1].content, "first reply");
+        assert_eq!(messages[2].content, "second (in progress)");
+    }
+
+    #[test]
+    fn test_thread_serialization_round_trip() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        thread.start_turn("hello");
+        thread.complete_turn("world");
+        thread.last_response_id = Some("resp_abc123".to_string());
+
+        let json = serde_json::to_string(&thread).unwrap();
+        let restored: Thread = serde_json::from_str(&json).unwrap();
+
+        assert_eq!(restored.id, thread.id);
+        assert_eq!(restored.session_id, thread.session_id);
+        assert_eq!(restored.turns.len(), 1);
+        assert_eq!(restored.turns[0].user_input, "hello");
+        assert_eq!(restored.turns[0].response, Some("world".to_string()));
+        assert_eq!(restored.last_response_id, Some("resp_abc123".to_string()));
+    }
+
+    #[test]
+    fn test_session_serialization_round_trip() {
+        let mut session = Session::new("user-ser");
+        session.create_thread();
+        session.auto_approve_tool("echo");
+
+        let json = serde_json::to_string(&session).unwrap();
+        let restored: Session = serde_json::from_str(&json).unwrap();
+
+        assert_eq!(restored.user_id, "user-ser");
+        assert_eq!(restored.threads.len(), 1);
+        assert!(restored.is_tool_auto_approved("echo"));
+        assert!(!restored.is_tool_auto_approved("shell"));
+    }
+
+    #[test]
+    fn test_auto_approved_tools() {
+        let mut session = Session::new("user-1");
+
+        assert!(!session.is_tool_auto_approved("shell"));
+        session.auto_approve_tool("shell");
+        assert!(session.is_tool_auto_approved("shell"));
+
+        // Idempotent
+        session.auto_approve_tool("shell");
+        assert_eq!(session.auto_approved_tools.len(), 1);
+    }
+
+    #[test]
+    fn test_turn_tool_call_error() {
+        let mut turn = Turn::new(0, "test");
+        turn.record_tool_call("http", serde_json::json!({"url": "example.com"}));
+        turn.record_tool_error("timeout");
+
+        assert_eq!(turn.tool_calls.len(), 1);
+        assert_eq!(turn.tool_calls[0].error, Some("timeout".to_string()));
+        assert!(turn.tool_calls[0].result.is_none());
+    }
+
+    #[test]
+    fn test_turn_number_increments() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Before any turns, turn_number() is 1 (1-indexed for display)
+        assert_eq!(thread.turn_number(), 1);
+
+        thread.start_turn("first");
+        thread.complete_turn("done");
+        assert_eq!(thread.turn_number(), 2);
+
+        thread.start_turn("second");
+        assert_eq!(thread.turn_number(), 3);
+    }
+
+    #[test]
+    fn test_complete_turn_on_empty_thread() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Completing a turn when there are no turns should be a safe no-op
+        thread.complete_turn("phantom response");
+        assert_eq!(thread.state, ThreadState::Idle);
+        assert!(thread.turns.is_empty());
+    }
+
+    #[test]
+    fn test_fail_turn_on_empty_thread() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        // Failing a turn when there are no turns should be a safe no-op
+        thread.fail_turn("phantom error");
+        assert_eq!(thread.state, ThreadState::Idle);
+        assert!(thread.turns.is_empty());
+    }
+
+    #[test]
+    fn test_pending_approval_flow() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        let approval = PendingApproval {
+            request_id: Uuid::new_v4(),
+            tool_name: "shell".to_string(),
+            parameters: serde_json::json!({"command": "rm -rf /"}),
+            description: "dangerous command".to_string(),
+            tool_call_id: "call_123".to_string(),
+            context_messages: vec![ChatMessage::user("do it")],
+        };
+
+        thread.await_approval(approval);
+        assert_eq!(thread.state, ThreadState::AwaitingApproval);
+        assert!(thread.pending_approval.is_some());
+
+        let taken = thread.take_pending_approval();
+        assert!(taken.is_some());
+        assert_eq!(taken.unwrap().tool_name, "shell");
+        assert!(thread.pending_approval.is_none());
+    }
+
+    #[test]
+    fn test_clear_pending_approval() {
+        let mut thread = Thread::new(Uuid::new_v4());
+
+        let approval = PendingApproval {
+            request_id: Uuid::new_v4(),
+            tool_name: "http".to_string(),
+            parameters: serde_json::json!({}),
+            description: "test".to_string(),
+            tool_call_id: "call_456".to_string(),
+            context_messages: vec![],
+        };
+
+        thread.await_approval(approval);
+        thread.clear_pending_approval();
+
+        assert_eq!(thread.state, ThreadState::Idle);
+        assert!(thread.pending_approval.is_none());
+    }
+
+    #[test]
+    fn test_active_thread_accessors() {
+        let mut session = Session::new("user-1");
+
+        assert!(session.active_thread().is_none());
+        assert!(session.active_thread_mut().is_none());
+
+        let tid = session.create_thread().id;
+
+        assert!(session.active_thread().is_some());
+        assert_eq!(session.active_thread().unwrap().id, tid);
+
+        // Mutably modify through accessor
+        session.active_thread_mut().unwrap().start_turn("test");
+        assert_eq!(
+            session.active_thread().unwrap().state,
+            ThreadState::Processing
+        );
+    }
 }

src/agent/session_manager.rs

@@ -110,6 +110,41 @@ impl SessionManager {
         (session, thread_id)
     }
 
+    /// Register a hydrated thread so subsequent `resolve_thread` calls find it.
+    ///
+    /// Inserts into the thread_map and creates an undo manager for the thread.
+    pub async fn register_thread(
+        &self,
+        user_id: &str,
+        channel: &str,
+        thread_id: Uuid,
+        session: Arc<Mutex<Session>>,
+    ) {
+        let key = ThreadKey {
+            user_id: user_id.to_string(),
+            channel: channel.to_string(),
+            external_thread_id: Some(thread_id.to_string()),
+        };
+
+        {
+            let mut thread_map = self.thread_map.write().await;
+            thread_map.insert(key, thread_id);
+        }
+
+        {
+            let mut undo_managers = self.undo_managers.write().await;
+            undo_managers
+                .entry(thread_id)
+                .or_insert_with(|| Arc::new(Mutex::new(UndoManager::new())));
+        }
+
+        // Ensure the session is tracked
+        {
+            let mut sessions = self.sessions.write().await;
+            sessions.entry(user_id.to_string()).or_insert(session);
+        }
+    }
+
     /// Get undo manager for a thread.
     pub async fn get_undo_manager(&self, thread_id: Uuid) -> Arc<Mutex<UndoManager>> {
         // Fast path
@@ -296,4 +331,344 @@ mod tests {
             .await;
         assert_eq!(pruned, 0);
     }
+
+    #[tokio::test]
+    async fn test_register_thread() {
+        use crate::agent::session::{Session, Thread};
+
+        let manager = SessionManager::new();
+        let thread_id = Uuid::new_v4();
+
+        // Create a session with a hydrated thread
+        let session = Arc::new(Mutex::new(Session::new("user-hydrate")));
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(thread_id, sess.id);
+            sess.threads.insert(thread_id, thread);
+            sess.active_thread = Some(thread_id);
+        }
+
+        // Register the thread
+        manager
+            .register_thread("user-hydrate", "gateway", thread_id, Arc::clone(&session))
+            .await;
+
+        // resolve_thread should find it (using the UUID as external_thread_id)
+        let (resolved_session, resolved_tid) = manager
+            .resolve_thread("user-hydrate", "gateway", Some(&thread_id.to_string()))
+            .await;
+        assert_eq!(resolved_tid, thread_id);
+
+        // Should be the same session object
+        let sess = resolved_session.lock().await;
+        assert!(sess.threads.contains_key(&thread_id));
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_with_explicit_external_id() {
+        let manager = SessionManager::new();
+
+        // Two calls with the same explicit external thread ID should resolve
+        // to the same internal thread.
+        let (_, t1) = manager
+            .resolve_thread("user-1", "gateway", Some("ext-abc"))
+            .await;
+        let (_, t2) = manager
+            .resolve_thread("user-1", "gateway", Some("ext-abc"))
+            .await;
+        assert_eq!(t1, t2);
+
+        // A different external ID on the same channel/user gets a new thread.
+        let (_, t3) = manager
+            .resolve_thread("user-1", "gateway", Some("ext-xyz"))
+            .await;
+        assert_ne!(t1, t3);
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_none_vs_some_external_id() {
+        let manager = SessionManager::new();
+
+        // None external_thread_id is a distinct key from Some("ext-1").
+        let (_, t_none) = manager.resolve_thread("user-1", "cli", None).await;
+        let (_, t_some) = manager.resolve_thread("user-1", "cli", Some("ext-1")).await;
+        assert_ne!(t_none, t_some);
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_different_users_isolated() {
+        let manager = SessionManager::new();
+
+        let (_, t1) = manager
+            .resolve_thread("user-a", "gateway", Some("same-ext"))
+            .await;
+        let (_, t2) = manager
+            .resolve_thread("user-b", "gateway", Some("same-ext"))
+            .await;
+
+        // Same channel + same external ID but different users = different threads
+        assert_ne!(t1, t2);
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_different_channels_isolated() {
+        let manager = SessionManager::new();
+
+        let (_, t1) = manager
+            .resolve_thread("user-1", "gateway", Some("thread-x"))
+            .await;
+        let (_, t2) = manager
+            .resolve_thread("user-1", "telegram", Some("thread-x"))
+            .await;
+
+        // Same user + same external ID but different channels = different threads
+        assert_ne!(t1, t2);
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_stale_mapping_creates_new_thread() {
+        let manager = SessionManager::new();
+
+        // Create a thread normally
+        let (session, original_tid) = manager
+            .resolve_thread("user-1", "gateway", Some("ext-1"))
+            .await;
+
+        // Simulate the thread being removed from the session (e.g. pruned)
+        {
+            let mut sess = session.lock().await;
+            sess.threads.remove(&original_tid);
+        }
+
+        // Next resolve should detect the stale mapping and create a fresh thread
+        let (_, new_tid) = manager
+            .resolve_thread("user-1", "gateway", Some("ext-1"))
+            .await;
+        assert_ne!(original_tid, new_tid);
+
+        // The new thread should actually exist in the session
+        let sess = session.lock().await;
+        assert!(sess.threads.contains_key(&new_tid));
+    }
+
+    #[tokio::test]
+    async fn test_register_thread_preserves_uuid_on_resolve() {
+        use crate::agent::session::{Session, Thread};
+
+        let manager = SessionManager::new();
+        let known_uuid = Uuid::new_v4();
+
+        let session = Arc::new(Mutex::new(Session::new("user-web")));
+        let session_id = {
+            let sess = session.lock().await;
+            sess.id
+        };
+
+        // Simulate hydration: create thread with a known UUID
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(known_uuid, session_id);
+            sess.threads.insert(known_uuid, thread);
+        }
+
+        // Register it
+        manager
+            .register_thread("user-web", "gateway", known_uuid, Arc::clone(&session))
+            .await;
+
+        // resolve_thread with UUID as external_thread_id MUST return the same UUID,
+        // not mint a new one (this was the root cause of the "wrong conversation" bug)
+        let (_, resolved) = manager
+            .resolve_thread("user-web", "gateway", Some(&known_uuid.to_string()))
+            .await;
+        assert_eq!(resolved, known_uuid);
+    }
+
+    #[tokio::test]
+    async fn test_register_thread_idempotent() {
+        use crate::agent::session::{Session, Thread};
+
+        let manager = SessionManager::new();
+        let tid = Uuid::new_v4();
+
+        let session = Arc::new(Mutex::new(Session::new("user-idem")));
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(tid, sess.id);
+            sess.threads.insert(tid, thread);
+        }
+
+        // Register twice
+        manager
+            .register_thread("user-idem", "gateway", tid, Arc::clone(&session))
+            .await;
+        manager
+            .register_thread("user-idem", "gateway", tid, Arc::clone(&session))
+            .await;
+
+        // Should still resolve to the same thread
+        let (_, resolved) = manager
+            .resolve_thread("user-idem", "gateway", Some(&tid.to_string()))
+            .await;
+        assert_eq!(resolved, tid);
+    }
+
+    #[tokio::test]
+    async fn test_register_thread_creates_undo_manager() {
+        use crate::agent::session::{Session, Thread};
+
+        let manager = SessionManager::new();
+        let tid = Uuid::new_v4();
+
+        let session = Arc::new(Mutex::new(Session::new("user-undo")));
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(tid, sess.id);
+            sess.threads.insert(tid, thread);
+        }
+
+        manager
+            .register_thread("user-undo", "gateway", tid, Arc::clone(&session))
+            .await;
+
+        // Undo manager should exist for the registered thread
+        let undo = manager.get_undo_manager(tid).await;
+        let undo2 = manager.get_undo_manager(tid).await;
+        assert!(Arc::ptr_eq(&undo, &undo2));
+    }
+
+    #[tokio::test]
+    async fn test_register_thread_stores_session() {
+        use crate::agent::session::{Session, Thread};
+
+        let manager = SessionManager::new();
+        let tid = Uuid::new_v4();
+
+        let session = Arc::new(Mutex::new(Session::new("user-new")));
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(tid, sess.id);
+            sess.threads.insert(tid, thread);
+        }
+
+        // The user has no session yet in the manager
+        {
+            let sessions = manager.sessions.read().await;
+            assert!(!sessions.contains_key("user-new"));
+        }
+
+        manager
+            .register_thread("user-new", "gateway", tid, Arc::clone(&session))
+            .await;
+
+        // Now the session should be tracked
+        {
+            let sessions = manager.sessions.read().await;
+            assert!(sessions.contains_key("user-new"));
+        }
+    }
+
+    #[tokio::test]
+    async fn test_multiple_threads_per_user() {
+        let manager = SessionManager::new();
+
+        let (_, t1) = manager
+            .resolve_thread("user-1", "gateway", Some("thread-a"))
+            .await;
+        let (_, t2) = manager
+            .resolve_thread("user-1", "gateway", Some("thread-b"))
+            .await;
+        let (session, t3) = manager
+            .resolve_thread("user-1", "gateway", Some("thread-c"))
+            .await;
+
+        // All three should be distinct
+        assert_ne!(t1, t2);
+        assert_ne!(t2, t3);
+        assert_ne!(t1, t3);
+
+        // All three should exist in the same session
+        let sess = session.lock().await;
+        assert!(sess.threads.contains_key(&t1));
+        assert!(sess.threads.contains_key(&t2));
+        assert!(sess.threads.contains_key(&t3));
+    }
+
+    #[tokio::test]
+    async fn test_prune_cleans_thread_map_and_undo_managers() {
+        let manager = SessionManager::new();
+
+        let (stale_session, stale_tid) = manager.resolve_thread("user-stale", "cli", None).await;
+
+        // Backdate the session
+        {
+            let mut sess = stale_session.lock().await;
+            sess.last_active_at = chrono::Utc::now() - chrono::TimeDelta::seconds(86400 * 30);
+        }
+
+        // Verify thread_map and undo_managers have entries
+        {
+            let tm = manager.thread_map.read().await;
+            assert!(!tm.is_empty());
+        }
+        {
+            let um = manager.undo_managers.read().await;
+            assert!(um.contains_key(&stale_tid));
+        }
+
+        let pruned = manager
+            .prune_stale_sessions(std::time::Duration::from_secs(86400 * 7))
+            .await;
+        assert_eq!(pruned, 1);
+
+        // Thread map and undo managers should be cleaned up
+        {
+            let tm = manager.thread_map.read().await;
+            assert!(tm.is_empty());
+        }
+        {
+            let um = manager.undo_managers.read().await;
+            assert!(!um.contains_key(&stale_tid));
+        }
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_active_thread_set() {
+        let manager = SessionManager::new();
+
+        let (session, thread_id) = manager
+            .resolve_thread("user-1", "gateway", Some("ext-1"))
+            .await;
+
+        // The resolved thread should be set as the active thread
+        let sess = session.lock().await;
+        assert_eq!(sess.active_thread, Some(thread_id));
+    }
+
+    #[tokio::test]
+    async fn test_register_then_resolve_different_channel_creates_new() {
+        use crate::agent::session::{Session, Thread};
+
+        let manager = SessionManager::new();
+        let tid = Uuid::new_v4();
+
+        let session = Arc::new(Mutex::new(Session::new("user-cross")));
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(tid, sess.id);
+            sess.threads.insert(tid, thread);
+        }
+
+        // Register on "gateway" channel
+        manager
+            .register_thread("user-cross", "gateway", tid, Arc::clone(&session))
+            .await;
+
+        // Resolve on a different channel with the same UUID string should NOT
+        // find the registered thread (channel is part of the key)
+        let (_, resolved) = manager
+            .resolve_thread("user-cross", "telegram", Some(&tid.to_string()))
+            .await;
+        assert_ne!(resolved, tid);
+    }
 }

src/agent/submission.rs

@@ -43,6 +43,49 @@ impl SubmissionParser {
         if lower == "/thread new" || lower == "/new" {
             return Submission::NewThread;
         }
+        // System commands (bypass thread-state checks)
+        if lower == "/help" || lower == "/?" {
+            return Submission::SystemCommand {
+                command: "help".to_string(),
+                args: vec![],
+            };
+        }
+        if lower == "/version" {
+            return Submission::SystemCommand {
+                command: "version".to_string(),
+                args: vec![],
+            };
+        }
+        if lower == "/tools" {
+            return Submission::SystemCommand {
+                command: "tools".to_string(),
+                args: vec![],
+            };
+        }
+        if lower == "/ping" {
+            return Submission::SystemCommand {
+                command: "ping".to_string(),
+                args: vec![],
+            };
+        }
+        if lower == "/debug" {
+            return Submission::SystemCommand {
+                command: "debug".to_string(),
+                args: vec![],
+            };
+        }
+        if lower.starts_with("/model") {
+            let args: Vec<String> = trimmed
+                .split_whitespace()
+                .skip(1)
+                .map(|s| s.to_string())
+                .collect();
+            return Submission::SystemCommand {
+                command: "model".to_string(),
+                args,
+            };
+        }
+
         if lower == "/quit" || lower == "/exit" || lower == "/shutdown" {
             return Submission::Quit;
         }
@@ -172,6 +215,15 @@ pub enum Submission {
 
     /// Quit the agent. Bypasses thread-state checks.
     Quit,
+
+    /// System command (help, model, version, tools, ping, debug).
+    /// Bypasses thread-state checks and safety validation.
+    SystemCommand {
+        /// The command name (e.g. "help", "model", "version").
+        command: String,
+        /// Arguments to the command.
+        args: Vec<String>,
+    },
 }
 
 impl Submission {
@@ -238,6 +290,7 @@ impl Submission {
                 | Self::Heartbeat
                 | Self::Summarize
                 | Self::Suggest
+                | Self::SystemCommand { .. }
         )
     }
 }
@@ -504,6 +557,84 @@ mod tests {
         );
     }
 
+    #[test]
+    fn test_parser_system_command_help() {
+        let submission = SubmissionParser::parse("/help");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "help" && args.is_empty())
+        );
+
+        let submission = SubmissionParser::parse("/?");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, .. } if command == "help")
+        );
+
+        let submission = SubmissionParser::parse("/HELP");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, .. } if command == "help")
+        );
+    }
+
+    #[test]
+    fn test_parser_system_command_model() {
+        // No args: show current model
+        let submission = SubmissionParser::parse("/model");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "model" && args.is_empty())
+        );
+
+        // With args: switch model
+        let submission = SubmissionParser::parse("/model gpt-4o");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "model" && args == vec!["gpt-4o"])
+        );
+
+        // Case insensitive command, preserves arg case
+        let submission = SubmissionParser::parse("/MODEL Claude-3.5");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "model" && args == vec!["Claude-3.5"])
+        );
+    }
+
+    #[test]
+    fn test_parser_system_command_version() {
+        let submission = SubmissionParser::parse("/version");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "version" && args.is_empty())
+        );
+    }
+
+    #[test]
+    fn test_parser_system_command_tools() {
+        let submission = SubmissionParser::parse("/tools");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "tools" && args.is_empty())
+        );
+    }
+
+    #[test]
+    fn test_parser_system_command_ping() {
+        let submission = SubmissionParser::parse("/ping");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "ping" && args.is_empty())
+        );
+    }
+
+    #[test]
+    fn test_parser_system_command_debug() {
+        let submission = SubmissionParser::parse("/debug");
+        assert!(
+            matches!(submission, Submission::SystemCommand { command, args } if command == "debug" && args.is_empty())
+        );
+    }
+
+    #[test]
+    fn test_parser_system_command_is_control() {
+        let submission = SubmissionParser::parse("/help");
+        assert!(submission.is_control());
+        assert!(!submission.starts_turn());
+    }
+
     #[test]
     fn test_parser_quit() {
         assert!(matches!(SubmissionParser::parse("/quit"), Submission::Quit));

src/agent/worker.rs

@@ -272,7 +272,10 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
                             ));
                         }
                     }
-                    RespondResult::ToolCalls(tool_calls) => {
+                    RespondResult::ToolCalls {
+                        tool_calls,
+                        content,
+                    } => {
                         // Model returned tool calls - execute them
                         tracing::debug!(
                             "Job {} respond_with_tools returned {} tool calls",
@@ -280,6 +283,14 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
                             tool_calls.len()
                         );
 
+                        // Add assistant message with tool_calls (OpenAI protocol)
+                        reason_ctx
+                            .messages
+                            .push(ChatMessage::assistant_with_tool_calls(
+                                content,
+                                tool_calls.clone(),
+                            ));
+
                         for tc in tool_calls {
                             let result = self.execute_tool(&tc.name, &tc.arguments).await;
 
@@ -417,14 +428,51 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
             .into());
         }
 
-        // Execute with timeout and timing
+        tracing::debug!(
+            tool = %tool_name,
+            params = %params,
+            job = %job_id,
+            "Tool call started"
+        );
+
+        // Execute with per-tool timeout and timing
+        let tool_timeout = tool.execution_timeout();
         let start = std::time::Instant::now();
-        let result = tokio::time::timeout(Duration::from_secs(60), async {
+        let result = tokio::time::timeout(tool_timeout, async {
             tool.execute(params.clone(), &job_ctx).await
         })
         .await;
         let elapsed = start.elapsed();
 
+        match &result {
+            Ok(Ok(output)) => {
+                let result_str = serde_json::to_string(&output.result)
+                    .unwrap_or_else(|_| "<serialize error>".to_string());
+                tracing::debug!(
+                    tool = %tool_name,
+                    elapsed_ms = elapsed.as_millis() as u64,
+                    result = %result_str,
+                    "Tool call succeeded"
+                );
+            }
+            Ok(Err(e)) => {
+                tracing::debug!(
+                    tool = %tool_name,
+                    elapsed_ms = elapsed.as_millis() as u64,
+                    error = %e,
+                    "Tool call failed"
+                );
+            }
+            Err(_) => {
+                tracing::debug!(
+                    tool = %tool_name,
+                    elapsed_ms = elapsed.as_millis() as u64,
+                    timeout_secs = tool_timeout.as_secs(),
+                    "Tool call timed out"
+                );
+            }
+        }
+
         // Record action in memory and get the ActionRecord for persistence
         let action = match &result {
             Ok(Ok(output)) => {
@@ -479,7 +527,7 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
         let output = result
             .map_err(|_| crate::error::ToolError::Timeout {
                 name: tool_name.to_string(),
-                timeout: Duration::from_secs(60),
+                timeout: tool_timeout,
             })?
             .map_err(|e| crate::error::ToolError::ExecutionFailed {
                 name: tool_name.to_string(),

src/bootstrap.rs

@@ -0,0 +1,325 @@
+//! Bootstrap configuration for IronClaw.
+//!
+//! These are the only settings that MUST live on disk because they're needed
+//! before the database connection is established. Everything else lives in the
+//! `settings` table in PostgreSQL.
+//!
+//! File: `~/.ironclaw/bootstrap.json`
+
+use std::path::PathBuf;
+
+use serde::{Deserialize, Serialize};
+
+use crate::settings::KeySource;
+
+/// Minimal config needed to connect to the database and decrypt secrets.
+///
+/// This is the only JSON file IronClaw reads from disk at startup.
+/// All other configuration lives in the `settings` table in PostgreSQL.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct BootstrapConfig {
+    /// Database connection URL (postgres://...).
+    #[serde(default)]
+    pub database_url: Option<String>,
+
+    /// Database connection pool size.
+    #[serde(default)]
+    pub database_pool_size: Option<usize>,
+
+    /// Source for the secrets master key.
+    #[serde(default)]
+    pub secrets_master_key_source: KeySource,
+
+    /// Whether onboarding wizard has been completed.
+    #[serde(default)]
+    pub onboard_completed: bool,
+}
+
+impl Default for BootstrapConfig {
+    fn default() -> Self {
+        Self {
+            database_url: None,
+            database_pool_size: None,
+            secrets_master_key_source: KeySource::None,
+            onboard_completed: false,
+        }
+    }
+}
+
+impl BootstrapConfig {
+    /// Default bootstrap file path: `~/.ironclaw/bootstrap.json`.
+    pub fn default_path() -> PathBuf {
+        dirs::home_dir()
+            .unwrap_or_else(|| PathBuf::from("."))
+            .join(".ironclaw")
+            .join("bootstrap.json")
+    }
+
+    /// Legacy settings.json path (for migration detection).
+    pub fn legacy_settings_path() -> PathBuf {
+        dirs::home_dir()
+            .unwrap_or_else(|| PathBuf::from("."))
+            .join(".ironclaw")
+            .join("settings.json")
+    }
+
+    /// Load from the default path, falling back to legacy settings.json,
+    /// then to defaults if neither exists.
+    pub fn load() -> Self {
+        let bootstrap_path = Self::default_path();
+        if bootstrap_path.exists() {
+            return Self::load_from(&bootstrap_path);
+        }
+
+        // Fall back to legacy settings.json (extract just the 4 bootstrap fields)
+        let legacy_path = Self::legacy_settings_path();
+        if legacy_path.exists() {
+            return Self::load_from_legacy(&legacy_path);
+        }
+
+        Self::default()
+    }
+
+    /// Load from a specific path.
+    pub fn load_from(path: &PathBuf) -> Self {
+        match std::fs::read_to_string(path) {
+            Ok(data) => serde_json::from_str(&data).unwrap_or_default(),
+            Err(_) => Self::default(),
+        }
+    }
+
+    /// Extract bootstrap fields from a legacy settings.json.
+    fn load_from_legacy(path: &PathBuf) -> Self {
+        match std::fs::read_to_string(path) {
+            Ok(data) => {
+                // The legacy Settings struct is a superset; serde will ignore extra fields.
+                serde_json::from_str(&data).unwrap_or_default()
+            }
+            Err(_) => Self::default(),
+        }
+    }
+
+    /// Save to the default path.
+    pub fn save(&self) -> std::io::Result<()> {
+        self.save_to(&Self::default_path())
+    }
+
+    /// Save to a specific path.
+    pub fn save_to(&self, path: &PathBuf) -> std::io::Result<()> {
+        if let Some(parent) = path.parent() {
+            std::fs::create_dir_all(parent)?;
+        }
+        let json = serde_json::to_string_pretty(self)
+            .map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidData, e.to_string()))?;
+        std::fs::write(path, json)
+    }
+}
+
+/// One-time migration from disk config files to the database settings table.
+///
+/// On first boot after upgrade, checks if:
+/// 1. `~/.ironclaw/settings.json` exists
+/// 2. The DB settings table is empty for this user
+///
+/// If both conditions hold, migrates settings, MCP servers, and session data
+/// to the database, writes `bootstrap.json`, and renames old files to `.migrated`.
+pub async fn migrate_disk_to_db(
+    store: &crate::history::Store,
+    user_id: &str,
+) -> Result<(), MigrationError> {
+    let legacy_settings_path = BootstrapConfig::legacy_settings_path();
+    if !legacy_settings_path.exists() {
+        tracing::debug!("No legacy settings.json found, skipping disk-to-DB migration");
+        return Ok(());
+    }
+
+    // Only migrate if DB is empty for this user
+    let has_settings = store.has_settings(user_id).await.map_err(|e| {
+        MigrationError::Database(format!("Failed to check existing settings: {}", e))
+    })?;
+    if has_settings {
+        tracing::debug!(
+            "DB already has settings for user '{}', skipping migration",
+            user_id
+        );
+        return Ok(());
+    }
+
+    tracing::info!("Migrating disk settings to database...");
+
+    // 1. Load and migrate settings.json
+    let settings = crate::settings::Settings::load_from(&legacy_settings_path);
+    let db_map = settings.to_db_map();
+    if !db_map.is_empty() {
+        store
+            .set_all_settings(user_id, &db_map)
+            .await
+            .map_err(|e| {
+                MigrationError::Database(format!("Failed to write settings to DB: {}", e))
+            })?;
+        tracing::info!("Migrated {} settings to database", db_map.len());
+    }
+
+    // 2. Write bootstrap.json with the 4 essential fields
+    let bootstrap = BootstrapConfig {
+        database_url: settings.database_url.clone(),
+        database_pool_size: settings.database_pool_size,
+        secrets_master_key_source: settings.secrets_master_key_source,
+        onboard_completed: settings.onboard_completed,
+    };
+    bootstrap
+        .save()
+        .map_err(|e| MigrationError::Io(format!("Failed to write bootstrap.json: {}", e)))?;
+    tracing::info!("Wrote bootstrap.json");
+
+    // 3. Migrate mcp-servers.json if it exists
+    let ironclaw_dir = dirs::home_dir()
+        .unwrap_or_else(|| PathBuf::from("."))
+        .join(".ironclaw");
+    let mcp_path = ironclaw_dir.join("mcp-servers.json");
+    if mcp_path.exists() {
+        match std::fs::read_to_string(&mcp_path) {
+            Ok(content) => match serde_json::from_str::<serde_json::Value>(&content) {
+                Ok(value) => {
+                    store
+                        .set_setting(user_id, "mcp_servers", &value)
+                        .await
+                        .map_err(|e| {
+                            MigrationError::Database(format!(
+                                "Failed to write MCP servers to DB: {}",
+                                e
+                            ))
+                        })?;
+                    tracing::info!("Migrated mcp-servers.json to database");
+
+                    rename_to_migrated(&mcp_path);
+                }
+                Err(e) => {
+                    tracing::warn!("Failed to parse mcp-servers.json: {}", e);
+                }
+            },
+            Err(e) => {
+                tracing::warn!("Failed to read mcp-servers.json: {}", e);
+            }
+        }
+    }
+
+    // 4. Migrate session.json if it exists
+    let session_path = ironclaw_dir.join("session.json");
+    if session_path.exists() {
+        match std::fs::read_to_string(&session_path) {
+            Ok(content) => match serde_json::from_str::<serde_json::Value>(&content) {
+                Ok(value) => {
+                    store
+                        .set_setting(user_id, "nearai.session", &value)
+                        .await
+                        .map_err(|e| {
+                            MigrationError::Database(format!(
+                                "Failed to write session to DB: {}",
+                                e
+                            ))
+                        })?;
+                    tracing::info!("Migrated session.json to database");
+
+                    rename_to_migrated(&session_path);
+                }
+                Err(e) => {
+                    tracing::warn!("Failed to parse session.json: {}", e);
+                }
+            },
+            Err(e) => {
+                tracing::warn!("Failed to read session.json: {}", e);
+            }
+        }
+    }
+
+    // 5. Rename settings.json to .migrated (don't delete, safety net)
+    rename_to_migrated(&legacy_settings_path);
+
+    tracing::info!("Disk-to-DB migration complete");
+    Ok(())
+}
+
+/// Rename a file to `<name>.migrated` as a safety net.
+fn rename_to_migrated(path: &PathBuf) {
+    let mut migrated = path.as_os_str().to_owned();
+    migrated.push(".migrated");
+    if let Err(e) = std::fs::rename(path, &migrated) {
+        tracing::warn!("Failed to rename {} to .migrated: {}", path.display(), e);
+    }
+}
+
+/// Errors that can occur during disk-to-DB migration.
+#[derive(Debug, thiserror::Error)]
+pub enum MigrationError {
+    #[error("Database error: {0}")]
+    Database(String),
+    #[error("IO error: {0}")]
+    Io(String),
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::tempdir;
+
+    #[test]
+    fn test_bootstrap_save_load() {
+        let dir = tempdir().unwrap();
+        let path = dir.path().join("bootstrap.json");
+
+        let config = BootstrapConfig {
+            database_url: Some("postgres://localhost/test".to_string()),
+            database_pool_size: Some(5),
+            secrets_master_key_source: KeySource::Keychain,
+            onboard_completed: true,
+        };
+
+        config.save_to(&path).unwrap();
+
+        let loaded = BootstrapConfig::load_from(&path);
+        assert_eq!(
+            loaded.database_url,
+            Some("postgres://localhost/test".to_string())
+        );
+        assert_eq!(loaded.database_pool_size, Some(5));
+        assert_eq!(loaded.secrets_master_key_source, KeySource::Keychain);
+        assert!(loaded.onboard_completed);
+    }
+
+    #[test]
+    fn test_bootstrap_from_legacy_settings() {
+        let dir = tempdir().unwrap();
+        let path = dir.path().join("settings.json");
+
+        // Write a legacy settings.json with many extra fields
+        let legacy = serde_json::json!({
+            "database_url": "postgres://localhost/ironclaw",
+            "database_pool_size": 10,
+            "secrets_master_key_source": "keychain",
+            "onboard_completed": true,
+            "selected_model": "claude-3-5-sonnet",
+            "agent": { "name": "testbot", "max_parallel_jobs": 3 },
+            "heartbeat": { "enabled": true }
+        });
+        std::fs::write(&path, serde_json::to_string_pretty(&legacy).unwrap()).unwrap();
+
+        let config = BootstrapConfig::load_from_legacy(&path);
+        assert_eq!(
+            config.database_url,
+            Some("postgres://localhost/ironclaw".to_string())
+        );
+        assert_eq!(config.database_pool_size, Some(10));
+        assert_eq!(config.secrets_master_key_source, KeySource::Keychain);
+        assert!(config.onboard_completed);
+    }
+
+    #[test]
+    fn test_bootstrap_defaults() {
+        let config = BootstrapConfig::default();
+        assert!(config.database_url.is_none());
+        assert!(config.database_pool_size.is_none());
+        assert_eq!(config.secrets_master_key_source, KeySource::None);
+        assert!(!config.onboard_completed);
+    }
+}

src/channels/channel.rs

@@ -114,6 +114,12 @@ pub enum StatusUpdate {
     StreamChunk(String),
     /// General status message.
     Status(String),
+    /// A sandbox job has started (shown as a clickable card in the UI).
+    JobStarted {
+        job_id: String,
+        title: String,
+        browse_url: String,
+    },
     /// Tool requires user approval before execution.
     ApprovalNeeded {
         request_id: String,
@@ -121,6 +127,19 @@ pub enum StatusUpdate {
         description: String,
         parameters: serde_json::Value,
     },
+    /// Extension needs user authentication (token or OAuth).
+    AuthRequired {
+        extension_name: String,
+        instructions: Option<String>,
+        auth_url: Option<String>,
+        setup_url: Option<String>,
+    },
+    /// Extension authentication completed.
+    AuthCompleted {
+        extension_name: String,
+        success: bool,
+        message: String,
+    },
 }
 
 /// Trait for message channels.

src/channels/repl.rs

@@ -42,12 +42,25 @@ const SLASH_COMMANDS: &[&str] = &[
     "/quit",
     "/exit",
     "/debug",
+    "/model",
     "/undo",
     "/redo",
     "/clear",
     "/compact",
     "/new",
     "/interrupt",
+    "/version",
+    "/tools",
+    "/ping",
+    "/job",
+    "/status",
+    "/cancel",
+    "/list",
+    "/heartbeat",
+    "/summarize",
+    "/suggest",
+    "/thread",
+    "/resume",
 ];
 
 /// Rustyline helper for slash-command tab completion.
@@ -295,10 +308,11 @@ impl Channel for ReplChannel {
                             continue;
                         }
 
-                        // Handle local REPL commands
+                        // Handle local REPL commands (only commands that need
+                        // immediate local handling stay here)
                         match line.to_lowercase().as_str() {
                             "/quit" | "/exit" => break,
-                            "/help" | "/?" => {
+                            "/help" => {
                                 print_help();
                                 continue;
                             }
@@ -413,6 +427,15 @@ impl Channel for ReplChannel {
                 print!("{chunk}");
                 let _ = io::stdout().flush();
             }
+            StatusUpdate::JobStarted {
+                job_id,
+                title,
+                browse_url,
+            } => {
+                eprintln!(
+                    "  \x1b[36m[job]\x1b[0m {title} \x1b[90m({job_id})\x1b[0m \x1b[4m{browse_url}\x1b[0m"
+                );
+            }
             StatusUpdate::Status(msg) => {
                 if debug || msg.contains("approval") || msg.contains("Approval") {
                     eprintln!("  \x1b[90m{msg}\x1b[0m");
@@ -472,6 +495,33 @@ impl Channel for ReplChannel {
                 eprintln!("  {bot_border}");
                 eprintln!();
             }
+            StatusUpdate::AuthRequired {
+                extension_name,
+                instructions,
+                setup_url,
+                ..
+            } => {
+                eprintln!();
+                eprintln!("\x1b[33m  Authentication required for {extension_name}\x1b[0m");
+                if let Some(ref instr) = instructions {
+                    eprintln!("  {instr}");
+                }
+                if let Some(ref url) = setup_url {
+                    eprintln!("  \x1b[4m{url}\x1b[0m");
+                }
+                eprintln!();
+            }
+            StatusUpdate::AuthCompleted {
+                extension_name,
+                success,
+                message,
+            } => {
+                if success {
+                    eprintln!("\x1b[32m  {extension_name}: {message}\x1b[0m");
+                } else {
+                    eprintln!("\x1b[31m  {extension_name}: {message}\x1b[0m");
+                }
+            }
         }
         Ok(())
     }

src/channels/wasm/bundled.rs

@@ -1,68 +1,125 @@
-//! Bundled WASM channels that can be installed locally.
+//! Known WASM channels that can be installed from build artifacts.
+//!
+//! Instead of embedding WASM binaries in the host binary via include_bytes!,
+//! channels are compiled separately and installed from their build output
+//! directories during onboarding.
+//!
+//! Channel source layout:
+//!   channels-src/<name>/
+//!     target/wasm32-wasip2/release/<name>_channel.wasm
+//!     <name>.capabilities.json
 
-use std::path::Path;
+use std::path::{Path, PathBuf};
 
 use tokio::fs;
 
-#[derive(Clone, Copy)]
-struct BundledChannel {
-    name: &'static str,
-    wasm: &'static [u8],
-    capabilities: &'static [u8],
+/// Compile-time project root, used to locate channels-src/ in dev builds.
+const CARGO_MANIFEST_DIR: &str = env!("CARGO_MANIFEST_DIR");
+
+/// Known channel names and their crate names (for locating build artifacts).
+const KNOWN_CHANNELS: &[(&str, &str)] = &[
+    ("telegram", "telegram_channel"),
+    ("slack", "slack_channel"),
+    ("whatsapp", "whatsapp_channel"),
+];
+
+/// Names of known channels that can be installed.
+pub fn bundled_channel_names() -> Vec<&'static str> {
+    KNOWN_CHANNELS.iter().map(|(name, _)| *name).collect()
 }
 
-/// Names of bundled channels shipped with IronClaw.
-pub fn bundled_channel_names() -> &'static [&'static str] {
-    &["telegram"]
+/// Resolve the channels source directory.
+///
+/// Checks (in order):
+/// 1. `IRONCLAW_CHANNELS_SRC` env var
+/// 2. `<CARGO_MANIFEST_DIR>/channels-src/` (dev builds)
+fn channels_src_dir() -> PathBuf {
+    if let Ok(dir) = std::env::var("IRONCLAW_CHANNELS_SRC") {
+        return PathBuf::from(dir);
+    }
+    PathBuf::from(CARGO_MANIFEST_DIR).join("channels-src")
 }
 
-/// Install a bundled channel into a channels directory.
+/// Locate the build artifacts for a channel.
+///
+/// Returns (wasm_path, capabilities_path) or an error if files are missing.
+fn locate_channel_artifacts(name: &str) -> Result<(PathBuf, PathBuf), String> {
+    let (_, crate_name) = KNOWN_CHANNELS
+        .iter()
+        .find(|(n, _)| *n == name)
+        .ok_or_else(|| format!("Unknown channel '{}'", name))?;
+
+    let src_dir = channels_src_dir();
+    let channel_dir = src_dir.join(name);
+
+    let wasm_path = channel_dir
+        .join("target/wasm32-wasip2/release")
+        .join(format!("{}.wasm", crate_name));
+
+    let caps_path = channel_dir.join(format!("{}.capabilities.json", name));
+
+    if !wasm_path.exists() {
+        return Err(format!(
+            "Channel '{}' WASM not found at {}. Build it first:\n  \
+             cd {} && cargo build --target wasm32-wasip2 --release",
+            name,
+            wasm_path.display(),
+            channel_dir.display()
+        ));
+    }
+
+    if !caps_path.exists() {
+        return Err(format!(
+            "Channel '{}' capabilities not found at {}",
+            name,
+            caps_path.display()
+        ));
+    }
+
+    Ok((wasm_path, caps_path))
+}
+
+/// Install a channel from build artifacts into the channels directory.
 pub async fn install_bundled_channel(
     name: &str,
     target_dir: &Path,
     force: bool,
 ) -> Result<(), String> {
-    let channel = bundled_channel(name)
-        .ok_or_else(|| format!("Unknown bundled channel '{}'", name.to_lowercase()))?;
+    let (wasm_src, caps_src) = locate_channel_artifacts(name)?;
 
     fs::create_dir_all(target_dir)
         .await
         .map_err(|e| format!("Failed to create channels directory: {}", e))?;
 
-    let wasm_path = target_dir.join(format!("{}.wasm", channel.name));
-    let caps_path = target_dir.join(format!("{}.capabilities.json", channel.name));
+    let wasm_dst = target_dir.join(format!("{}.wasm", name));
+    let caps_dst = target_dir.join(format!("{}.capabilities.json", name));
 
-    let has_existing = wasm_path.exists() || caps_path.exists();
+    let has_existing = wasm_dst.exists() || caps_dst.exists();
     if has_existing && !force {
         return Err(format!(
             "Channel '{}' already exists at {}",
-            channel.name,
+            name,
             target_dir.display()
         ));
     }
 
-    fs::write(&wasm_path, channel.wasm)
+    fs::copy(&wasm_src, &wasm_dst)
         .await
-        .map_err(|e| format!("Failed to write {}: {}", wasm_path.display(), e))?;
-    fs::write(&caps_path, channel.capabilities)
+        .map_err(|e| format!("Failed to copy {}: {}", wasm_src.display(), e))?;
+    fs::copy(&caps_src, &caps_dst)
         .await
-        .map_err(|e| format!("Failed to write {}: {}", caps_path.display(), e))?;
+        .map_err(|e| format!("Failed to copy {}: {}", caps_src.display(), e))?;
 
     Ok(())
 }
 
-fn bundled_channel(name: &str) -> Option<BundledChannel> {
-    if name.eq_ignore_ascii_case("telegram") {
-        Some(BundledChannel {
-            name: "telegram",
-            wasm: include_bytes!("../../../channels-src/telegram/telegram.wasm"),
-            capabilities: include_bytes!(
-                "../../../channels-src/telegram/telegram.capabilities.json"
-            ),
-        })
-    } else {
-        None
-    }
+/// Check which known channels have build artifacts available.
+pub fn available_channel_names() -> Vec<&'static str> {
+    KNOWN_CHANNELS
+        .iter()
+        .filter(|(name, _)| locate_channel_artifacts(name).is_ok())
+        .map(|(name, _)| *name)
+        .collect()
 }
 
 #[cfg(test)]
@@ -73,31 +130,35 @@ mod tests {
     use super::*;
 
     #[test]
-    fn test_bundled_channel_names_contains_telegram() {
-        assert!(bundled_channel_names().contains(&"telegram"));
+    fn test_known_channels_includes_all_three() {
+        let names = bundled_channel_names();
+        assert!(names.contains(&"telegram"));
+        assert!(names.contains(&"slack"));
+        assert!(names.contains(&"whatsapp"));
+    }
+
+    #[test]
+    fn test_channels_src_dir_default() {
+        let dir = channels_src_dir();
+        assert!(dir.ends_with("channels-src"));
+    }
+
+    #[test]
+    fn test_locate_unknown_channel_errors() {
+        assert!(locate_channel_artifacts("nonexistent").is_err());
     }
 
     #[tokio::test]
-    async fn test_install_bundled_channel_writes_files() {
-        let dir = tempdir().unwrap();
-
-        install_bundled_channel("telegram", dir.path(), false)
-            .await
-            .unwrap();
-
-        assert!(dir.path().join("telegram.wasm").exists());
-        assert!(dir.path().join("telegram.capabilities.json").exists());
-    }
-
-    #[tokio::test]
-    async fn test_install_bundled_channel_refuses_overwrite_without_force() {
+    async fn test_install_refuses_overwrite_without_force() {
         let dir = tempdir().unwrap();
         let wasm_path = dir.path().join("telegram.wasm");
         fs::write(&wasm_path, b"custom").await.unwrap();
 
         let result = install_bundled_channel("telegram", dir.path(), false).await;
+        // Either fails because artifacts missing OR because file exists
         assert!(result.is_err());
 
+        // Original file should be untouched
         let existing = fs::read(&wasm_path).await.unwrap();
         assert_eq!(existing, b"custom");
     }

src/channels/wasm/loader.rs

@@ -16,16 +16,21 @@ use crate::channels::wasm::error::WasmChannelError;
 use crate::channels::wasm::runtime::WasmChannelRuntime;
 use crate::channels::wasm::schema::ChannelCapabilitiesFile;
 use crate::channels::wasm::wrapper::WasmChannel;
+use crate::pairing::PairingStore;
 
 /// Loads WASM channels from the filesystem.
 pub struct WasmChannelLoader {
     runtime: Arc<WasmChannelRuntime>,
+    pairing_store: Arc<PairingStore>,
 }
 
 impl WasmChannelLoader {
-    /// Create a new loader with the given runtime.
-    pub fn new(runtime: Arc<WasmChannelRuntime>) -> Self {
-        Self { runtime }
+    /// Create a new loader with the given runtime and pairing store.
+    pub fn new(runtime: Arc<WasmChannelRuntime>, pairing_store: Arc<PairingStore>) -> Self {
+        Self {
+            runtime,
+            pairing_store,
+        }
     }
 
     /// Load a single WASM channel from a file pair.
@@ -114,7 +119,13 @@ impl WasmChannelLoader {
             .await?;
 
         // Create the channel
-        let channel = WasmChannel::new(self.runtime.clone(), prepared, capabilities, config_json);
+        let channel = WasmChannel::new(
+            self.runtime.clone(),
+            prepared,
+            capabilities,
+            config_json,
+            self.pairing_store.clone(),
+        );
 
         tracing::info!(
             name = name,
@@ -352,6 +363,7 @@ mod tests {
 
     use crate::channels::wasm::loader::{WasmChannelLoader, discover_channels};
     use crate::channels::wasm::runtime::{WasmChannelRuntime, WasmChannelRuntimeConfig};
+    use crate::pairing::PairingStore;
     use std::sync::Arc;
 
     #[tokio::test]
@@ -408,7 +420,7 @@ mod tests {
     async fn test_loader_invalid_name() {
         let config = WasmChannelRuntimeConfig::for_testing();
         let runtime = Arc::new(WasmChannelRuntime::new(config).unwrap());
-        let loader = WasmChannelLoader::new(runtime);
+        let loader = WasmChannelLoader::new(runtime, Arc::new(PairingStore::new()));
 
         let dir = TempDir::new().unwrap();
         let wasm_path = dir.path().join("test.wasm");

src/channels/wasm/mod.rs

@@ -89,7 +89,7 @@ mod schema;
 mod wrapper;
 
 // Core types
-pub use bundled::{bundled_channel_names, install_bundled_channel};
+pub use bundled::{available_channel_names, bundled_channel_names, install_bundled_channel};
 pub use capabilities::{ChannelCapabilities, EmitRateLimitConfig, HttpEndpointConfig, PollConfig};
 pub use error::WasmChannelError;
 pub use host::{ChannelEmitRateLimiter, ChannelHostState, EmittedMessage};

src/channels/wasm/router.rs

@@ -478,6 +478,7 @@ mod tests {
         PreparedChannelModule, WasmChannelRuntime, WasmChannelRuntimeConfig,
     };
     use crate::channels::wasm::wrapper::WasmChannel;
+    use crate::pairing::PairingStore;
     use crate::tools::wasm::ResourceLimits;
 
     fn create_test_channel(name: &str) -> Arc<WasmChannel> {
@@ -499,6 +500,7 @@ mod tests {
             prepared,
             capabilities,
             "{}".to_string(),
+            Arc::new(PairingStore::new()),
         ))
     }
 

src/channels/wasm/wrapper.rs

@@ -48,6 +48,7 @@ use crate::channels::wasm::runtime::{PreparedChannelModule, WasmChannelRuntime};
 use crate::channels::wasm::schema::ChannelConfig;
 use crate::channels::{Channel, IncomingMessage, MessageStream, OutgoingResponse, StatusUpdate};
 use crate::error::ChannelError;
+use crate::pairing::PairingStore;
 use crate::safety::LeakDetector;
 use crate::tools::wasm::LogLevel;
 use crate::tools::wasm::WasmResourceLimiter;
@@ -73,6 +74,8 @@ struct ChannelStoreData {
     /// Injected credentials for URL substitution (e.g., bot tokens).
     /// Keys are placeholder names like "TELEGRAM_BOT_TOKEN".
     credentials: HashMap<String, String>,
+    /// Pairing store for DM pairing (guest access control).
+    pairing_store: Arc<PairingStore>,
 }
 
 impl ChannelStoreData {
@@ -81,6 +84,7 @@ impl ChannelStoreData {
         channel_name: &str,
         capabilities: ChannelCapabilities,
         credentials: HashMap<String, String>,
+        pairing_store: Arc<PairingStore>,
     ) -> Self {
         // Create a minimal WASI context (no filesystem, no env vars for security)
         let wasi = WasiCtxBuilder::new().build();
@@ -91,6 +95,7 @@ impl ChannelStoreData {
             wasi,
             table: ResourceTable::new(),
             credentials,
+            pairing_store,
         }
     }
 
@@ -141,6 +146,22 @@ impl ChannelStoreData {
 
         result
     }
+
+    /// Replace injected credential values with `[REDACTED]` in text.
+    ///
+    /// Prevents credentials from leaking through error messages, logs, or
+    /// return values to WASM. reqwest::Error includes the full URL in its
+    /// Display output, so any error from an injected-URL request will
+    /// contain the raw credential unless we scrub it.
+    fn redact_credentials(&self, text: &str) -> String {
+        let mut result = text.to_string();
+        for (name, value) in &self.credentials {
+            if !value.is_empty() {
+                result = result.replace(value, &format!("[REDACTED:{}]", name));
+            }
+        }
+        result
+    }
 }
 
 // Implement WasiView to provide WASI context and resource table
@@ -187,6 +208,7 @@ impl near::agent::channel_host::Host for ChannelStoreData {
         url: String,
         headers_json: String,
         body: Option<Vec<u8>>,
+        timeout_ms: Option<u32>,
     ) -> Result<near::agent::channel_host::HttpResponse, String> {
         tracing::info!(
             method = %method,
@@ -276,12 +298,21 @@ impl near::agent::channel_host::Host for ChannelStoreData {
                 request = request.body(body_bytes);
             }
 
-            // Send request with timeout
-            let response = request
-                .timeout(std::time::Duration::from_secs(30))
-                .send()
-                .await
-                .map_err(|e| format!("HTTP request failed: {}", e))?;
+            // Send request with caller-specified timeout (default 30s).
+            // Cap at callback_timeout to prevent outliving the host wrapper.
+            let timeout = std::time::Duration::from_millis(timeout_ms.unwrap_or(30_000) as u64);
+            let response = request.timeout(timeout).send().await.map_err(|e| {
+                // Walk the full error chain so we get the actual root cause
+                // (DNS, TLS, connection refused, etc.) instead of just
+                // "error sending request for url (...)".
+                let mut chain = format!("HTTP request failed: {}", e);
+                let mut source = std::error::Error::source(&e);
+                while let Some(cause) = source {
+                    chain.push_str(&format!(" -> {}", cause));
+                    source = cause.source();
+                }
+                chain
+            })?;
 
             let status = response.status().as_u16();
             let response_headers: std::collections::HashMap<String, String> = response
@@ -330,6 +361,11 @@ impl near::agent::channel_host::Host for ChannelStoreData {
             })
         });
 
+        // Scrub credential values from error messages before logging or returning
+        // to WASM. reqwest::Error includes the full URL (with injected credentials)
+        // in its Display output.
+        let result = result.map_err(|e| self.redact_credentials(&e));
+
         match &result {
             Ok(resp) => {
                 tracing::info!(status = resp.status, "http_request completed successfully");
@@ -372,6 +408,43 @@ impl near::agent::channel_host::Host for ChannelStoreData {
             }
         }
     }
+
+    fn pairing_upsert_request(
+        &mut self,
+        channel: String,
+        id: String,
+        meta_json: String,
+    ) -> Result<near::agent::channel_host::PairingUpsertResult, String> {
+        let meta = if meta_json.is_empty() {
+            None
+        } else {
+            serde_json::from_str(&meta_json).ok()
+        };
+        match self.pairing_store.upsert_request(&channel, &id, meta) {
+            Ok(r) => Ok(near::agent::channel_host::PairingUpsertResult {
+                code: r.code,
+                created: r.created,
+            }),
+            Err(e) => Err(e.to_string()),
+        }
+    }
+
+    fn pairing_is_allowed(
+        &mut self,
+        channel: String,
+        id: String,
+        username: Option<String>,
+    ) -> Result<bool, String> {
+        self.pairing_store
+            .is_sender_allowed(&channel, &id, username.as_deref())
+            .map_err(|e| e.to_string())
+    }
+
+    fn pairing_read_allow_from(&mut self, channel: String) -> Result<Vec<String>, String> {
+        self.pairing_store
+            .read_allow_from(&channel)
+            .map_err(|e| e.to_string())
+    }
 }
 
 /// A WASM-based channel implementing the Channel trait.
@@ -424,6 +497,9 @@ pub struct WasmChannel {
     /// Background task that repeats typing indicators every 4 seconds.
     /// Telegram's "typing..." indicator expires after ~5s, so we refresh it.
     typing_task: RwLock<Option<tokio::task::JoinHandle<()>>>,
+
+    /// Pairing store for DM pairing (guest access control).
+    pairing_store: Arc<PairingStore>,
 }
 
 impl WasmChannel {
@@ -433,6 +509,7 @@ impl WasmChannel {
         prepared: Arc<PreparedChannelModule>,
         capabilities: ChannelCapabilities,
         config_json: String,
+        pairing_store: Arc<PairingStore>,
     ) -> Self {
         let name = prepared.name.clone();
         let rate_limiter = ChannelEmitRateLimiter::new(capabilities.emit_rate_limit.clone());
@@ -452,6 +529,7 @@ impl WasmChannel {
             endpoints: RwLock::new(Vec::new()),
             credentials: Arc::new(RwLock::new(HashMap::new())),
             typing_task: RwLock::new(None),
+            pairing_store,
         }
     }
 
@@ -533,6 +611,7 @@ impl WasmChannel {
         prepared: &PreparedChannelModule,
         capabilities: &ChannelCapabilities,
         credentials: HashMap<String, String>,
+        pairing_store: Arc<PairingStore>,
     ) -> Result<Store<ChannelStoreData>, WasmChannelError> {
         let engine = runtime.engine();
         let limits = &prepared.limits;
@@ -543,6 +622,7 @@ impl WasmChannel {
             &prepared.name,
             capabilities.clone(),
             credentials,
+            pairing_store,
         );
         let mut store = Store::new(engine, store_data);
 
@@ -643,12 +723,18 @@ impl WasmChannel {
         let timeout = self.runtime.config().callback_timeout;
         let channel_name = self.name.clone();
         let credentials = self.get_credentials().await;
+        let pairing_store = self.pairing_store.clone();
 
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials,
+                    pairing_store,
+                )?;
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
 
                 // Call on_start using the generated typed interface
@@ -753,6 +839,7 @@ impl WasmChannel {
         let capabilities = self.capabilities.clone();
         let timeout = self.runtime.config().callback_timeout;
         let credentials = self.get_credentials().await;
+        let pairing_store = self.pairing_store.clone();
 
         // Prepare request data
         let method = method.to_string();
@@ -766,8 +853,13 @@ impl WasmChannel {
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials,
+                    pairing_store,
+                )?;
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
 
                 // Build the WIT request type
@@ -840,12 +932,18 @@ impl WasmChannel {
         let timeout = self.runtime.config().callback_timeout;
         let channel_name = self.name.clone();
         let credentials = self.get_credentials().await;
+        let pairing_store = self.pairing_store.clone();
 
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials,
+                    pairing_store,
+                )?;
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
 
                 // Call on_poll using the generated typed interface
@@ -929,6 +1027,7 @@ impl WasmChannel {
         let timeout = self.runtime.config().callback_timeout;
         let channel_name = self.name.clone();
         let credentials = self.get_credentials().await;
+        let pairing_store = self.pairing_store.clone();
 
         // Prepare response data
         let message_id_str = message_id.to_string();
@@ -942,8 +1041,13 @@ impl WasmChannel {
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
                 tracing::info!("Creating WASM store for on_respond");
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials,
+                    pairing_store,
+                )?;
 
                 tracing::info!("Instantiating WASM component for on_respond");
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
@@ -1036,13 +1140,19 @@ impl WasmChannel {
         let timeout = self.runtime.config().callback_timeout;
         let channel_name = self.name.clone();
         let credentials = self.get_credentials().await;
+        let pairing_store = self.pairing_store.clone();
 
         let wit_update = status_to_wit(status, metadata);
 
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials,
+                    pairing_store,
+                )?;
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
 
                 let channel_iface = instance.near_agent_channel();
@@ -1080,12 +1190,14 @@ impl WasmChannel {
     ///
     /// Static method for use by the background typing repeat task (which
     /// doesn't have access to `&self`).
+    #[allow(clippy::too_many_arguments)]
     async fn execute_status(
         channel_name: &str,
         runtime: &Arc<WasmChannelRuntime>,
         prepared: &Arc<PreparedChannelModule>,
         capabilities: &ChannelCapabilities,
         credentials: &RwLock<HashMap<String, String>>,
+        pairing_store: Arc<PairingStore>,
         timeout: Duration,
         wit_update: wit_channel::StatusUpdate,
     ) -> Result<(), WasmChannelError> {
@@ -1101,8 +1213,13 @@ impl WasmChannel {
 
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials_snapshot)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials_snapshot,
+                    pairing_store,
+                )?;
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
 
                 let channel_iface = instance.near_agent_channel();
@@ -1170,6 +1287,7 @@ impl WasmChannel {
                 let prepared = Arc::clone(&self.prepared);
                 let capabilities = self.capabilities.clone();
                 let credentials = self.credentials.clone();
+                let pairing_store = self.pairing_store.clone();
                 let callback_timeout = self.runtime.config().callback_timeout;
                 let wit_update = status_to_wit(&status, metadata);
 
@@ -1189,6 +1307,7 @@ impl WasmChannel {
                             &prepared,
                             &capabilities,
                             &credentials,
+                            pairing_store.clone(),
                             callback_timeout,
                             wit_update_clone,
                         )
@@ -1319,6 +1438,7 @@ impl WasmChannel {
         let message_tx = self.message_tx.clone();
         let rate_limiter = self.rate_limiter.clone();
         let credentials = self.credentials.clone();
+        let pairing_store = self.pairing_store.clone();
         let callback_timeout = self.runtime.config().callback_timeout;
 
         tokio::spawn(async move {
@@ -1340,6 +1460,7 @@ impl WasmChannel {
                             &prepared,
                             &capabilities,
                             &credentials,
+                            pairing_store.clone(),
                             callback_timeout,
                         ).await;
 
@@ -1391,6 +1512,7 @@ impl WasmChannel {
         prepared: &Arc<PreparedChannelModule>,
         capabilities: &ChannelCapabilities,
         credentials: &RwLock<HashMap<String, String>>,
+        pairing_store: Arc<PairingStore>,
         timeout: Duration,
     ) -> Result<Vec<EmittedMessage>, WasmChannelError> {
         // Skip if no WASM bytes (testing mode)
@@ -1411,8 +1533,13 @@ impl WasmChannel {
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
             tokio::task::spawn_blocking(move || {
-                let mut store =
-                    Self::create_store(&runtime, &prepared, &capabilities, credentials_snapshot)?;
+                let mut store = Self::create_store(
+                    &runtime,
+                    &prepared,
+                    &capabilities,
+                    credentials_snapshot,
+                    pairing_store,
+                )?;
                 let instance = Self::instantiate_component(&runtime, &prepared, &mut store)?;
 
                 // Call on_poll using the generated typed interface
@@ -1858,6 +1985,29 @@ fn status_to_wit(status: &StatusUpdate, metadata: &serde_json::Value) -> wit_cha
             message: format!("Approval needed: {} - {}", tool_name, description),
             metadata_json,
         },
+        StatusUpdate::JobStarted { job_id, title, .. } => wit_channel::StatusUpdate {
+            status: wit_channel::StatusType::Thinking,
+            message: format!("Job started: {} ({})", title, job_id),
+            metadata_json,
+        },
+        StatusUpdate::AuthRequired { extension_name, .. } => wit_channel::StatusUpdate {
+            status: wit_channel::StatusType::Thinking,
+            message: format!("Auth required: {}", extension_name),
+            metadata_json,
+        },
+        StatusUpdate::AuthCompleted {
+            extension_name,
+            success,
+            ..
+        } => wit_channel::StatusUpdate {
+            status: wit_channel::StatusType::Thinking,
+            message: format!(
+                "Auth {}: {}",
+                if *success { "completed" } else { "failed" },
+                extension_name
+            ),
+            metadata_json,
+        },
     }
 }
 
@@ -1929,6 +2079,7 @@ mod tests {
         PreparedChannelModule, WasmChannelRuntime, WasmChannelRuntimeConfig,
     };
     use crate::channels::wasm::wrapper::{HttpResponse, WasmChannel};
+    use crate::pairing::PairingStore;
     use crate::tools::wasm::ResourceLimits;
 
     fn create_test_channel() -> WasmChannel {
@@ -1944,7 +2095,13 @@ mod tests {
 
         let capabilities = ChannelCapabilities::for_channel("test").with_path("/webhook/test");
 
-        WasmChannel::new(runtime, prepared, capabilities, "{}".to_string())
+        WasmChannel::new(
+            runtime,
+            prepared,
+            capabilities,
+            "{}".to_string(),
+            Arc::new(PairingStore::new()),
+        )
     }
 
     #[test]
@@ -2019,6 +2176,7 @@ mod tests {
             &prepared,
             &capabilities,
             &credentials,
+            Arc::new(PairingStore::new()),
             timeout,
         )
         .await;
@@ -2112,7 +2270,13 @@ mod tests {
             .with_path("/webhook/poll")
             .with_polling(1000);
 
-        let channel = WasmChannel::new(runtime, prepared, capabilities, "{}".to_string());
+        let channel = WasmChannel::new(
+            runtime,
+            prepared,
+            capabilities,
+            "{}".to_string(),
+            Arc::new(PairingStore::new()),
+        );
 
         // Start the channel
         let _stream = channel.start().await.expect("Channel should start");
@@ -2350,4 +2514,89 @@ mod tests {
         assert_eq!(cloned.message, "hello");
         assert_eq!(cloned.metadata_json, "{\"a\":1}");
     }
+
+    #[test]
+    fn test_redact_credentials_replaces_values() {
+        use super::ChannelStoreData;
+
+        let mut creds = std::collections::HashMap::new();
+        creds.insert(
+            "TELEGRAM_BOT_TOKEN".to_string(),
+            "8218490433:AAEZeUxwqZ5OO3mOCXv7fKvpdhDgsmBBNis".to_string(),
+        );
+        creds.insert("OTHER_SECRET".to_string(), "s3cret".to_string());
+
+        let store = ChannelStoreData::new(
+            1024 * 1024,
+            "test",
+            ChannelCapabilities::default(),
+            creds,
+            Arc::new(PairingStore::new()),
+        );
+
+        let error = "HTTP request failed: error sending request for url \
+            (https://api.telegram.org/bot8218490433:AAEZeUxwqZ5OO3mOCXv7fKvpdhDgsmBBNis/getUpdates)";
+
+        let redacted = store.redact_credentials(error);
+
+        assert!(
+            !redacted.contains("8218490433:AAEZeUxwqZ5OO3mOCXv7fKvpdhDgsmBBNis"),
+            "credential value should be redacted"
+        );
+        assert!(
+            redacted.contains("[REDACTED:TELEGRAM_BOT_TOKEN]"),
+            "redacted text should contain placeholder name"
+        );
+        assert!(
+            !redacted.contains("s3cret"),
+            "other credentials should also be redacted"
+        );
+    }
+
+    #[test]
+    fn test_redact_credentials_no_op_without_credentials() {
+        use super::ChannelStoreData;
+
+        let store = ChannelStoreData::new(
+            1024 * 1024,
+            "test",
+            ChannelCapabilities::default(),
+            std::collections::HashMap::new(),
+            Arc::new(PairingStore::new()),
+        );
+
+        let input = "some error message";
+        assert_eq!(store.redact_credentials(input), input);
+    }
+
+    #[test]
+    fn test_redact_credentials_skips_empty_values() {
+        use super::ChannelStoreData;
+
+        let mut creds = std::collections::HashMap::new();
+        creds.insert("EMPTY_TOKEN".to_string(), String::new());
+
+        let store = ChannelStoreData::new(
+            1024 * 1024,
+            "test",
+            ChannelCapabilities::default(),
+            creds,
+            Arc::new(PairingStore::new()),
+        );
+
+        let input = "should not match anything";
+        assert_eq!(store.redact_credentials(input), input);
+    }
+
+    /// Verify that the block_on-inside-spawn_blocking pattern used by the WASM
+    /// channel HTTP host function doesn't deadlock or panic.
+    #[tokio::test]
+    async fn test_block_on_inside_spawn_blocking_does_not_deadlock() {
+        let result = tokio::task::spawn_blocking(|| {
+            tokio::runtime::Handle::current().block_on(async { 42 })
+        })
+        .await
+        .expect("spawn_blocking panicked");
+        assert_eq!(result, 42);
+    }
 }

src/channels/web/mod.rs

@@ -10,7 +10,7 @@
 //!         ◄── GET  /api/chat/events ── SSE stream
 //!         ─── GET  /api/chat/ws ─────► WebSocket (bidirectional)
 //!         ─── GET  /api/memory/* ────► Workspace
-//!         ─── GET  /api/jobs/* ──────► ContextManager
+//!         ─── GET  /api/jobs/* ──────► Database
 //!         ◄── GET  / ───────────────── Static HTML/CSS/JS
 //! ```
 
@@ -31,9 +31,10 @@ use tokio_stream::wrappers::ReceiverStream;
 use crate::agent::SessionManager;
 use crate::channels::{Channel, IncomingMessage, MessageStream, OutgoingResponse, StatusUpdate};
 use crate::config::GatewayConfig;
-use crate::context::ContextManager;
 use crate::error::ChannelError;
 use crate::extensions::ExtensionManager;
+use crate::history::Store;
+use crate::orchestrator::job_manager::ContainerJobManager;
 use crate::tools::ToolRegistry;
 use crate::workspace::Workspace;
 
@@ -70,11 +71,13 @@ impl GatewayChannel {
             msg_tx: tokio::sync::RwLock::new(None),
             sse: SseManager::new(),
             workspace: None,
-            context_manager: None,
             session_manager: None,
             log_broadcaster: None,
             extension_manager: None,
             tool_registry: None,
+            store: None,
+            job_manager: None,
+            prompt_queue: None,
             user_id: config.user_id.clone(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: Some(Arc::new(ws::WsConnectionTracker::new())),
@@ -93,11 +96,13 @@ impl GatewayChannel {
             msg_tx: tokio::sync::RwLock::new(None),
             sse: SseManager::new(),
             workspace: self.state.workspace.clone(),
-            context_manager: self.state.context_manager.clone(),
             session_manager: self.state.session_manager.clone(),
             log_broadcaster: self.state.log_broadcaster.clone(),
             extension_manager: self.state.extension_manager.clone(),
             tool_registry: self.state.tool_registry.clone(),
+            store: self.state.store.clone(),
+            job_manager: self.state.job_manager.clone(),
+            prompt_queue: self.state.prompt_queue.clone(),
             user_id: self.state.user_id.clone(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: self.state.ws_tracker.clone(),
@@ -112,12 +117,6 @@ impl GatewayChannel {
         self
     }
 
-    /// Inject the context manager for the jobs API.
-    pub fn with_context_manager(mut self, cm: Arc<ContextManager>) -> Self {
-        self.rebuild_state(|s| s.context_manager = Some(cm));
-        self
-    }
-
     /// Inject the session manager for thread/session info.
     pub fn with_session_manager(mut self, sm: Arc<SessionManager>) -> Self {
         self.rebuild_state(|s| s.session_manager = Some(sm));
@@ -142,6 +141,34 @@ impl GatewayChannel {
         self
     }
 
+    /// Inject the database store for sandbox job persistence.
+    pub fn with_store(mut self, store: Arc<Store>) -> Self {
+        self.rebuild_state(|s| s.store = Some(store));
+        self
+    }
+
+    /// Inject the container job manager for sandbox operations.
+    pub fn with_job_manager(mut self, jm: Arc<ContainerJobManager>) -> Self {
+        self.rebuild_state(|s| s.job_manager = Some(jm));
+        self
+    }
+
+    /// Inject the prompt queue for Claude Code follow-up prompts.
+    pub fn with_prompt_queue(
+        mut self,
+        pq: Arc<
+            tokio::sync::Mutex<
+                std::collections::HashMap<
+                    uuid::Uuid,
+                    std::collections::VecDeque<crate::orchestrator::api::PendingPrompt>,
+                >,
+            >,
+        >,
+    ) -> Self {
+        self.rebuild_state(|s| s.prompt_queue = Some(pq));
+        self
+    }
+
     /// Get the auth token (for printing to console on startup).
     pub fn auth_token(&self) -> &str {
         &self.auth_token
@@ -173,11 +200,7 @@ impl Channel for GatewayChannel {
                 ),
             })?;
 
-        let bound_addr =
-            server::start_server(addr, self.state.clone(), self.auth_token.clone()).await?;
-
-        tracing::info!("Web gateway listening on http://{}", bound_addr);
-        tracing::info!("Auth token: {}", self.auth_token);
+        server::start_server(addr, self.state.clone(), self.auth_token.clone()).await?;
 
         Ok(Box::pin(ReceiverStream::new(rx)))
     }
@@ -200,17 +223,48 @@ impl Channel for GatewayChannel {
     async fn send_status(
         &self,
         status: StatusUpdate,
-        _metadata: &serde_json::Value,
+        metadata: &serde_json::Value,
     ) -> Result<(), ChannelError> {
+        let thread_id = metadata
+            .get("thread_id")
+            .and_then(|v| v.as_str())
+            .map(String::from);
         let event = match status {
-            StatusUpdate::Thinking(msg) => SseEvent::Thinking { message: msg },
-            StatusUpdate::ToolStarted { name } => SseEvent::ToolStarted { name },
-            StatusUpdate::ToolCompleted { name, success } => {
-                SseEvent::ToolCompleted { name, success }
-            }
-            StatusUpdate::ToolResult { name, preview } => SseEvent::ToolResult { name, preview },
-            StatusUpdate::StreamChunk(content) => SseEvent::StreamChunk { content },
-            StatusUpdate::Status(msg) => SseEvent::Status { message: msg },
+            StatusUpdate::Thinking(msg) => SseEvent::Thinking {
+                message: msg,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::ToolStarted { name } => SseEvent::ToolStarted {
+                name,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::ToolCompleted { name, success } => SseEvent::ToolCompleted {
+                name,
+                success,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::ToolResult { name, preview } => SseEvent::ToolResult {
+                name,
+                preview,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::StreamChunk(content) => SseEvent::StreamChunk {
+                content,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::Status(msg) => SseEvent::Status {
+                message: msg,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::JobStarted {
+                job_id,
+                title,
+                browse_url,
+            } => SseEvent::JobStarted {
+                job_id,
+                title,
+                browse_url,
+            },
             StatusUpdate::ApprovalNeeded {
                 request_id,
                 tool_name,
@@ -223,6 +277,26 @@ impl Channel for GatewayChannel {
                 parameters: serde_json::to_string_pretty(&parameters)
                     .unwrap_or_else(|_| parameters.to_string()),
             },
+            StatusUpdate::AuthRequired {
+                extension_name,
+                instructions,
+                auth_url,
+                setup_url,
+            } => SseEvent::AuthRequired {
+                extension_name,
+                instructions,
+                auth_url,
+                setup_url,
+            },
+            StatusUpdate::AuthCompleted {
+                extension_name,
+                success,
+                message,
+            } => SseEvent::AuthCompleted {
+                extension_name,
+                success,
+                message,
+            },
         };
 
         self.state.sse.broadcast(event);

src/channels/web/sse.rs

@@ -79,7 +79,15 @@ impl SseManager {
                     SseEvent::StreamChunk { .. } => "stream_chunk",
                     SseEvent::Status { .. } => "status",
                     SseEvent::ApprovalNeeded { .. } => "approval_needed",
+                    SseEvent::AuthRequired { .. } => "auth_required",
+                    SseEvent::AuthCompleted { .. } => "auth_completed",
                     SseEvent::Error { .. } => "error",
+                    SseEvent::JobStarted { .. } => "job_started",
+                    SseEvent::JobMessage { .. } => "job_message",
+                    SseEvent::JobToolUse { .. } => "job_tool_use",
+                    SseEvent::JobToolResult { .. } => "job_tool_result",
+                    SseEvent::JobStatus { .. } => "job_status",
+                    SseEvent::JobResult { .. } => "job_result",
                     SseEvent::Heartbeat => "heartbeat",
                 };
                 Ok(Event::default().event(event_type).data(data))
@@ -152,13 +160,14 @@ mod tests {
 
         manager.broadcast(SseEvent::Status {
             message: "test".to_string(),
+            thread_id: None,
         });
 
         let event = rx.next().await;
         assert!(event.is_some());
         let event = event.unwrap().unwrap();
         match event {
-            SseEvent::Status { message } => assert_eq!(message, "test"),
+            SseEvent::Status { message, .. } => assert_eq!(message, "test"),
             _ => panic!("unexpected event type"),
         }
     }
@@ -172,11 +181,12 @@ mod tests {
 
         manager.broadcast(SseEvent::Thinking {
             message: "working".to_string(),
+            thread_id: None,
         });
 
         let event = stream.next().await.unwrap();
         match event {
-            SseEvent::Thinking { message } => assert_eq!(message, "working"),
+            SseEvent::Thinking { message, .. } => assert_eq!(message, "working"),
             _ => panic!("Expected Thinking event"),
         }
     }

src/channels/web/static/index.html

@@ -10,12 +10,19 @@
 <body>
   <!-- Auth Screen -->
   <div id="auth-screen">
-    <h1>IronClaw</h1>
-    <div class="auth-form">
-      <input type="password" id="token-input" placeholder="Auth token" autofocus>
-      <button onclick="authenticate()">Connect</button>
+    <div class="auth-card-login">
+      <div class="auth-brand">
+        <h1>IronClaw</h1>
+        <p class="auth-tagline">Secure AI Assistant</p>
+      </div>
+      <div class="auth-form">
+        <label for="token-input">Gateway Token</label>
+        <input type="password" id="token-input" placeholder="Paste your auth token" autofocus>
+        <button onclick="authenticate()">Connect</button>
+      </div>
+      <div id="auth-error"></div>
+      <p class="auth-hint">Enter the GATEWAY_AUTH_TOKEN from your .env configuration.</p>
     </div>
-    <div id="auth-error"></div>
   </div>
 
   <!-- Main App (hidden until authenticated) -->
@@ -26,16 +33,33 @@
       <button data-tab="memory">Memory</button>
       <button data-tab="jobs">Jobs</button>
       <button data-tab="logs">Logs</button>
+      <button data-tab="routines">Routines</button>
       <button data-tab="extensions">Extensions</button>
       <div class="spacer"></div>
-      <div class="status">
+      <div class="status" id="gateway-status-trigger">
         <div class="dot" id="sse-dot"></div>
         <span id="sse-status">Connected</span>
+        <div class="gateway-popover" id="gateway-popover"></div>
       </div>
     </div>
 
     <!-- Chat Tab -->
     <div class="tab-panel active" id="tab-chat">
+      <div class="thread-sidebar" id="thread-sidebar">
+        <div class="thread-sidebar-header">
+          <span>Threads</span>
+          <button class="thread-new-btn" onclick="createNewThread()" title="New thread (Ctrl/Cmd+N)">+</button>
+          <button class="thread-toggle-btn" id="thread-toggle-btn" onclick="toggleThreadSidebar()" title="Toggle sidebar">&laquo;</button>
+        </div>
+        <div class="assistant-item" id="assistant-thread" onclick="switchToAssistant()">
+          <span class="assistant-label">Assistant</span>
+          <span class="assistant-meta" id="assistant-meta"></span>
+        </div>
+        <div class="threads-section-header">
+          <span>Conversations</span>
+        </div>
+        <div class="thread-list" id="thread-list"></div>
+      </div>
       <div class="chat-container">
         <div class="chat-messages" id="chat-messages"></div>
         <div class="chat-status" id="chat-status"></div>
@@ -56,10 +80,20 @@
           <div class="memory-tree" id="memory-tree"></div>
         </div>
         <div class="memory-content">
-          <div class="memory-breadcrumb" id="memory-breadcrumb">workspace /</div>
+          <div class="memory-breadcrumb" id="memory-breadcrumb">
+            <span id="memory-breadcrumb-path">workspace /</span>
+            <button class="memory-edit-btn" id="memory-edit-btn" style="display:none" onclick="startMemoryEdit()">Edit</button>
+          </div>
           <div class="memory-viewer" id="memory-viewer">
             <div class="empty">Select a file to view its contents</div>
           </div>
+          <div class="memory-editor" id="memory-editor" style="display:none">
+            <textarea id="memory-edit-textarea"></textarea>
+            <div class="memory-editor-actions">
+              <button class="btn-save" onclick="saveMemoryEdit()">Save</button>
+              <button class="btn-cancel-edit" onclick="cancelMemoryEdit()">Cancel</button>
+            </div>
+          </div>
         </div>
       </div>
     </div>
@@ -73,6 +107,7 @@
             <tr>
               <th>ID</th>
               <th>Title</th>
+              <th>Source</th>
               <th>Status</th>
               <th>Created</th>
               <th>Actions</th>
@@ -104,9 +139,48 @@
       </div>
     </div>
 
+    <!-- Routines Tab -->
+    <div class="tab-panel" id="tab-routines">
+      <div class="routines-container">
+        <div class="routines-summary" id="routines-summary"></div>
+        <table class="routines-table" id="routines-table">
+          <thead>
+            <tr>
+              <th>Name</th>
+              <th>Trigger</th>
+              <th>Action</th>
+              <th>Last Run</th>
+              <th>Next Run</th>
+              <th>Runs</th>
+              <th>Status</th>
+              <th>Actions</th>
+            </tr>
+          </thead>
+          <tbody id="routines-tbody"></tbody>
+        </table>
+        <div class="empty-state" id="routines-empty" style="display:none">
+          No routines configured. Ask the assistant to create one.
+        </div>
+        <div class="routine-detail" id="routine-detail" style="display:none"></div>
+      </div>
+    </div>
+
     <!-- Extensions Tab -->
     <div class="tab-panel" id="tab-extensions">
       <div class="extensions-container">
+        <div class="extensions-section">
+          <h3>Install Extension</h3>
+          <div class="ext-install-form" id="ext-install-form">
+            <input type="text" id="ext-install-name" placeholder="Extension name (required)">
+            <input type="text" id="ext-install-url" placeholder="URL (optional)">
+            <select id="ext-install-kind">
+              <option value="mcp_server">MCP Server</option>
+              <option value="wasm_tool">WASM Tool</option>
+              <option value="wasm_channel">WASM Channel</option>
+            </select>
+            <button onclick="installExtension()">Install</button>
+          </div>
+        </div>
         <div class="extensions-section">
           <h3>Installed Extensions</h3>
           <div class="extensions-list" id="extensions-list">
@@ -130,6 +204,7 @@
     </div>
   </div>
 
+  <div id="toasts"></div>
   <script src="/app.js"></script>
 </body>
 </html>

src/channels/web/types.rs

@@ -24,10 +24,17 @@ pub struct ThreadInfo {
     pub turn_count: usize,
     pub created_at: String,
     pub updated_at: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub title: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub thread_type: Option<String>,
 }
 
 #[derive(Debug, Serialize)]
 pub struct ThreadListResponse {
+    /// The pinned assistant thread (always present after first load).
+    pub assistant_thread: Option<ThreadInfo>,
+    /// Regular conversation threads.
     pub threads: Vec<ThreadInfo>,
     pub active_thread: Option<Uuid>,
 }
@@ -54,6 +61,12 @@ pub struct ToolCallInfo {
 pub struct HistoryResponse {
     pub thread_id: Uuid,
     pub turns: Vec<TurnInfo>,
+    /// Whether there are older messages available.
+    #[serde(default)]
+    pub has_more: bool,
+    /// Cursor for the next page (ISO8601 timestamp of the oldest message returned).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub oldest_timestamp: Option<String>,
 }
 
 // --- Approval ---
@@ -63,6 +76,8 @@ pub struct ApprovalRequest {
     pub request_id: String,
     /// "approve", "always", or "deny"
     pub action: String,
+    /// Thread that owns the pending approval (so the agent loop finds the right session).
+    pub thread_id: Option<String>,
 }
 
 // --- SSE Event Types ---
@@ -73,17 +88,49 @@ pub enum SseEvent {
     #[serde(rename = "response")]
     Response { content: String, thread_id: String },
     #[serde(rename = "thinking")]
-    Thinking { message: String },
+    Thinking {
+        message: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
     #[serde(rename = "tool_started")]
-    ToolStarted { name: String },
+    ToolStarted {
+        name: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
     #[serde(rename = "tool_completed")]
-    ToolCompleted { name: String, success: bool },
+    ToolCompleted {
+        name: String,
+        success: bool,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
     #[serde(rename = "tool_result")]
-    ToolResult { name: String, preview: String },
+    ToolResult {
+        name: String,
+        preview: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
     #[serde(rename = "stream_chunk")]
-    StreamChunk { content: String },
+    StreamChunk {
+        content: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
     #[serde(rename = "status")]
-    Status { message: String },
+    Status {
+        message: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+    #[serde(rename = "job_started")]
+    JobStarted {
+        job_id: String,
+        title: String,
+        browse_url: String,
+    },
     #[serde(rename = "approval_needed")]
     ApprovalNeeded {
         request_id: String,
@@ -91,10 +138,59 @@ pub enum SseEvent {
         description: String,
         parameters: String,
     },
+    #[serde(rename = "auth_required")]
+    AuthRequired {
+        extension_name: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        instructions: Option<String>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        auth_url: Option<String>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        setup_url: Option<String>,
+    },
+    #[serde(rename = "auth_completed")]
+    AuthCompleted {
+        extension_name: String,
+        success: bool,
+        message: String,
+    },
     #[serde(rename = "error")]
-    Error { message: String },
+    Error {
+        message: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
     #[serde(rename = "heartbeat")]
     Heartbeat,
+
+    // Sandbox job streaming events (worker + Claude Code bridge)
+    #[serde(rename = "job_message")]
+    JobMessage {
+        job_id: String,
+        role: String,
+        content: String,
+    },
+    #[serde(rename = "job_tool_use")]
+    JobToolUse {
+        job_id: String,
+        tool_name: String,
+        input: serde_json::Value,
+    },
+    #[serde(rename = "job_tool_result")]
+    JobToolResult {
+        job_id: String,
+        tool_name: String,
+        output: String,
+    },
+    #[serde(rename = "job_status")]
+    JobStatus { job_id: String, message: String },
+    #[serde(rename = "job_result")]
+    JobResult {
+        job_id: String,
+        status: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        session_id: Option<String>,
+    },
 }
 
 // --- Memory ---
@@ -188,6 +284,54 @@ pub struct JobSummaryResponse {
     pub stuck: usize,
 }
 
+#[derive(Debug, Serialize)]
+pub struct JobDetailResponse {
+    pub id: Uuid,
+    pub title: String,
+    pub description: String,
+    pub state: String,
+    pub user_id: String,
+    pub created_at: String,
+    pub started_at: Option<String>,
+    pub completed_at: Option<String>,
+    pub elapsed_secs: Option<u64>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub project_dir: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub browse_url: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub job_mode: Option<String>,
+    pub transitions: Vec<TransitionInfo>,
+}
+
+// --- Project Files ---
+
+#[derive(Debug, Serialize)]
+pub struct ProjectFileEntry {
+    pub name: String,
+    pub path: String,
+    pub is_dir: bool,
+}
+
+#[derive(Debug, Serialize)]
+pub struct ProjectFilesResponse {
+    pub entries: Vec<ProjectFileEntry>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct ProjectFileReadResponse {
+    pub path: String,
+    pub content: String,
+}
+
+#[derive(Debug, Serialize)]
+pub struct TransitionInfo {
+    pub from: String,
+    pub to: String,
+    pub timestamp: String,
+    pub reason: Option<String>,
+}
+
 // --- Extensions ---
 
 #[derive(Debug, Serialize)]
@@ -262,6 +406,21 @@ impl ActionResponse {
     }
 }
 
+// --- Auth Token ---
+
+/// Request to submit an auth token for an extension (dedicated endpoint).
+#[derive(Debug, Deserialize)]
+pub struct AuthTokenRequest {
+    pub extension_name: String,
+    pub token: String,
+}
+
+/// Request to cancel an in-progress auth flow.
+#[derive(Debug, Deserialize)]
+pub struct AuthCancelRequest {
+    pub extension_name: String,
+}
+
 // --- WebSocket ---
 
 /// Message sent by a WebSocket client to the server.
@@ -280,7 +439,18 @@ pub enum WsClientMessage {
         request_id: String,
         /// "approve", "always", or "deny"
         action: String,
+        /// Thread that owns the pending approval.
+        thread_id: Option<String>,
     },
+    /// Submit an auth token for an extension (bypasses message pipeline).
+    #[serde(rename = "auth_token")]
+    AuthToken {
+        extension_name: String,
+        token: String,
+    },
+    /// Cancel an in-progress auth flow.
+    #[serde(rename = "auth_cancel")]
+    AuthCancel { extension_name: String },
     /// Client heartbeat ping.
     #[serde(rename = "ping")]
     Ping,
@@ -314,11 +484,20 @@ impl WsServerMessage {
             SseEvent::Thinking { .. } => "thinking",
             SseEvent::ToolStarted { .. } => "tool_started",
             SseEvent::ToolCompleted { .. } => "tool_completed",
+            SseEvent::ToolResult { .. } => "tool_result",
             SseEvent::StreamChunk { .. } => "stream_chunk",
             SseEvent::Status { .. } => "status",
+            SseEvent::JobStarted { .. } => "job_started",
             SseEvent::ApprovalNeeded { .. } => "approval_needed",
+            SseEvent::AuthRequired { .. } => "auth_required",
+            SseEvent::AuthCompleted { .. } => "auth_completed",
             SseEvent::Error { .. } => "error",
             SseEvent::Heartbeat => "heartbeat",
+            SseEvent::JobMessage { .. } => "job_message",
+            SseEvent::JobToolUse { .. } => "job_tool_use",
+            SseEvent::JobToolResult { .. } => "job_tool_result",
+            SseEvent::JobStatus { .. } => "job_status",
+            SseEvent::JobResult { .. } => "job_result",
         };
         let data = serde_json::to_value(event).unwrap_or(serde_json::Value::Null);
         WsServerMessage::Event {
@@ -328,6 +507,96 @@ impl WsServerMessage {
     }
 }
 
+// --- Routines ---
+
+#[derive(Debug, Serialize)]
+pub struct RoutineInfo {
+    pub id: Uuid,
+    pub name: String,
+    pub description: String,
+    pub enabled: bool,
+    pub trigger_type: String,
+    pub trigger_summary: String,
+    pub action_type: String,
+    pub last_run_at: Option<String>,
+    pub next_fire_at: Option<String>,
+    pub run_count: u64,
+    pub consecutive_failures: u32,
+    pub status: String,
+}
+
+#[derive(Debug, Serialize)]
+pub struct RoutineListResponse {
+    pub routines: Vec<RoutineInfo>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct RoutineSummaryResponse {
+    pub total: u64,
+    pub enabled: u64,
+    pub disabled: u64,
+    pub failing: u64,
+    pub runs_today: u64,
+}
+
+#[derive(Debug, Serialize)]
+pub struct RoutineDetailResponse {
+    pub id: Uuid,
+    pub name: String,
+    pub description: String,
+    pub enabled: bool,
+    pub trigger: serde_json::Value,
+    pub action: serde_json::Value,
+    pub guardrails: serde_json::Value,
+    pub notify: serde_json::Value,
+    pub last_run_at: Option<String>,
+    pub next_fire_at: Option<String>,
+    pub run_count: u64,
+    pub consecutive_failures: u32,
+    pub created_at: String,
+    pub recent_runs: Vec<RoutineRunInfo>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct RoutineRunInfo {
+    pub id: Uuid,
+    pub trigger_type: String,
+    pub started_at: String,
+    pub completed_at: Option<String>,
+    pub status: String,
+    pub result_summary: Option<String>,
+    pub tokens_used: Option<i32>,
+}
+
+// --- Settings ---
+
+#[derive(Debug, Serialize)]
+pub struct SettingResponse {
+    pub key: String,
+    pub value: serde_json::Value,
+    pub updated_at: String,
+}
+
+#[derive(Debug, Serialize)]
+pub struct SettingsListResponse {
+    pub settings: Vec<SettingResponse>,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct SettingWriteRequest {
+    pub value: serde_json::Value,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct SettingsImportRequest {
+    pub settings: std::collections::HashMap<String, serde_json::Value>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct SettingsExportResponse {
+    pub settings: std::collections::HashMap<String, serde_json::Value>,
+}
+
 // --- Health ---
 
 #[derive(Debug, Serialize)]
@@ -370,12 +639,36 @@ mod tests {
 
     #[test]
     fn test_ws_client_approval_parse() {
-        let json = r#"{"type":"approval","request_id":"abc-123","action":"approve"}"#;
+        let json =
+            r#"{"type":"approval","request_id":"abc-123","action":"approve","thread_id":"t1"}"#;
         let msg: WsClientMessage = serde_json::from_str(json).unwrap();
         match msg {
-            WsClientMessage::Approval { request_id, action } => {
+            WsClientMessage::Approval {
+                request_id,
+                action,
+                thread_id,
+            } => {
                 assert_eq!(request_id, "abc-123");
                 assert_eq!(action, "approve");
+                assert_eq!(thread_id.as_deref(), Some("t1"));
+            }
+            _ => panic!("Expected Approval variant"),
+        }
+    }
+
+    #[test]
+    fn test_ws_client_approval_parse_no_thread() {
+        let json = r#"{"type":"approval","request_id":"abc-123","action":"deny"}"#;
+        let msg: WsClientMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            WsClientMessage::Approval {
+                request_id,
+                action,
+                thread_id,
+            } => {
+                assert_eq!(request_id, "abc-123");
+                assert_eq!(action, "deny");
+                assert!(thread_id.is_none());
             }
             _ => panic!("Expected Approval variant"),
         }
@@ -436,6 +729,7 @@ mod tests {
     fn test_ws_server_from_sse_thinking() {
         let sse = SseEvent::Thinking {
             message: "reasoning...".to_string(),
+            thread_id: None,
         };
         let ws = WsServerMessage::from_sse_event(&sse);
         match ws {
@@ -476,4 +770,115 @@ mod tests {
             _ => panic!("Expected Event variant"),
         }
     }
+
+    // ---- Auth type tests ----
+
+    #[test]
+    fn test_ws_client_auth_token_parse() {
+        let json = r#"{"type":"auth_token","extension_name":"notion","token":"sk-123"}"#;
+        let msg: WsClientMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            WsClientMessage::AuthToken {
+                extension_name,
+                token,
+            } => {
+                assert_eq!(extension_name, "notion");
+                assert_eq!(token, "sk-123");
+            }
+            _ => panic!("Expected AuthToken variant"),
+        }
+    }
+
+    #[test]
+    fn test_ws_client_auth_cancel_parse() {
+        let json = r#"{"type":"auth_cancel","extension_name":"notion"}"#;
+        let msg: WsClientMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            WsClientMessage::AuthCancel { extension_name } => {
+                assert_eq!(extension_name, "notion");
+            }
+            _ => panic!("Expected AuthCancel variant"),
+        }
+    }
+
+    #[test]
+    fn test_sse_auth_required_serialize() {
+        let event = SseEvent::AuthRequired {
+            extension_name: "notion".to_string(),
+            instructions: Some("Get your token from...".to_string()),
+            auth_url: None,
+            setup_url: Some("https://notion.so/integrations".to_string()),
+        };
+        let json = serde_json::to_string(&event).unwrap();
+        let parsed: serde_json::Value = serde_json::from_str(&json).unwrap();
+        assert_eq!(parsed["type"], "auth_required");
+        assert_eq!(parsed["extension_name"], "notion");
+        assert_eq!(parsed["instructions"], "Get your token from...");
+        assert!(parsed.get("auth_url").is_none());
+        assert_eq!(parsed["setup_url"], "https://notion.so/integrations");
+    }
+
+    #[test]
+    fn test_sse_auth_completed_serialize() {
+        let event = SseEvent::AuthCompleted {
+            extension_name: "notion".to_string(),
+            success: true,
+            message: "notion authenticated (3 tools loaded)".to_string(),
+        };
+        let json = serde_json::to_string(&event).unwrap();
+        let parsed: serde_json::Value = serde_json::from_str(&json).unwrap();
+        assert_eq!(parsed["type"], "auth_completed");
+        assert_eq!(parsed["extension_name"], "notion");
+        assert_eq!(parsed["success"], true);
+    }
+
+    #[test]
+    fn test_ws_server_from_sse_auth_required() {
+        let sse = SseEvent::AuthRequired {
+            extension_name: "openai".to_string(),
+            instructions: Some("Enter API key".to_string()),
+            auth_url: None,
+            setup_url: None,
+        };
+        let ws = WsServerMessage::from_sse_event(&sse);
+        match ws {
+            WsServerMessage::Event { event_type, data } => {
+                assert_eq!(event_type, "auth_required");
+                assert_eq!(data["extension_name"], "openai");
+            }
+            _ => panic!("Expected Event variant"),
+        }
+    }
+
+    #[test]
+    fn test_ws_server_from_sse_auth_completed() {
+        let sse = SseEvent::AuthCompleted {
+            extension_name: "slack".to_string(),
+            success: false,
+            message: "Invalid token".to_string(),
+        };
+        let ws = WsServerMessage::from_sse_event(&sse);
+        match ws {
+            WsServerMessage::Event { event_type, data } => {
+                assert_eq!(event_type, "auth_completed");
+                assert_eq!(data["success"], false);
+            }
+            _ => panic!("Expected Event variant"),
+        }
+    }
+
+    #[test]
+    fn test_auth_token_request_deserialize() {
+        let json = r#"{"extension_name":"telegram","token":"bot12345"}"#;
+        let req: AuthTokenRequest = serde_json::from_str(json).unwrap();
+        assert_eq!(req.extension_name, "telegram");
+        assert_eq!(req.token, "bot12345");
+    }
+
+    #[test]
+    fn test_auth_cancel_request_deserialize() {
+        let json = r#"{"extension_name":"telegram"}"#;
+        let req: AuthCancelRequest = serde_json::from_str(json).unwrap();
+        assert_eq!(req.extension_name, "telegram");
+    }
 }

src/channels/web/ws.rs

@@ -170,7 +170,11 @@ async fn handle_client_message(
                     .await;
             }
         }
-        WsClientMessage::Approval { request_id, action } => {
+        WsClientMessage::Approval {
+            request_id,
+            action,
+            thread_id,
+        } => {
             let (approved, always) = match action.as_str() {
                 "approve" => (true, false),
                 "always" => (true, true),
@@ -214,12 +218,71 @@ async fn handle_client_message(
                 }
             };
 
-            let msg = IncomingMessage::new("gateway", user_id, content);
+            let mut msg = IncomingMessage::new("gateway", user_id, content);
+            if let Some(ref tid) = thread_id {
+                msg = msg.with_thread(tid);
+            }
             let tx_guard = state.msg_tx.read().await;
             if let Some(ref tx) = *tx_guard {
                 let _ = tx.send(msg).await;
             }
         }
+        WsClientMessage::AuthToken {
+            extension_name,
+            token,
+        } => {
+            if let Some(ref ext_mgr) = state.extension_manager {
+                match ext_mgr.auth(&extension_name, Some(&token)).await {
+                    Ok(result) if result.status == "authenticated" => {
+                        let msg = match ext_mgr.activate(&extension_name).await {
+                            Ok(r) => format!(
+                                "{} authenticated ({} tools loaded)",
+                                extension_name,
+                                r.tools_loaded.len()
+                            ),
+                            Err(e) => format!(
+                                "{} authenticated but activation failed: {}",
+                                extension_name, e
+                            ),
+                        };
+                        crate::channels::web::server::clear_auth_mode(state).await;
+                        state
+                            .sse
+                            .broadcast(crate::channels::web::types::SseEvent::AuthCompleted {
+                                extension_name,
+                                success: true,
+                                message: msg,
+                            });
+                    }
+                    Ok(result) => {
+                        state
+                            .sse
+                            .broadcast(crate::channels::web::types::SseEvent::AuthRequired {
+                                extension_name,
+                                instructions: result.instructions,
+                                auth_url: result.auth_url,
+                                setup_url: result.setup_url,
+                            });
+                    }
+                    Err(e) => {
+                        let _ = direct_tx
+                            .send(WsServerMessage::Error {
+                                message: format!("Auth failed: {}", e),
+                            })
+                            .await;
+                    }
+                }
+            } else {
+                let _ = direct_tx
+                    .send(WsServerMessage::Error {
+                        message: "Extension manager not available".to_string(),
+                    })
+                    .await;
+            }
+        }
+        WsClientMessage::AuthCancel { .. } => {
+            crate::channels::web::server::clear_auth_mode(state).await;
+        }
         WsClientMessage::Ping => {
             let _ = direct_tx.send(WsServerMessage::Pong).await;
         }
@@ -328,6 +391,7 @@ mod tests {
             WsClientMessage::Approval {
                 request_id: request_id.to_string(),
                 action: "approve".to_string(),
+                thread_id: Some("thread-42".to_string()),
             },
             &state,
             "user1",
@@ -338,6 +402,8 @@ mod tests {
         let incoming = agent_rx.recv().await.unwrap();
         // The content should be a serialized ExecApproval
         assert!(incoming.content.contains("ExecApproval"));
+        // Thread should be forwarded onto the IncomingMessage.
+        assert_eq!(incoming.thread_id.as_deref(), Some("thread-42"));
     }
 
     #[tokio::test]
@@ -349,6 +415,7 @@ mod tests {
             WsClientMessage::Approval {
                 request_id: Uuid::new_v4().to_string(),
                 action: "maybe".to_string(),
+                thread_id: None,
             },
             &state,
             "user1",
@@ -374,6 +441,7 @@ mod tests {
             WsClientMessage::Approval {
                 request_id: "not-a-uuid".to_string(),
                 action: "approve".to_string(),
+                thread_id: None,
             },
             &state,
             "user1",
@@ -398,11 +466,13 @@ mod tests {
             msg_tx: tokio::sync::RwLock::new(msg_tx),
             sse: SseManager::new(),
             workspace: None,
-            context_manager: None,
             session_manager: None,
             log_broadcaster: None,
             extension_manager: None,
             tool_registry: None,
+            store: None,
+            job_manager: None,
+            prompt_queue: None,
             user_id: "test".to_string(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: Some(Arc::new(WsConnectionTracker::new())),

src/cli/config.rs

@@ -1,6 +1,7 @@
 //! Configuration management CLI commands.
 //!
 //! Commands for viewing and modifying settings.
+//! Settings are stored in PostgreSQL (env > DB > default).
 
 use clap::Subcommand;
 
@@ -36,41 +37,82 @@ pub enum ConfigCommand {
         path: String,
     },
 
-    /// Show the settings file path
+    /// Show the settings storage info
     Path,
 }
 
 /// Run a config command.
-pub fn run_config_command(cmd: ConfigCommand) -> anyhow::Result<()> {
+///
+/// Connects to the database to read/write settings. Falls back to disk
+/// if the database is not available.
+pub async fn run_config_command(cmd: ConfigCommand) -> anyhow::Result<()> {
+    let _ = dotenvy::dotenv();
+
+    // Try to connect to the DB for settings access
+    let store = match connect_store().await {
+        Ok(s) => Some(s),
+        Err(e) => {
+            eprintln!(
+                "Warning: Could not connect to database ({}), using disk fallback",
+                e
+            );
+            None
+        }
+    };
+
     match cmd {
-        ConfigCommand::List { filter } => list_settings(filter),
-        ConfigCommand::Get { path } => get_setting(&path),
-        ConfigCommand::Set { path, value } => set_setting(&path, &value),
-        ConfigCommand::Reset { path } => reset_setting(&path),
-        ConfigCommand::Path => show_path(),
+        ConfigCommand::List { filter } => list_settings(store.as_ref(), filter).await,
+        ConfigCommand::Get { path } => get_setting(store.as_ref(), &path).await,
+        ConfigCommand::Set { path, value } => set_setting(store.as_ref(), &path, &value).await,
+        ConfigCommand::Reset { path } => reset_setting(store.as_ref(), &path).await,
+        ConfigCommand::Path => show_path(store.is_some()),
     }
 }
 
+/// Bootstrap a DB connection for config commands.
+async fn connect_store() -> anyhow::Result<crate::history::Store> {
+    let config = crate::config::Config::from_env()
+        .await
+        .map_err(|e| anyhow::anyhow!("{}", e))?;
+    let store = crate::history::Store::new(&config.database).await?;
+    store.run_migrations().await?;
+    Ok(store)
+}
+
+const DEFAULT_USER_ID: &str = "default";
+
+/// Load settings: DB if available, else disk.
+async fn load_settings(store: Option<&crate::history::Store>) -> Settings {
+    if let Some(store) = store {
+        match store.get_all_settings(DEFAULT_USER_ID).await {
+            Ok(map) if !map.is_empty() => return Settings::from_db_map(&map),
+            _ => {}
+        }
+    }
+    Settings::load()
+}
+
 /// List all settings.
-fn list_settings(filter: Option<String>) -> anyhow::Result<()> {
-    let settings = Settings::load();
+async fn list_settings(
+    store: Option<&crate::history::Store>,
+    filter: Option<String>,
+) -> anyhow::Result<()> {
+    let settings = load_settings(store).await;
     let all = settings.list();
 
-    // Find the longest key for alignment
     let max_key_len = all.iter().map(|(k, _)| k.len()).max().unwrap_or(0);
 
-    println!("Settings:");
+    let source = if store.is_some() { "database" } else { "disk" };
+    println!("Settings (source: {}):", source);
     println!();
 
     for (key, value) in all {
-        // Skip if filter is set and doesn't match
         if let Some(ref f) = filter {
             if !key.starts_with(f) {
                 continue;
             }
         }
 
-        // Truncate long values for display
         let display_value = if value.len() > 60 {
             format!("{}...", &value[..57])
         } else {
@@ -84,8 +126,8 @@ fn list_settings(filter: Option<String>) -> anyhow::Result<()> {
 }
 
 /// Get a specific setting.
-fn get_setting(path: &str) -> anyhow::Result<()> {
-    let settings = Settings::load();
+async fn get_setting(store: Option<&crate::history::Store>, path: &str) -> anyhow::Result<()> {
+    let settings = load_settings(store).await;
 
     match settings.get(path) {
         Some(value) => {
@@ -99,67 +141,92 @@ fn get_setting(path: &str) -> anyhow::Result<()> {
 }
 
 /// Set a setting value.
-fn set_setting(path: &str, value: &str) -> anyhow::Result<()> {
-    let mut settings = Settings::load();
+async fn set_setting(
+    store: Option<&crate::history::Store>,
+    path: &str,
+    value: &str,
+) -> anyhow::Result<()> {
+    let mut settings = load_settings(store).await;
 
-    // Try to set the value
     settings
         .set(path, value)
         .map_err(|e| anyhow::anyhow!("{}", e))?;
 
-    // Save to disk
-    settings.save()?;
+    // Save to DB if available, otherwise disk
+    if let Some(store) = store {
+        let json_value = match serde_json::from_str::<serde_json::Value>(value) {
+            Ok(v) => v,
+            Err(_) => serde_json::Value::String(value.to_string()),
+        };
+        store
+            .set_setting(DEFAULT_USER_ID, path, &json_value)
+            .await
+            .map_err(|e| anyhow::anyhow!("Failed to save to database: {}", e))?;
+    } else {
+        settings.save()?;
+    }
 
     println!("Set {} = {}", path, value);
     Ok(())
 }
 
 /// Reset a setting to default.
-fn reset_setting(path: &str) -> anyhow::Result<()> {
-    let mut settings = Settings::load();
-
-    // Get the default value for display
+async fn reset_setting(store: Option<&crate::history::Store>, path: &str) -> anyhow::Result<()> {
     let default = Settings::default();
     let default_value = default
         .get(path)
         .ok_or_else(|| anyhow::anyhow!("Unknown setting: {}", path))?;
 
-    // Reset it
-    settings.reset(path).map_err(|e| anyhow::anyhow!("{}", e))?;
-
-    // Save to disk
-    settings.save()?;
+    // Delete from DB (falling back to default) or reset on disk
+    if let Some(store) = store {
+        store
+            .delete_setting(DEFAULT_USER_ID, path)
+            .await
+            .map_err(|e| anyhow::anyhow!("Failed to delete setting from database: {}", e))?;
+    } else {
+        let mut settings = Settings::load();
+        settings.reset(path).map_err(|e| anyhow::anyhow!("{}", e))?;
+        settings.save()?;
+    }
 
     println!("Reset {} to default: {}", path, default_value);
     Ok(())
 }
 
-/// Show the settings file path.
-fn show_path() -> anyhow::Result<()> {
-    let path = Settings::default_path();
-    println!("{}", path.display());
-
-    if path.exists() {
-        let metadata = std::fs::metadata(&path)?;
-        println!("  Size: {} bytes", metadata.len());
-        if let Ok(modified) = metadata.modified() {
-            use std::time::SystemTime;
-            let duration = SystemTime::now()
-                .duration_since(modified)
-                .unwrap_or_default();
-            let secs = duration.as_secs();
-            if secs < 60 {
-                println!("  Modified: {} seconds ago", secs);
-            } else if secs < 3600 {
-                println!("  Modified: {} minutes ago", secs / 60);
-            } else if secs < 86400 {
-                println!("  Modified: {} hours ago", secs / 3600);
-            } else {
-                println!("  Modified: {} days ago", secs / 86400);
-            }
-        }
+/// Show the settings storage info.
+fn show_path(has_db: bool) -> anyhow::Result<()> {
+    if has_db {
+        println!("Settings stored in: PostgreSQL (settings table)");
+        println!(
+            "Bootstrap config:   {}",
+            crate::bootstrap::BootstrapConfig::default_path().display()
+        );
     } else {
-        println!("  (does not exist, using defaults)");
+        let path = Settings::default_path();
+        println!("Settings stored in: {} (disk fallback)", path.display());
+
+        if path.exists() {
+            let metadata = std::fs::metadata(&path)?;
+            println!("  Size: {} bytes", metadata.len());
+            if let Ok(modified) = metadata.modified() {
+                use std::time::SystemTime;
+                let duration = SystemTime::now()
+                    .duration_since(modified)
+                    .unwrap_or_default();
+                let secs = duration.as_secs();
+                if secs < 60 {
+                    println!("  Modified: {} seconds ago", secs);
+                } else if secs < 3600 {
+                    println!("  Modified: {} minutes ago", secs / 60);
+                } else if secs < 86400 {
+                    println!("  Modified: {} hours ago", secs / 3600);
+                } else {
+                    println!("  Modified: {} days ago", secs / 86400);
+                }
+            }
+        } else {
+            println!("  (does not exist, using defaults)");
+        }
     }
 
     Ok(())

src/cli/mcp.rs

@@ -13,9 +13,7 @@ use crate::secrets::{PostgresSecretsStore, SecretsCrypto, SecretsStore};
 use crate::tools::mcp::{
     McpClient, McpServerConfig, McpSessionManager, OAuthConfig,
     auth::{authorize_mcp_server, is_authenticated},
-    config::{
-        add_mcp_server, get_mcp_server, load_mcp_servers, remove_mcp_server, save_mcp_servers,
-    },
+    config::{self, McpServersFile},
 };
 
 #[derive(Subcommand, Debug, Clone)]
@@ -173,8 +171,11 @@ async fn add_server(
     // Validate
     config.validate()?;
 
-    // Save
-    add_mcp_server(config).await?;
+    // Save (DB if available, else disk)
+    let store = connect_store().await;
+    let mut servers = load_servers(store.as_ref()).await?;
+    servers.upsert(config);
+    save_servers(store.as_ref(), &servers).await?;
 
     println!();
     println!("  ✓ Added MCP server '{}'", name);
@@ -192,7 +193,12 @@ async fn add_server(
 
 /// Remove an MCP server.
 async fn remove_server(name: String) -> anyhow::Result<()> {
-    remove_mcp_server(&name).await?;
+    let store = connect_store().await;
+    let mut servers = load_servers(store.as_ref()).await?;
+    if !servers.remove(&name) {
+        anyhow::bail!("Server '{}' not found", name);
+    }
+    save_servers(store.as_ref(), &servers).await?;
 
     println!();
     println!("  ✓ Removed MCP server '{}'", name);
@@ -203,7 +209,8 @@ async fn remove_server(name: String) -> anyhow::Result<()> {
 
 /// List configured MCP servers.
 async fn list_servers(verbose: bool) -> anyhow::Result<()> {
-    let servers = load_mcp_servers().await?;
+    let store = connect_store().await;
+    let servers = load_servers(store.as_ref()).await?;
 
     if servers.servers.is_empty() {
         println!();
@@ -261,7 +268,12 @@ async fn list_servers(verbose: bool) -> anyhow::Result<()> {
 /// Authenticate with an MCP server.
 async fn auth_server(name: String, user_id: String) -> anyhow::Result<()> {
     // Get server config
-    let server = get_mcp_server(&name).await?;
+    let store = connect_store().await;
+    let servers = load_servers(store.as_ref()).await?;
+    let server = servers
+        .get(&name)
+        .cloned()
+        .ok_or_else(|| anyhow::anyhow!("Server '{}' not found", name))?;
 
     // Initialize secrets store
     let secrets = get_secrets_store().await?;
@@ -329,7 +341,12 @@ async fn auth_server(name: String, user_id: String) -> anyhow::Result<()> {
 /// Test connection to an MCP server.
 async fn test_server(name: String, user_id: String) -> anyhow::Result<()> {
     // Get server config
-    let server = get_mcp_server(&name).await?;
+    let store = connect_store().await;
+    let servers = load_servers(store.as_ref()).await?;
+    let server = servers
+        .get(&name)
+        .cloned()
+        .ok_or_else(|| anyhow::anyhow!("Server '{}' not found", name))?;
 
     println!();
     println!("  Testing connection to '{}'...", name);
@@ -420,7 +437,8 @@ async fn test_server(name: String, user_id: String) -> anyhow::Result<()> {
 
 /// Toggle server enabled/disabled state.
 async fn toggle_server(name: String, enable: bool, disable: bool) -> anyhow::Result<()> {
-    let mut servers = load_mcp_servers().await?;
+    let store = connect_store().await;
+    let mut servers = load_servers(store.as_ref()).await?;
 
     let server = servers
         .get_mut(&name)
@@ -435,7 +453,7 @@ async fn toggle_server(name: String, enable: bool, disable: bool) -> anyhow::Res
     };
 
     server.enabled = new_state;
-    save_mcp_servers(&servers).await?;
+    save_servers(store.as_ref(), &servers).await?;
 
     let status = if new_state { "enabled" } else { "disabled" };
     println!();
@@ -445,9 +463,40 @@ async fn toggle_server(name: String, enable: bool, disable: bool) -> anyhow::Res
     Ok(())
 }
 
+const DEFAULT_USER_ID: &str = "default";
+
+/// Try to connect to the database store for DB-backed config.
+async fn connect_store() -> Option<Store> {
+    let config = Config::from_env().await.ok()?;
+    let store = Store::new(&config.database).await.ok()?;
+    store.run_migrations().await.ok()?;
+    Some(store)
+}
+
+/// Load MCP servers (DB if available, else disk).
+async fn load_servers(store: Option<&Store>) -> Result<McpServersFile, config::ConfigError> {
+    if let Some(store) = store {
+        config::load_mcp_servers_from_db(store, DEFAULT_USER_ID).await
+    } else {
+        config::load_mcp_servers().await
+    }
+}
+
+/// Save MCP servers (DB if available, else disk).
+async fn save_servers(
+    store: Option<&Store>,
+    servers: &McpServersFile,
+) -> Result<(), config::ConfigError> {
+    if let Some(store) = store {
+        config::save_mcp_servers_to_db(store, DEFAULT_USER_ID, servers).await
+    } else {
+        config::save_mcp_servers(servers).await
+    }
+}
+
 /// Initialize and return the secrets store.
 async fn get_secrets_store() -> anyhow::Result<Arc<dyn SecretsStore + Send + Sync>> {
-    let config = Config::from_env()?;
+    let config = Config::from_env().await?;
 
     let master_key = config.secrets.master_key().ok_or_else(|| {
         anyhow::anyhow!(

src/cli/mod.rs

@@ -12,12 +12,14 @@
 mod config;
 mod mcp;
 pub mod memory;
+mod pairing;
 pub mod status;
 mod tool;
 
 pub use config::{ConfigCommand, run_config_command};
 pub use mcp::{McpCommand, run_mcp_command};
 pub use memory::{MemoryCommand, run_memory_command};
+pub use pairing::{PairingCommand, run_pairing_command, run_pairing_command_with_store};
 pub use status::run_status_command;
 pub use tool::{ToolCommand, run_tool_command};
 
@@ -86,8 +88,48 @@ pub enum Command {
     #[command(subcommand)]
     Memory(MemoryCommand),
 
+    /// DM pairing (approve inbound requests from unknown senders)
+    #[command(subcommand)]
+    Pairing(PairingCommand),
+
     /// Show system health and diagnostics
     Status,
+
+    /// Run as a sandboxed worker inside a Docker container (internal use).
+    /// This is invoked automatically by the orchestrator, not by users directly.
+    Worker {
+        /// Job ID to execute.
+        #[arg(long)]
+        job_id: uuid::Uuid,
+
+        /// URL of the orchestrator's internal API.
+        #[arg(long, default_value = "http://host.docker.internal:50051")]
+        orchestrator_url: String,
+
+        /// Maximum iterations before stopping.
+        #[arg(long, default_value = "50")]
+        max_iterations: u32,
+    },
+
+    /// Run as a Claude Code bridge inside a Docker container (internal use).
+    /// Spawns the `claude` CLI and streams output back to the orchestrator.
+    ClaudeBridge {
+        /// Job ID to execute.
+        #[arg(long)]
+        job_id: uuid::Uuid,
+
+        /// URL of the orchestrator's internal API.
+        #[arg(long, default_value = "http://host.docker.internal:50051")]
+        orchestrator_url: String,
+
+        /// Maximum agentic turns for Claude Code.
+        #[arg(long, default_value = "50")]
+        max_turns: u32,
+
+        /// Claude model to use (e.g. "sonnet", "opus").
+        #[arg(long, default_value = "sonnet")]
+        model: String,
+    },
 }
 
 impl Cli {

src/cli/pairing.rs

@@ -0,0 +1,187 @@
+//! DM pairing CLI commands.
+//!
+//! Manage pairing requests for channels (Telegram, Slack, etc.).
+
+use clap::Subcommand;
+
+use crate::pairing::PairingStore;
+
+/// Pairing subcommands.
+#[derive(Subcommand, Debug, Clone)]
+pub enum PairingCommand {
+    /// List pending pairing requests
+    List {
+        /// Channel name (e.g., telegram, slack)
+        #[arg(required = true)]
+        channel: String,
+
+        /// Output as JSON
+        #[arg(long)]
+        json: bool,
+    },
+
+    /// Approve a pairing request by code
+    Approve {
+        /// Channel name (e.g., telegram, slack)
+        #[arg(required = true)]
+        channel: String,
+
+        /// Pairing code (e.g., ABC12345)
+        #[arg(required = true)]
+        code: String,
+    },
+}
+
+/// Run pairing CLI command.
+pub fn run_pairing_command(cmd: PairingCommand) -> Result<(), String> {
+    run_pairing_command_with_store(&PairingStore::new(), cmd)
+}
+
+/// Run pairing CLI command with a given store (for testing).
+pub fn run_pairing_command_with_store(
+    store: &PairingStore,
+    cmd: PairingCommand,
+) -> Result<(), String> {
+    match cmd {
+        PairingCommand::List { channel, json } => run_list(store, &channel, json),
+        PairingCommand::Approve { channel, code } => run_approve(store, &channel, &code),
+    }
+}
+
+fn run_list(store: &PairingStore, channel: &str, json: bool) -> Result<(), String> {
+    let requests = store.list_pending(channel).map_err(|e| e.to_string())?;
+
+    if json {
+        println!(
+            "{}",
+            serde_json::to_string_pretty(&requests).map_err(|e| e.to_string())?
+        );
+        return Ok(());
+    }
+
+    if requests.is_empty() {
+        println!("No pending {} pairing requests.", channel);
+        return Ok(());
+    }
+
+    println!("Pairing requests ({}):", requests.len());
+    for r in &requests {
+        let meta = r
+            .meta
+            .as_ref()
+            .and_then(|m| m.as_object())
+            .map(|o| {
+                o.iter()
+                    .filter_map(|(k, v)| v.as_str().map(|s| format!("{}={}", k, s)))
+                    .collect::<Vec<_>>()
+                    .join(", ")
+            })
+            .unwrap_or_default();
+        println!("  {}  {}  {}  {}", r.code, r.id, meta, r.created_at);
+    }
+
+    Ok(())
+}
+
+fn run_approve(store: &PairingStore, channel: &str, code: &str) -> Result<(), String> {
+    match store.approve(channel, code) {
+        Ok(Some(entry)) => {
+            println!("Approved {} sender {}.", channel, entry.id);
+            Ok(())
+        }
+        Ok(None) => Err(format!(
+            "No pending pairing request found for code: {}",
+            code
+        )),
+        Err(crate::pairing::PairingStoreError::ApproveRateLimited) => Err(
+            "Too many failed approve attempts. Wait a few minutes before trying again.".to_string(),
+        ),
+        Err(e) => Err(e.to_string()),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::TempDir;
+
+    fn test_store() -> (PairingStore, TempDir) {
+        let dir = TempDir::new().unwrap();
+        let store = PairingStore::with_base_dir(dir.path().to_path_buf());
+        (store, dir)
+    }
+
+    #[test]
+    fn test_list_empty_returns_ok() {
+        let (store, _) = test_store();
+        let result = run_pairing_command_with_store(
+            &store,
+            PairingCommand::List {
+                channel: "telegram".to_string(),
+                json: false,
+            },
+        );
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_list_json_empty_returns_ok() {
+        let (store, _) = test_store();
+        let result = run_pairing_command_with_store(
+            &store,
+            PairingCommand::List {
+                channel: "telegram".to_string(),
+                json: true,
+            },
+        );
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_approve_invalid_code_returns_err() {
+        let (store, _) = test_store();
+        // Create a pending request so the pairing file exists, then approve with wrong code
+        store.upsert_request("telegram", "user1", None).unwrap();
+
+        let result = run_pairing_command_with_store(
+            &store,
+            PairingCommand::Approve {
+                channel: "telegram".to_string(),
+                code: "BADCODE1".to_string(),
+            },
+        );
+        assert!(result.is_err());
+        assert!(result.unwrap_err().contains("No pending pairing request"));
+    }
+
+    #[test]
+    fn test_approve_valid_code_returns_ok() {
+        let (store, _) = test_store();
+        let r = store.upsert_request("telegram", "user1", None).unwrap();
+        assert!(r.created);
+
+        let result = run_pairing_command_with_store(
+            &store,
+            PairingCommand::Approve {
+                channel: "telegram".to_string(),
+                code: r.code,
+            },
+        );
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_list_with_pending_returns_ok() {
+        let (store, _) = test_store();
+        store.upsert_request("telegram", "user1", None).unwrap();
+
+        let result = run_pairing_command_with_store(
+            &store,
+            PairingCommand::List {
+                channel: "telegram".to_string(),
+                json: false,
+            },
+        );
+        assert!(result.is_ok());
+    }
+}

src/cli/status.rs

@@ -47,7 +47,7 @@ pub async fn run_status_command() -> anyhow::Result<()> {
     print!("  Secrets:     ");
     let secrets_configured = settings.secrets_master_key_source != crate::settings::KeySource::None
         || std::env::var("SECRETS_MASTER_KEY").is_ok()
-        || crate::secrets::keychain::has_master_key();
+        || crate::secrets::keychain::has_master_key().await;
     if secrets_configured {
         println!("configured ({:?})", settings.secrets_master_key_source);
     } else {

src/cli/tool.rs

@@ -715,7 +715,7 @@ async fn auth_tool(name: String, dir: Option<PathBuf>, user_id: String) -> anyho
     println!();
 
     // Initialize secrets store
-    let config = Config::from_env()?;
+    let config = Config::from_env().await?;
     let master_key = config.secrets.master_key().ok_or_else(|| {
         anyhow::anyhow!(
             "SECRETS_MASTER_KEY not set. Run 'ironclaw onboard' first or set it in .env"

src/config.rs

@@ -1,4 +1,9 @@
 //! Configuration for IronClaw.
+//!
+//! Settings are loaded with priority: env var > database > default.
+//! The database replaces the old `settings.json` file for all settings
+//! except the 4 bootstrap fields (database_url, pool_size, secrets key
+//! source, onboard_completed) which live in `~/.ironclaw/bootstrap.json`.
 
 use std::path::PathBuf;
 use std::time::Duration;
@@ -6,6 +11,7 @@ use std::time::Duration;
 use secrecy::{ExposeSecret, SecretString};
 
 use crate::error::ConfigError;
+use crate::settings::Settings;
 
 /// Main configuration for the agent.
 #[derive(Debug, Clone)]
@@ -21,28 +27,67 @@ pub struct Config {
     pub secrets: SecretsConfig,
     pub builder: BuilderModeConfig,
     pub heartbeat: HeartbeatConfig,
+    pub routines: RoutineConfig,
     pub sandbox: SandboxModeConfig,
+    pub claude_code: ClaudeCodeConfig,
 }
 
 impl Config {
-    /// Load configuration from environment variables.
-    pub fn from_env() -> Result<Self, ConfigError> {
-        // Load .env file if present (ignore errors if not found)
+    /// Load configuration from environment variables and the database.
+    ///
+    /// Priority: env var > DB settings > default.
+    /// This is the primary way to load config after DB is connected.
+    pub async fn from_db(
+        store: &crate::history::Store,
+        user_id: &str,
+        bootstrap: &crate::bootstrap::BootstrapConfig,
+    ) -> Result<Self, ConfigError> {
         let _ = dotenvy::dotenv();
 
+        // Load all settings from DB into a Settings struct
+        let db_settings = match store.get_all_settings(user_id).await {
+            Ok(map) => Settings::from_db_map(&map),
+            Err(e) => {
+                tracing::warn!("Failed to load settings from DB, using defaults: {}", e);
+                Settings::default()
+            }
+        };
+
+        Self::build(bootstrap, &db_settings).await
+    }
+
+    /// Load configuration from environment variables only (no database).
+    ///
+    /// Used during early startup before the database is connected,
+    /// and by CLI commands that don't have DB access.
+    /// Falls back to legacy `settings.json` on disk if present.
+    pub async fn from_env() -> Result<Self, ConfigError> {
+        let _ = dotenvy::dotenv();
+        let bootstrap = crate::bootstrap::BootstrapConfig::load();
+        let settings = Settings::load();
+        Self::build(&bootstrap, &settings).await
+    }
+
+    /// Build config from bootstrap + settings (shared by from_env and from_db).
+    async fn build(
+        bootstrap: &crate::bootstrap::BootstrapConfig,
+        settings: &Settings,
+    ) -> Result<Self, ConfigError> {
         Ok(Self {
-            database: DatabaseConfig::from_env()?,
-            llm: LlmConfig::from_env()?,
-            embeddings: EmbeddingsConfig::from_env()?,
-            tunnel: TunnelConfig::from_env()?,
-            channels: ChannelsConfig::from_env()?,
-            agent: AgentConfig::from_env()?,
-            safety: SafetyConfig::from_env()?,
-            wasm: WasmConfig::from_env()?,
-            secrets: SecretsConfig::from_env()?,
-            builder: BuilderModeConfig::from_env()?,
-            heartbeat: HeartbeatConfig::from_env()?,
-            sandbox: SandboxModeConfig::from_env()?,
+            database: DatabaseConfig::resolve(bootstrap)?,
+            llm: LlmConfig::resolve(settings)?,
+            embeddings: EmbeddingsConfig::resolve(settings)?,
+            tunnel: TunnelConfig::resolve(settings)?,
+            channels: ChannelsConfig::resolve(settings)?,
+            agent: AgentConfig::resolve(settings)?,
+            safety: SafetyConfig::resolve()?,
+            wasm: WasmConfig::resolve()?,
+            secrets: SecretsConfig::resolve(bootstrap).await?,
+            builder: BuilderModeConfig::resolve()?,
+            heartbeat: HeartbeatConfig::resolve(settings)?,
+            routines: RoutineConfig::resolve()?,
+            sandbox: SandboxModeConfig::resolve()?,
+            claude_code: ClaudeCodeConfig::resolve()?,
         })
     }
 }
@@ -51,48 +96,17 @@ impl Config {
 ///
 /// Used by channels and tools that need public webhook endpoints.
 /// The tunnel URL is shared across all channels (Telegram, Slack, etc.).
-///
-/// # Security Notes
-///
-/// **Webhook endpoints** (e.g., `/webhook/telegram`) should NOT use tunnel-level
-/// authentication because webhook providers (Telegram, Slack, GitHub) need
-/// unauthenticated access to POST updates. Security for webhooks comes from:
-/// - Webhook signature verification (provider-specific secrets)
-/// - IP allowlisting (if supported by provider)
-///
-/// **Non-webhook endpoints** (admin APIs, health checks) CAN be protected using
-/// tunnel provider features:
-/// - ngrok: Basic Auth, OAuth, IP restrictions
-/// - Cloudflare: Access policies, mTLS
-///
-/// These protections are configured in the tunnel provider, not here.
-///
-/// # Supported Providers
-///
-/// - **ngrok**: `ngrok http 8080` -> `https://abc123.ngrok.io`
-/// - **Cloudflare Tunnel**: `cloudflared tunnel --url http://localhost:8080`
-/// - **localtunnel**: `lt --port 8080`
-/// - Any service that provides a public HTTPS URL to localhost
 #[derive(Debug, Clone, Default)]
 pub struct TunnelConfig {
     /// Public URL from tunnel provider (e.g., "https://abc123.ngrok.io").
-    ///
-    /// When set, channels that support webhooks will register their endpoints
-    /// with this base URL instead of using polling.
     pub public_url: Option<String>,
 }
 
 impl TunnelConfig {
-    fn from_env() -> Result<Self, ConfigError> {
-        // Priority: env var > settings file
-        let public_url = optional_env("TUNNEL_URL")?.or_else(|| {
-            crate::settings::Settings::load()
-                .tunnel
-                .public_url
-                .filter(|s| !s.is_empty())
-        });
+    fn resolve(settings: &Settings) -> Result<Self, ConfigError> {
+        let public_url = optional_env("TUNNEL_URL")?
+            .or_else(|| settings.tunnel.public_url.clone().filter(|s| !s.is_empty()));
 
-        // Validate URL format if provided
         if let Some(ref url) = public_url {
             if !url.starts_with("https://") {
                 return Err(ConfigError::InvalidValue {
@@ -111,8 +125,6 @@ impl TunnelConfig {
     }
 
     /// Get the webhook URL for a given path.
-    ///
-    /// Returns `None` if no tunnel is configured.
     pub fn webhook_url(&self, path: &str) -> Option<String> {
         self.public_url.as_ref().map(|base| {
             let base = base.trim_end_matches('/');
@@ -130,18 +142,14 @@ pub struct DatabaseConfig {
 }
 
 impl DatabaseConfig {
-    fn from_env() -> Result<Self, ConfigError> {
-        let settings = crate::settings::Settings::load();
-
-        // Priority: env var > settings > error (required)
+    fn resolve(bootstrap: &crate::bootstrap::BootstrapConfig) -> Result<Self, ConfigError> {
         let url = optional_env("DATABASE_URL")?
-            .or(settings.database_url.clone())
+            .or_else(|| bootstrap.database_url.clone())
             .ok_or_else(|| ConfigError::MissingRequired {
                 key: "database_url".to_string(),
                 hint: "Run 'ironclaw onboard' or set DATABASE_URL environment variable".to_string(),
             })?;
 
-        // Priority: env var > settings > default
         let pool_size = optional_env("DATABASE_POOL_SIZE")?
             .map(|s| s.parse())
             .transpose()
@@ -149,7 +157,7 @@ impl DatabaseConfig {
                 key: "DATABASE_POOL_SIZE".to_string(),
                 message: format!("must be a positive integer: {e}"),
             })?
-            .or(settings.database_pool_size)
+            .or(bootstrap.database_pool_size)
             .unwrap_or(10);
 
         Ok(Self {
@@ -164,10 +172,102 @@ impl DatabaseConfig {
     }
 }
 
-/// LLM provider configuration (NEAR AI only).
+/// Which LLM backend to use.
+///
+/// Defaults to `NearAi` to keep IronClaw close to the NEAR ecosystem.
+/// Users can override with `LLM_BACKEND` env var to use their own API keys.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub enum LlmBackend {
+    /// NEAR AI proxy (default) -- session or API key auth
+    #[default]
+    NearAi,
+    /// Direct OpenAI API
+    OpenAi,
+    /// Direct Anthropic API
+    Anthropic,
+    /// Local Ollama instance
+    Ollama,
+    /// Any OpenAI-compatible endpoint (e.g. vLLM, LiteLLM, Together)
+    OpenAiCompatible,
+}
+
+impl std::str::FromStr for LlmBackend {
+    type Err = String;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        match s.to_lowercase().as_str() {
+            "nearai" | "near_ai" | "near" => Ok(Self::NearAi),
+            "openai" | "open_ai" => Ok(Self::OpenAi),
+            "anthropic" | "claude" => Ok(Self::Anthropic),
+            "ollama" => Ok(Self::Ollama),
+            "openai_compatible" | "openai-compatible" | "compatible" => Ok(Self::OpenAiCompatible),
+            _ => Err(format!(
+                "invalid LLM backend '{}', expected one of: nearai, openai, anthropic, ollama, openai_compatible",
+                s
+            )),
+        }
+    }
+}
+
+impl std::fmt::Display for LlmBackend {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::NearAi => write!(f, "nearai"),
+            Self::OpenAi => write!(f, "openai"),
+            Self::Anthropic => write!(f, "anthropic"),
+            Self::Ollama => write!(f, "ollama"),
+            Self::OpenAiCompatible => write!(f, "openai_compatible"),
+        }
+    }
+}
+
+/// Configuration for direct OpenAI API access.
+#[derive(Debug, Clone)]
+pub struct OpenAiDirectConfig {
+    pub api_key: SecretString,
+    pub model: String,
+}
+
+/// Configuration for direct Anthropic API access.
+#[derive(Debug, Clone)]
+pub struct AnthropicDirectConfig {
+    pub api_key: SecretString,
+    pub model: String,
+}
+
+/// Configuration for local Ollama.
+#[derive(Debug, Clone)]
+pub struct OllamaConfig {
+    pub base_url: String,
+    pub model: String,
+}
+
+/// Configuration for any OpenAI-compatible endpoint.
+#[derive(Debug, Clone)]
+pub struct OpenAiCompatibleConfig {
+    pub base_url: String,
+    pub api_key: Option<SecretString>,
+    pub model: String,
+}
+
+/// LLM provider configuration.
+///
+/// NEAR AI remains the default backend. Users can switch to other providers
+/// by setting `LLM_BACKEND` (e.g. `openai`, `anthropic`, `ollama`).
 #[derive(Debug, Clone)]
 pub struct LlmConfig {
+    /// Which backend to use (default: NearAi)
+    pub backend: LlmBackend,
+    /// NEAR AI config (always populated for NEAR AI embeddings, etc.)
     pub nearai: NearAiConfig,
+    /// Direct OpenAI config (populated when backend=openai)
+    pub openai: Option<OpenAiDirectConfig>,
+    /// Direct Anthropic config (populated when backend=anthropic)
+    pub anthropic: Option<AnthropicDirectConfig>,
+    /// Ollama config (populated when backend=ollama)
+    pub ollama: Option<OllamaConfig>,
+    /// OpenAI-compatible config (populated when backend=openai_compatible)
+    pub openai_compatible: Option<OpenAiCompatibleConfig>,
 }
 
 /// API mode for NEAR AI.
@@ -215,42 +315,110 @@ pub struct NearAiConfig {
 }
 
 impl LlmConfig {
-    fn from_env() -> Result<Self, ConfigError> {
-        let api_key = optional_env("NEARAI_API_KEY")?.map(SecretString::from);
+    fn resolve(settings: &Settings) -> Result<Self, ConfigError> {
+        // Determine backend (default: NearAi)
+        let backend: LlmBackend = if let Some(b) = optional_env("LLM_BACKEND")? {
+            b.parse().map_err(|e| ConfigError::InvalidValue {
+                key: "LLM_BACKEND".to_string(),
+                message: e,
+            })?
+        } else {
+            LlmBackend::NearAi
+        };
+
+        // Always resolve NEAR AI config (used as fallback and for embeddings)
+        let nearai_api_key = optional_env("NEARAI_API_KEY")?.map(SecretString::from);
 
-        // Determine API mode: explicit setting, or infer from API key presence
         let api_mode = if let Some(mode_str) = optional_env("NEARAI_API_MODE")? {
             mode_str.parse().map_err(|e| ConfigError::InvalidValue {
                 key: "NEARAI_API_MODE".to_string(),
                 message: e,
             })?
-        } else if api_key.is_some() {
-            // If API key is provided, default to chat_completions mode
+        } else if nearai_api_key.is_some() {
             NearAiApiMode::ChatCompletions
         } else {
             NearAiApiMode::Responses
         };
 
-        Ok(Self {
-            nearai: NearAiConfig {
-                // Load model from saved settings first, then env, then default
-                model: crate::settings::Settings::load()
-                    .selected_model
-                    .or_else(|| optional_env("NEARAI_MODEL").ok().flatten())
-                    .unwrap_or_else(|| {
-                        "fireworks::accounts/fireworks/models/llama4-maverick-instruct-basic"
-                            .to_string()
-                    }),
-                base_url: optional_env("NEARAI_BASE_URL")?
-                    .unwrap_or_else(|| "https://cloud-api.near.ai".to_string()),
-                auth_base_url: optional_env("NEARAI_AUTH_URL")?
-                    .unwrap_or_else(|| "https://private.near.ai".to_string()),
-                session_path: optional_env("NEARAI_SESSION_PATH")?
-                    .map(PathBuf::from)
-                    .unwrap_or_else(default_session_path),
-                api_mode,
+        let nearai = NearAiConfig {
+            model: optional_env("NEARAI_MODEL")?
+                .or_else(|| settings.selected_model.clone())
+                .unwrap_or_else(|| {
+                    "fireworks::accounts/fireworks/models/llama4-maverick-instruct-basic"
+                        .to_string()
+                }),
+            base_url: optional_env("NEARAI_BASE_URL")?
+                .unwrap_or_else(|| "https://cloud-api.near.ai".to_string()),
+            auth_base_url: optional_env("NEARAI_AUTH_URL")?
+                .unwrap_or_else(|| "https://private.near.ai".to_string()),
+            session_path: optional_env("NEARAI_SESSION_PATH")?
+                .map(PathBuf::from)
+                .unwrap_or_else(default_session_path),
+            api_mode,
+            api_key: nearai_api_key,
+        };
+
+        // Resolve provider-specific configs based on backend
+        let openai = if backend == LlmBackend::OpenAi {
+            let api_key = optional_env("OPENAI_API_KEY")?
+                .map(SecretString::from)
+                .ok_or_else(|| ConfigError::MissingRequired {
+                    key: "OPENAI_API_KEY".to_string(),
+                    hint: "Set OPENAI_API_KEY when LLM_BACKEND=openai".to_string(),
+                })?;
+            let model = optional_env("OPENAI_MODEL")?.unwrap_or_else(|| "gpt-4o".to_string());
+            Some(OpenAiDirectConfig { api_key, model })
+        } else {
+            None
+        };
+
+        let anthropic = if backend == LlmBackend::Anthropic {
+            let api_key = optional_env("ANTHROPIC_API_KEY")?
+                .map(SecretString::from)
+                .ok_or_else(|| ConfigError::MissingRequired {
+                    key: "ANTHROPIC_API_KEY".to_string(),
+                    hint: "Set ANTHROPIC_API_KEY when LLM_BACKEND=anthropic".to_string(),
+                })?;
+            let model = optional_env("ANTHROPIC_MODEL")?
+                .unwrap_or_else(|| "claude-sonnet-4-20250514".to_string());
+            Some(AnthropicDirectConfig { api_key, model })
+        } else {
+            None
+        };
+
+        let ollama = if backend == LlmBackend::Ollama {
+            let base_url = optional_env("OLLAMA_BASE_URL")?
+                .unwrap_or_else(|| "http://localhost:11434".to_string());
+            let model = optional_env("OLLAMA_MODEL")?.unwrap_or_else(|| "llama3".to_string());
+            Some(OllamaConfig { base_url, model })
+        } else {
+            None
+        };
+
+        let openai_compatible = if backend == LlmBackend::OpenAiCompatible {
+            let base_url =
+                optional_env("LLM_BASE_URL")?.ok_or_else(|| ConfigError::MissingRequired {
+                    key: "LLM_BASE_URL".to_string(),
+                    hint: "Set LLM_BASE_URL when LLM_BACKEND=openai_compatible".to_string(),
+                })?;
+            let api_key = optional_env("LLM_API_KEY")?.map(SecretString::from);
+            let model = optional_env("LLM_MODEL")?.unwrap_or_else(|| "default".to_string());
+            Some(OpenAiCompatibleConfig {
+                base_url,
                 api_key,
-            },
+                model,
+            })
+        } else {
+            None
+        };
+
+        Ok(Self {
+            backend,
+            nearai,
+            openai,
+            anthropic,
+            ollama,
+            openai_compatible,
         })
     }
 }
@@ -265,8 +433,6 @@ pub struct EmbeddingsConfig {
     /// OpenAI API key (for OpenAI provider).
     pub openai_api_key: Option<SecretString>,
     /// Model to use for embeddings.
-    /// For OpenAI: "text-embedding-3-small", "text-embedding-3-large", "text-embedding-ada-002"
-    /// For NEAR AI: Uses the configured session for auth.
     pub model: String,
 }
 
@@ -282,18 +448,15 @@ impl Default for EmbeddingsConfig {
 }
 
 impl EmbeddingsConfig {
-    fn from_env() -> Result<Self, ConfigError> {
-        let settings = crate::settings::Settings::load();
+    fn resolve(settings: &Settings) -> Result<Self, ConfigError> {
         let openai_api_key = optional_env("OPENAI_API_KEY")?.map(SecretString::from);
 
-        // Priority: env var > settings > default
         let provider = optional_env("EMBEDDING_PROVIDER")?
             .unwrap_or_else(|| settings.embeddings.provider.clone());
 
         let model =
             optional_env("EMBEDDING_MODEL")?.unwrap_or_else(|| settings.embeddings.model.clone());
 
-        // Priority: env var > settings > auto-detect from API key
         let enabled = optional_env("EMBEDDING_ENABLED")?
             .map(|s| s.parse())
             .transpose()
@@ -301,10 +464,7 @@ impl EmbeddingsConfig {
                 key: "EMBEDDING_ENABLED".to_string(),
                 message: format!("must be 'true' or 'false': {e}"),
             })?
-            .unwrap_or_else(|| {
-                // Check settings, or auto-enable if API key present
-                settings.embeddings.enabled || openai_api_key.is_some()
-            });
+            .unwrap_or_else(|| settings.embeddings.enabled || openai_api_key.is_some());
 
         Ok(Self {
             enabled,
@@ -338,6 +498,8 @@ pub struct ChannelsConfig {
     pub wasm_channels_dir: std::path::PathBuf,
     /// Whether WASM channels are enabled.
     pub wasm_channels_enabled: bool,
+    /// Telegram owner user ID. When set, the bot only responds to this user.
+    pub telegram_owner_id: Option<i64>,
 }
 
 #[derive(Debug, Clone)]
@@ -364,7 +526,7 @@ pub struct GatewayConfig {
 }
 
 impl ChannelsConfig {
-    fn from_env() -> Result<Self, ConfigError> {
+    fn resolve(settings: &Settings) -> Result<Self, ConfigError> {
         let http = if optional_env("HTTP_PORT")?.is_some() || optional_env("HTTP_HOST")?.is_some() {
             Some(HttpConfig {
                 host: optional_env("HTTP_HOST")?.unwrap_or_else(|| "0.0.0.0".to_string()),
@@ -385,7 +547,7 @@ impl ChannelsConfig {
 
         let gateway = if optional_env("GATEWAY_ENABLED")?
             .map(|s| s.to_lowercase() == "true" || s == "1")
-            .unwrap_or(false)
+            .unwrap_or(true)
         {
             Some(GatewayConfig {
                 host: optional_env("GATEWAY_HOST")?.unwrap_or_else(|| "127.0.0.1".to_string()),
@@ -425,6 +587,14 @@ impl ChannelsConfig {
                     message: format!("must be 'true' or 'false': {e}"),
                 })?
                 .unwrap_or(true),
+            telegram_owner_id: optional_env("TELEGRAM_OWNER_ID")?
+                .map(|s| s.parse())
+                .transpose()
+                .map_err(|e| ConfigError::InvalidValue {
+                    key: "TELEGRAM_OWNER_ID".to_string(),
+                    message: format!("must be an integer: {e}"),
+                })?
+                .or(settings.channels.telegram_owner_id),
         })
     }
 }
@@ -450,14 +620,13 @@ pub struct AgentConfig {
     pub use_planning: bool,
     /// Session idle timeout. Sessions inactive longer than this are pruned.
     pub session_idle_timeout: Duration,
+    /// Allow chat to use filesystem/shell tools directly (bypass sandbox).
+    pub allow_local_tools: bool,
 }
 
 impl AgentConfig {
-    fn from_env() -> Result<Self, ConfigError> {
-        let settings = crate::settings::Settings::load();
-
+    fn resolve(settings: &Settings) -> Result<Self, ConfigError> {
         Ok(Self {
-            // Priority: env var > settings > default
             name: optional_env("AGENT_NAME")?.unwrap_or_else(|| settings.agent.name.clone()),
             max_parallel_jobs: optional_env("AGENT_MAX_PARALLEL_JOBS")?
                 .map(|s| s.parse())
@@ -523,6 +692,14 @@ impl AgentConfig {
                     })?
                     .unwrap_or(settings.agent.session_idle_timeout_secs),
             ),
+            allow_local_tools: optional_env("ALLOW_LOCAL_TOOLS")?
+                .map(|s| s.parse())
+                .transpose()
+                .map_err(|e| ConfigError::InvalidValue {
+                    key: "ALLOW_LOCAL_TOOLS".to_string(),
+                    message: format!("must be 'true' or 'false': {e}"),
+                })?
+                .unwrap_or(false),
         })
     }
 }
@@ -535,7 +712,7 @@ pub struct SafetyConfig {
 }
 
 impl SafetyConfig {
-    fn from_env() -> Result<Self, ConfigError> {
+    fn resolve() -> Result<Self, ConfigError> {
         Ok(Self {
             max_output_length: parse_optional_env("SAFETY_MAX_OUTPUT_LENGTH", 100_000)?,
             injection_check_enabled: optional_env("SAFETY_INJECTION_CHECK_ENABLED")?
@@ -573,7 +750,6 @@ pub struct WasmConfig {
 #[derive(Clone, Default)]
 pub struct SecretsConfig {
     /// Master key for encrypting secrets.
-    /// Source determined by KeySource in settings.
     pub master_key: Option<SecretString>,
     /// Whether secrets management is enabled.
     pub enabled: bool,
@@ -592,20 +768,16 @@ impl std::fmt::Debug for SecretsConfig {
 }
 
 impl SecretsConfig {
-    fn from_env() -> Result<Self, ConfigError> {
+    async fn resolve(bootstrap: &crate::bootstrap::BootstrapConfig) -> Result<Self, ConfigError> {
         use crate::settings::KeySource;
 
-        let settings = crate::settings::Settings::load();
-
-        // Priority: env var > keychain (based on settings) > disabled
         let (master_key, source) = if let Some(env_key) = optional_env("SECRETS_MASTER_KEY")? {
-            // Env var takes priority (for CI/Docker)
             (Some(SecretString::from(env_key)), KeySource::Env)
         } else {
-            match settings.secrets_master_key_source {
+            match bootstrap.secrets_master_key_source {
                 KeySource::Keychain => {
-                    // Try to load from OS keychain
-                    match crate::secrets::keychain::get_master_key() {
+                    // Try to load from OS keychain (async on Linux)
+                    match crate::secrets::keychain::get_master_key().await {
                         Ok(key_bytes) => {
                             let key_hex: String =
                                 key_bytes.iter().map(|b| format!("{:02x}", b)).collect();
@@ -623,7 +795,6 @@ impl SecretsConfig {
                     }
                 }
                 KeySource::Env => {
-                    // Settings say env, but no env var found
                     tracing::warn!(
                         "Secrets configured for env var but SECRETS_MASTER_KEY not set."
                     );
@@ -635,7 +806,6 @@ impl SecretsConfig {
 
         let enabled = master_key.is_some();
 
-        // Validate master key length if provided
         if let Some(ref key) = master_key {
             if key.expose_secret().len() < 32 {
                 return Err(ConfigError::InvalidValue {
@@ -681,7 +851,7 @@ fn default_tools_dir() -> PathBuf {
 }
 
 impl WasmConfig {
-    fn from_env() -> Result<Self, ConfigError> {
+    fn resolve() -> Result<Self, ConfigError> {
         Ok(Self {
             enabled: optional_env("WASM_ENABLED")?
                 .map(|s| s.parse())
@@ -752,7 +922,7 @@ pub struct BuilderModeConfig {
 impl Default for BuilderModeConfig {
     fn default() -> Self {
         Self {
-            enabled: true, // Builder enabled by default
+            enabled: true,
             build_dir: None,
             max_iterations: 20,
             timeout_secs: 600,
@@ -762,7 +932,7 @@ impl Default for BuilderModeConfig {
 }
 
 impl BuilderModeConfig {
-    fn from_env() -> Result<Self, ConfigError> {
+    fn resolve() -> Result<Self, ConfigError> {
         Ok(Self {
             enabled: optional_env("BUILDER_ENABLED")?
                 .map(|s| s.parse())
@@ -771,7 +941,7 @@ impl BuilderModeConfig {
                     key: "BUILDER_ENABLED".to_string(),
                     message: format!("must be 'true' or 'false': {e}"),
                 })?
-                .unwrap_or(true), // Builder enabled by default
+                .unwrap_or(true),
             build_dir: optional_env("BUILDER_DIR")?.map(PathBuf::from),
             max_iterations: parse_optional_env("BUILDER_MAX_ITERATIONS", 20)?,
             timeout_secs: parse_optional_env("BUILDER_TIMEOUT_SECS", 600)?,
@@ -826,11 +996,8 @@ impl Default for HeartbeatConfig {
 }
 
 impl HeartbeatConfig {
-    fn from_env() -> Result<Self, ConfigError> {
-        let settings = crate::settings::Settings::load();
-
+    fn resolve(settings: &Settings) -> Result<Self, ConfigError> {
         Ok(Self {
-            // Priority: env var > settings > default
             enabled: optional_env("HEARTBEAT_ENABLED")?
                 .map(|s| s.parse())
                 .transpose()
@@ -848,9 +1015,55 @@ impl HeartbeatConfig {
                 })?
                 .unwrap_or(settings.heartbeat.interval_secs),
             notify_channel: optional_env("HEARTBEAT_NOTIFY_CHANNEL")?
-                .or(settings.heartbeat.notify_channel.clone()),
+                .or_else(|| settings.heartbeat.notify_channel.clone()),
             notify_user: optional_env("HEARTBEAT_NOTIFY_USER")?
-                .or(settings.heartbeat.notify_user.clone()),
+                .or_else(|| settings.heartbeat.notify_user.clone()),
+        })
+    }
+}
+
+/// Routines configuration.
+#[derive(Debug, Clone)]
+pub struct RoutineConfig {
+    /// Whether the routines system is enabled.
+    pub enabled: bool,
+    /// How often (seconds) to poll for cron routines that need firing.
+    pub cron_check_interval_secs: u64,
+    /// Max routines executing concurrently across all users.
+    pub max_concurrent_routines: usize,
+    /// Default cooldown between fires (seconds).
+    pub default_cooldown_secs: u64,
+    /// Max output tokens for lightweight routine LLM calls.
+    pub max_lightweight_tokens: u32,
+}
+
+impl Default for RoutineConfig {
+    fn default() -> Self {
+        Self {
+            enabled: true,
+            cron_check_interval_secs: 15,
+            max_concurrent_routines: 10,
+            default_cooldown_secs: 300,
+            max_lightweight_tokens: 4096,
+        }
+    }
+}
+
+impl RoutineConfig {
+    fn resolve() -> Result<Self, ConfigError> {
+        Ok(Self {
+            enabled: optional_env("ROUTINES_ENABLED")?
+                .map(|s| s.parse())
+                .transpose()
+                .map_err(|e| ConfigError::InvalidValue {
+                    key: "ROUTINES_ENABLED".to_string(),
+                    message: format!("must be 'true' or 'false': {e}"),
+                })?
+                .unwrap_or(true),
+            cron_check_interval_secs: parse_optional_env("ROUTINES_CRON_INTERVAL", 15)?,
+            max_concurrent_routines: parse_optional_env("ROUTINES_MAX_CONCURRENT", 10)?,
+            default_cooldown_secs: parse_optional_env("ROUTINES_DEFAULT_COOLDOWN", 300)?,
+            max_lightweight_tokens: parse_optional_env("ROUTINES_MAX_TOKENS", 4096)?,
         })
     }
 }
@@ -879,7 +1092,7 @@ pub struct SandboxModeConfig {
 impl Default for SandboxModeConfig {
     fn default() -> Self {
         Self {
-            enabled: true, // Enabled by default
+            enabled: true,
             policy: "readonly".to_string(),
             timeout_secs: 120,
             memory_limit_mb: 2048,
@@ -892,7 +1105,7 @@ impl Default for SandboxModeConfig {
 }
 
 impl SandboxModeConfig {
-    fn from_env() -> Result<Self, ConfigError> {
+    fn resolve() -> Result<Self, ConfigError> {
         let extra_domains = optional_env("SANDBOX_EXTRA_DOMAINS")?
             .map(|s| s.split(',').map(|d| d.trim().to_string()).collect())
             .unwrap_or_default();
@@ -948,6 +1161,60 @@ impl SandboxModeConfig {
     }
 }
 
+/// Claude Code sandbox configuration.
+#[derive(Debug, Clone)]
+pub struct ClaudeCodeConfig {
+    /// Whether Claude Code sandbox mode is available.
+    pub enabled: bool,
+    /// Host directory containing Claude auth session (mounted read-only).
+    pub config_dir: std::path::PathBuf,
+    /// Claude model to use (e.g. "sonnet", "opus").
+    pub model: String,
+    /// Maximum agentic turns before stopping.
+    pub max_turns: u32,
+    /// Memory limit in MB for Claude Code containers (heavier than workers).
+    pub memory_limit_mb: u64,
+}
+
+impl Default for ClaudeCodeConfig {
+    fn default() -> Self {
+        Self {
+            enabled: false,
+            config_dir: dirs::home_dir()
+                .unwrap_or_else(|| std::path::PathBuf::from("."))
+                .join(".claude"),
+            model: "sonnet".to_string(),
+            max_turns: 50,
+            memory_limit_mb: 4096,
+        }
+    }
+}
+
+impl ClaudeCodeConfig {
+    fn resolve() -> Result<Self, ConfigError> {
+        let defaults = Self::default();
+        Ok(Self {
+            enabled: optional_env("CLAUDE_CODE_ENABLED")?
+                .map(|s| s.parse())
+                .transpose()
+                .map_err(|e| ConfigError::InvalidValue {
+                    key: "CLAUDE_CODE_ENABLED".to_string(),
+                    message: format!("must be 'true' or 'false': {e}"),
+                })?
+                .unwrap_or(defaults.enabled),
+            config_dir: optional_env("CLAUDE_CONFIG_DIR")?
+                .map(std::path::PathBuf::from)
+                .unwrap_or(defaults.config_dir),
+            model: optional_env("CLAUDE_CODE_MODEL")?.unwrap_or(defaults.model),
+            max_turns: parse_optional_env("CLAUDE_CODE_MAX_TURNS", defaults.max_turns)?,
+            memory_limit_mb: parse_optional_env(
+                "CLAUDE_CODE_MEMORY_LIMIT_MB",
+                defaults.memory_limit_mb,
+            )?,
+        })
+    }
+}
+
 // Helper functions
 
 fn optional_env(key: &str) -> Result<Option<String>, ConfigError> {

src/error.rs

@@ -39,6 +39,12 @@ pub enum Error {
 
     #[error("Workspace error: {0}")]
     Workspace(#[from] WorkspaceError),
+
+    #[error("Orchestrator error: {0}")]
+    Orchestrator(#[from] OrchestratorError),
+
+    #[error("Worker error: {0}")]
+    Worker(#[from] WorkerError),
 }
 
 /// Configuration-related errors.
@@ -308,5 +314,52 @@ pub enum WorkspaceError {
     HeartbeatError { reason: String },
 }
 
+/// Orchestrator errors (internal API, container management).
+#[derive(Debug, thiserror::Error)]
+pub enum OrchestratorError {
+    #[error("Container creation failed for job {job_id}: {reason}")]
+    ContainerCreationFailed { job_id: Uuid, reason: String },
+
+    #[error("Container not found for job {job_id}")]
+    ContainerNotFound { job_id: Uuid },
+
+    #[error("Container for job {job_id} is in unexpected state: {state}")]
+    InvalidContainerState { job_id: Uuid, state: String },
+
+    #[error("Worker authentication failed: {reason}")]
+    AuthFailed { reason: String },
+
+    #[error("Internal API error: {reason}")]
+    ApiError { reason: String },
+
+    #[error("Docker error: {reason}")]
+    Docker { reason: String },
+
+    #[error("Job {job_id} timed out in container")]
+    ContainerTimeout { job_id: Uuid },
+}
+
+/// Worker errors (container-side execution).
+#[derive(Debug, thiserror::Error)]
+pub enum WorkerError {
+    #[error("Failed to connect to orchestrator at {url}: {reason}")]
+    ConnectionFailed { url: String, reason: String },
+
+    #[error("LLM proxy request failed: {reason}")]
+    LlmProxyFailed { reason: String },
+
+    #[error("Secret resolution failed for {secret_name}: {reason}")]
+    SecretResolveFailed { secret_name: String, reason: String },
+
+    #[error("Orchestrator returned error for job {job_id}: {reason}")]
+    OrchestratorRejected { job_id: Uuid, reason: String },
+
+    #[error("Worker execution failed: {reason}")]
+    ExecutionFailed { reason: String },
+
+    #[error("Missing worker token (IRONCLAW_WORKER_TOKEN not set)")]
+    MissingToken,
+}
+
 /// Result type alias for the agent.
 pub type Result<T> = std::result::Result<T, Error>;

src/extensions/manager.rs

@@ -23,9 +23,7 @@ use crate::tools::mcp::auth::{
     PkceChallenge, authorize_mcp_server, build_authorization_url, discover_full_oauth_metadata,
     find_available_port, is_authenticated, register_client,
 };
-use crate::tools::mcp::config::{
-    McpServerConfig, add_mcp_server, get_mcp_server, load_mcp_servers, remove_mcp_server,
-};
+use crate::tools::mcp::config::McpServerConfig;
 use crate::tools::mcp::session::McpSessionManager;
 use crate::tools::wasm::{WasmToolLoader, WasmToolRuntime, discover_tools};
 
@@ -58,6 +56,8 @@ pub struct ExtensionManager {
     /// Tunnel URL for remote OAuth callbacks (used in future iterations).
     _tunnel_url: Option<String>,
     user_id: String,
+    /// Optional database store for DB-backed MCP config.
+    store: Option<Arc<crate::history::Store>>,
 }
 
 impl ExtensionManager {
@@ -71,6 +71,7 @@ impl ExtensionManager {
         wasm_channels_dir: PathBuf,
         tunnel_url: Option<String>,
         user_id: String,
+        store: Option<Arc<crate::history::Store>>,
     ) -> Self {
         Self {
             registry: ExtensionRegistry::new(),
@@ -85,6 +86,7 @@ impl ExtensionManager {
             pending_auth: RwLock::new(HashMap::new()),
             _tunnel_url: tunnel_url,
             user_id,
+            store,
         }
     }
 
@@ -191,7 +193,7 @@ impl ExtensionManager {
 
         // List MCP servers
         if kind_filter.is_none() || kind_filter == Some(ExtensionKind::McpServer) {
-            match load_mcp_servers().await {
+            match self.load_mcp_servers().await {
                 Ok(servers) => {
                     for server in &servers.servers {
                         let authenticated =
@@ -304,7 +306,7 @@ impl ExtensionManager {
                 self.mcp_clients.write().await.remove(name);
 
                 // Remove from config
-                remove_mcp_server(name)
+                self.remove_mcp_server(name)
                     .await
                     .map_err(|e| ExtensionError::Config(e.to_string()))?;
 
@@ -342,6 +344,54 @@ impl ExtensionManager {
         }
     }
 
+    // ── MCP config helpers (DB with disk fallback) ─────────────────────
+
+    async fn load_mcp_servers(
+        &self,
+    ) -> Result<crate::tools::mcp::config::McpServersFile, crate::tools::mcp::config::ConfigError>
+    {
+        if let Some(ref store) = self.store {
+            crate::tools::mcp::config::load_mcp_servers_from_db(store, &self.user_id).await
+        } else {
+            crate::tools::mcp::config::load_mcp_servers().await
+        }
+    }
+
+    async fn get_mcp_server(
+        &self,
+        name: &str,
+    ) -> Result<McpServerConfig, crate::tools::mcp::config::ConfigError> {
+        let servers = self.load_mcp_servers().await?;
+        servers.get(name).cloned().ok_or_else(|| {
+            crate::tools::mcp::config::ConfigError::ServerNotFound {
+                name: name.to_string(),
+            }
+        })
+    }
+
+    async fn add_mcp_server(
+        &self,
+        config: McpServerConfig,
+    ) -> Result<(), crate::tools::mcp::config::ConfigError> {
+        config.validate()?;
+        if let Some(ref store) = self.store {
+            crate::tools::mcp::config::add_mcp_server_db(store, &self.user_id, config).await
+        } else {
+            crate::tools::mcp::config::add_mcp_server(config).await
+        }
+    }
+
+    async fn remove_mcp_server(
+        &self,
+        name: &str,
+    ) -> Result<(), crate::tools::mcp::config::ConfigError> {
+        if let Some(ref store) = self.store {
+            crate::tools::mcp::config::remove_mcp_server_db(store, &self.user_id, name).await
+        } else {
+            crate::tools::mcp::config::remove_mcp_server(name).await
+        }
+    }
+
     // ── Private helpers ──────────────────────────────────────────────────
 
     async fn install_from_entry(
@@ -381,7 +431,7 @@ impl ExtensionManager {
         url: &str,
     ) -> Result<InstallResult, ExtensionError> {
         // Check if already installed
-        if get_mcp_server(name).await.is_ok() {
+        if self.get_mcp_server(name).await.is_ok() {
             return Err(ExtensionError::AlreadyInstalled(name.to_string()));
         }
 
@@ -390,7 +440,7 @@ impl ExtensionManager {
             .validate()
             .map_err(|e| ExtensionError::InvalidUrl(e.to_string()))?;
 
-        add_mcp_server(config)
+        self.add_mcp_server(config)
             .await
             .map_err(|e| ExtensionError::Config(e.to_string()))?;
 
@@ -465,7 +515,8 @@ impl ExtensionManager {
         name: &str,
         token: Option<&str>,
     ) -> Result<AuthResult, ExtensionError> {
-        let server = get_mcp_server(name)
+        let server = self
+            .get_mcp_server(name)
             .await
             .map_err(|e| ExtensionError::NotInstalled(e.to_string()))?;
 
@@ -784,7 +835,8 @@ impl ExtensionManager {
             }
         }
 
-        let server = get_mcp_server(name)
+        let server = self
+            .get_mcp_server(name)
             .await
             .map_err(|e| ExtensionError::NotInstalled(e.to_string()))?;
 
@@ -893,7 +945,7 @@ impl ExtensionManager {
     /// Determine what kind of installed extension this is.
     async fn determine_installed_kind(&self, name: &str) -> Result<ExtensionKind, ExtensionError> {
         // Check MCP servers first
-        if get_mcp_server(name).await.is_ok() {
+        if self.get_mcp_server(name).await.is_ok() {
             return Ok(ExtensionKind::McpServer);
         }
 

src/history/mod.rs

@@ -9,4 +9,7 @@ mod analytics;
 mod store;
 
 pub use analytics::{JobStats, ToolStats};
-pub use store::{LlmCallRecord, Store};
+pub use store::{
+    ConversationMessage, ConversationSummary, JobEventRecord, LlmCallRecord, SandboxJobRecord,
+    SandboxJobSummary, Store,
+};

src/lib.rs

@@ -39,6 +39,7 @@
 //! - **Continuous learning** - Improve estimates from historical data
 
 pub mod agent;
+pub mod bootstrap;
 pub mod channels;
 pub mod cli;
 pub mod config;
@@ -49,12 +50,15 @@ pub mod evaluation;
 pub mod extensions;
 pub mod history;
 pub mod llm;
+pub mod orchestrator;
+pub mod pairing;
 pub mod safety;
 pub mod sandbox;
 pub mod secrets;
 pub mod settings;
 pub mod setup;
 pub mod tools;
+pub mod worker;
 pub mod workspace;
 
 pub use config::Config;

src/llm/costs.rs

@@ -0,0 +1,124 @@
+//! Per-model cost lookup table for multi-provider LLM support.
+//!
+//! Returns (input_cost_per_token, output_cost_per_token) as Decimal pairs.
+//! Ollama and other local models return zero cost.
+
+use rust_decimal::Decimal;
+use rust_decimal_macros::dec;
+
+/// Look up known per-token costs for a model by its identifier.
+///
+/// Returns `Some((input_cost, output_cost))` for known models, `None` otherwise.
+pub fn model_cost(model_id: &str) -> Option<(Decimal, Decimal)> {
+    // Normalize: strip provider prefixes (e.g., "openai/gpt-4o" -> "gpt-4o")
+    let id = model_id
+        .rsplit_once('/')
+        .map(|(_, name)| name)
+        .unwrap_or(model_id);
+
+    match id {
+        // OpenAI models -- prices per token (USD)
+        "gpt-4o" | "gpt-4o-2024-11-20" | "gpt-4o-2024-08-06" => {
+            Some((dec!(0.0000025), dec!(0.00001)))
+        }
+        "gpt-4o-mini" | "gpt-4o-mini-2024-07-18" => Some((dec!(0.00000015), dec!(0.0000006))),
+        "gpt-4-turbo" | "gpt-4-turbo-2024-04-09" => Some((dec!(0.00001), dec!(0.00003))),
+        "gpt-4" | "gpt-4-0613" => Some((dec!(0.00003), dec!(0.00006))),
+        "gpt-3.5-turbo" | "gpt-3.5-turbo-0125" => Some((dec!(0.0000005), dec!(0.0000015))),
+        "o1" | "o1-2024-12-17" => Some((dec!(0.000015), dec!(0.00006))),
+        "o1-mini" | "o1-mini-2024-09-12" => Some((dec!(0.000003), dec!(0.000012))),
+        "o3-mini" | "o3-mini-2025-01-31" => Some((dec!(0.0000011), dec!(0.0000044))),
+
+        // Anthropic models
+        "claude-3-5-sonnet-20241022" | "claude-3-5-sonnet-latest" | "claude-sonnet-4-20250514" => {
+            Some((dec!(0.000003), dec!(0.000015)))
+        }
+        "claude-3-5-haiku-20241022" | "claude-3-5-haiku-latest" => {
+            Some((dec!(0.0000008), dec!(0.000004)))
+        }
+        "claude-3-opus-20240229" | "claude-3-opus-latest" | "claude-opus-4-20250514" => {
+            Some((dec!(0.000015), dec!(0.000075)))
+        }
+        "claude-3-haiku-20240307" => Some((dec!(0.00000025), dec!(0.00000125))),
+
+        // Ollama / local models -- free
+        _ if is_local_model(id) => Some((Decimal::ZERO, Decimal::ZERO)),
+
+        _ => None,
+    }
+}
+
+/// Default cost for unknown models.
+pub fn default_cost() -> (Decimal, Decimal) {
+    // Conservative estimate: roughly GPT-4o pricing
+    (dec!(0.0000025), dec!(0.00001))
+}
+
+/// Heuristic to detect local/self-hosted models (Ollama, llama.cpp, etc.).
+fn is_local_model(model_id: &str) -> bool {
+    let lower = model_id.to_lowercase();
+    lower.starts_with("llama")
+        || lower.starts_with("mistral")
+        || lower.starts_with("mixtral")
+        || lower.starts_with("phi")
+        || lower.starts_with("gemma")
+        || lower.starts_with("qwen")
+        || lower.starts_with("codellama")
+        || lower.starts_with("deepseek")
+        || lower.starts_with("starcoder")
+        || lower.starts_with("vicuna")
+        || lower.starts_with("yi")
+        || lower.contains(":latest")
+        || lower.contains(":instruct")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_known_model_costs() {
+        let (input, output) = model_cost("gpt-4o").unwrap();
+        assert!(input > Decimal::ZERO);
+        assert!(output > input);
+    }
+
+    #[test]
+    fn test_claude_costs() {
+        let (input, output) = model_cost("claude-3-5-sonnet-20241022").unwrap();
+        assert!(input > Decimal::ZERO);
+        assert!(output > input);
+    }
+
+    #[test]
+    fn test_local_model_free() {
+        let (input, output) = model_cost("llama3").unwrap();
+        assert_eq!(input, Decimal::ZERO);
+        assert_eq!(output, Decimal::ZERO);
+    }
+
+    #[test]
+    fn test_ollama_tagged_model_free() {
+        let (input, output) = model_cost("mistral:latest").unwrap();
+        assert_eq!(input, Decimal::ZERO);
+        assert_eq!(output, Decimal::ZERO);
+    }
+
+    #[test]
+    fn test_unknown_model_returns_none() {
+        assert!(model_cost("some-totally-unknown-model-xyz").is_none());
+    }
+
+    #[test]
+    fn test_default_cost_nonzero() {
+        let (input, output) = default_cost();
+        assert!(input > Decimal::ZERO);
+        assert!(output > Decimal::ZERO);
+    }
+
+    #[test]
+    fn test_provider_prefix_stripped() {
+        // "openai/gpt-4o" should resolve to same as "gpt-4o"
+        assert_eq!(model_cost("openai/gpt-4o"), model_cost("gpt-4o"));
+    }
+}

src/llm/mod.rs

@@ -1,48 +1,172 @@
 //! LLM integration for the agent.
 //!
-//! Supports two API modes:
-//! - **Responses API** (chat-api): Session-based auth, uses `/v1/responses` endpoint
-//! - **Chat Completions API** (cloud-api): API key auth, uses `/v1/chat/completions` endpoint
+//! Supports multiple backends:
+//! - **NEAR AI** (default): Session-based or API key auth via NEAR AI proxy
+//! - **OpenAI**: Direct API access with your own key
+//! - **Anthropic**: Direct API access with your own key
+//! - **Ollama**: Local model inference
+//! - **OpenAI-compatible**: Any endpoint that speaks the OpenAI API
 
+mod costs;
 mod nearai;
 mod nearai_chat;
 mod provider;
 mod reasoning;
+mod rig_adapter;
 pub mod session;
 
 pub use nearai::{ModelInfo, NearAiProvider};
 pub use nearai_chat::NearAiChatProvider;
 pub use provider::{
-    ChatMessage, CompletionRequest, CompletionResponse, LlmProvider, Role, ToolCall,
-    ToolCompletionRequest, ToolCompletionResponse, ToolDefinition, ToolResult,
+    ChatMessage, CompletionRequest, CompletionResponse, FinishReason, LlmProvider, ModelMetadata,
+    Role, ToolCall, ToolCompletionRequest, ToolCompletionResponse, ToolDefinition, ToolResult,
 };
 pub use reasoning::{ActionPlan, Reasoning, ReasoningContext, RespondResult, ToolSelection};
+pub use rig_adapter::RigAdapter;
 pub use session::{SessionConfig, SessionManager, create_session_manager};
 
 use std::sync::Arc;
 
-use crate::config::{LlmConfig, NearAiApiMode};
+use rig::client::CompletionClient;
+use secrecy::ExposeSecret;
+
+use crate::config::{LlmBackend, LlmConfig, NearAiApiMode};
 use crate::error::LlmError;
 
 /// Create an LLM provider based on configuration.
 ///
-/// - For `Responses` mode: Requires a session manager for authentication
-/// - For `ChatCompletions` mode: Uses API key from config (session not needed)
+/// - `NearAi` backend: Uses session manager for authentication (Responses API)
+///   or API key (Chat Completions API)
+/// - Other backends: Use rig-core adapter with provider-specific clients
 pub fn create_llm_provider(
     config: &LlmConfig,
     session: Arc<SessionManager>,
+) -> Result<Arc<dyn LlmProvider>, LlmError> {
+    match config.backend {
+        LlmBackend::NearAi => create_nearai_provider(config, session),
+        LlmBackend::OpenAi => create_openai_provider(config),
+        LlmBackend::Anthropic => create_anthropic_provider(config),
+        LlmBackend::Ollama => create_ollama_provider(config),
+        LlmBackend::OpenAiCompatible => create_openai_compatible_provider(config),
+    }
+}
+
+fn create_nearai_provider(
+    config: &LlmConfig,
+    session: Arc<SessionManager>,
 ) -> Result<Arc<dyn LlmProvider>, LlmError> {
     match config.nearai.api_mode {
         NearAiApiMode::Responses => {
-            tracing::info!("Using Responses API (chat-api) with session auth");
+            tracing::info!("Using NEAR AI Responses API (chat-api) with session auth");
             Ok(Arc::new(NearAiProvider::new(
                 config.nearai.clone(),
                 session,
             )))
         }
         NearAiApiMode::ChatCompletions => {
-            tracing::info!("Using Chat Completions API (cloud-api) with API key auth");
+            tracing::info!("Using NEAR AI Chat Completions API (cloud-api) with API key auth");
             Ok(Arc::new(NearAiChatProvider::new(config.nearai.clone())?))
         }
     }
 }
+
+fn create_openai_provider(config: &LlmConfig) -> Result<Arc<dyn LlmProvider>, LlmError> {
+    let oai = config.openai.as_ref().ok_or_else(|| LlmError::AuthFailed {
+        provider: "openai".to_string(),
+    })?;
+
+    use rig::providers::openai;
+
+    let client: openai::Client =
+        openai::Client::new(oai.api_key.expose_secret()).map_err(|e| LlmError::RequestFailed {
+            provider: "openai".to_string(),
+            reason: format!("Failed to create OpenAI client: {}", e),
+        })?;
+
+    let model = client.completion_model(&oai.model);
+    tracing::info!("Using OpenAI direct API (model: {})", oai.model);
+    Ok(Arc::new(RigAdapter::new(model, &oai.model)))
+}
+
+fn create_anthropic_provider(config: &LlmConfig) -> Result<Arc<dyn LlmProvider>, LlmError> {
+    let anth = config
+        .anthropic
+        .as_ref()
+        .ok_or_else(|| LlmError::AuthFailed {
+            provider: "anthropic".to_string(),
+        })?;
+
+    use rig::providers::anthropic;
+
+    let client: anthropic::Client =
+        anthropic::Client::new(anth.api_key.expose_secret()).map_err(|e| {
+            LlmError::RequestFailed {
+                provider: "anthropic".to_string(),
+                reason: format!("Failed to create Anthropic client: {}", e),
+            }
+        })?;
+
+    let model = client.completion_model(&anth.model);
+    tracing::info!("Using Anthropic direct API (model: {})", anth.model);
+    Ok(Arc::new(RigAdapter::new(model, &anth.model)))
+}
+
+fn create_ollama_provider(config: &LlmConfig) -> Result<Arc<dyn LlmProvider>, LlmError> {
+    let oll = config.ollama.as_ref().ok_or_else(|| LlmError::AuthFailed {
+        provider: "ollama".to_string(),
+    })?;
+
+    use rig::client::Nothing;
+    use rig::providers::ollama;
+
+    let client: ollama::Client = ollama::Client::builder()
+        .base_url(&oll.base_url)
+        .api_key(Nothing)
+        .build()
+        .map_err(|e| LlmError::RequestFailed {
+            provider: "ollama".to_string(),
+            reason: format!("Failed to create Ollama client: {}", e),
+        })?;
+
+    let model = client.completion_model(&oll.model);
+    tracing::info!(
+        "Using Ollama (base_url: {}, model: {})",
+        oll.base_url,
+        oll.model
+    );
+    Ok(Arc::new(RigAdapter::new(model, &oll.model)))
+}
+
+fn create_openai_compatible_provider(config: &LlmConfig) -> Result<Arc<dyn LlmProvider>, LlmError> {
+    let compat = config
+        .openai_compatible
+        .as_ref()
+        .ok_or_else(|| LlmError::AuthFailed {
+            provider: "openai_compatible".to_string(),
+        })?;
+
+    use rig::providers::openai;
+
+    let api_key = compat
+        .api_key
+        .as_ref()
+        .map(|k| k.expose_secret().to_string())
+        .unwrap_or_else(|| "no-key".to_string());
+
+    let client: openai::Client = openai::Client::builder()
+        .base_url(&compat.base_url)
+        .api_key(api_key)
+        .build()
+        .map_err(|e| LlmError::RequestFailed {
+            provider: "openai_compatible".to_string(),
+            reason: format!("Failed to create OpenAI-compatible client: {}", e),
+        })?;
+
+    let model = client.completion_model(&compat.model);
+    tracing::info!(
+        "Using OpenAI-compatible endpoint (base_url: {}, model: {})",
+        compat.base_url,
+        compat.model
+    );
+    Ok(Arc::new(RigAdapter::new(model, &compat.model)))
+}

src/llm/nearai.rs

@@ -3,6 +3,7 @@
 //! This provider uses the NEAR AI chat-api which provides a unified interface
 //! to multiple LLM models (OpenAI, Anthropic, etc.) with user authentication.
 
+use std::collections::HashMap;
 use std::sync::Arc;
 
 use async_trait::async_trait;
@@ -31,11 +32,23 @@ pub struct ModelInfo {
     pub provider: Option<String>,
 }
 
+/// Per-thread chaining state: the last response ID and how many input
+/// messages were included in that request. This lets subsequent calls send
+/// only the delta (new messages since last call).
+struct ChainState {
+    response_id: String,
+    input_count: usize,
+}
+
 /// NEAR AI Chat API provider.
 pub struct NearAiProvider {
     client: Client,
     config: NearAiConfig,
     session: Arc<SessionManager>,
+    active_model: std::sync::RwLock<String>,
+    /// Per-thread response ID chaining state.
+    /// Key is thread_id from request metadata.
+    response_chains: std::sync::RwLock<HashMap<String, ChainState>>,
 }
 
 impl NearAiProvider {
@@ -46,13 +59,64 @@ impl NearAiProvider {
             .build()
             .unwrap_or_else(|_| Client::new());
 
+        let active_model = std::sync::RwLock::new(config.model.clone());
         Self {
             client,
             config,
             session,
+            active_model,
+            response_chains: std::sync::RwLock::new(HashMap::new()),
         }
     }
 
+    /// Seed a response chain for a thread (e.g. when restoring from DB).
+    pub fn seed_response_id(&self, thread_id: &str, response_id: String) {
+        let mut chains = self
+            .response_chains
+            .write()
+            .expect("response_chains lock poisoned");
+        chains.insert(
+            thread_id.to_string(),
+            ChainState {
+                response_id,
+                input_count: 0,
+            },
+        );
+    }
+
+    /// Get the last response ID for a thread (for persistence).
+    pub fn get_response_id(&self, thread_id: &str) -> Option<String> {
+        let chains = self
+            .response_chains
+            .read()
+            .expect("response_chains lock poisoned");
+        chains.get(thread_id).map(|c| c.response_id.clone())
+    }
+
+    /// Store a response chain state after a successful call.
+    fn store_chain(&self, thread_id: &str, response_id: String, input_count: usize) {
+        let mut chains = self
+            .response_chains
+            .write()
+            .expect("response_chains lock poisoned");
+        chains.insert(
+            thread_id.to_string(),
+            ChainState {
+                response_id,
+                input_count,
+            },
+        );
+    }
+
+    /// Clear the chain for a thread (on error / fallback).
+    fn clear_chain(&self, thread_id: &str) {
+        let mut chains = self
+            .response_chains
+            .write()
+            .expect("response_chains lock poisoned");
+        chains.remove(thread_id);
+    }
+
     fn api_url(&self, path: &str) -> String {
         format!(
             "{}/v1/{}",
@@ -291,18 +355,34 @@ impl NearAiProvider {
     }
 }
 
-/// Split messages into system instructions and non-system input messages.
+/// Split messages into system instructions and non-system input items.
 /// The OpenAI Responses API expects system prompts in an `instructions` field,
 /// not as a message with role "system" in the input array.
-fn split_messages(messages: Vec<ChatMessage>) -> (Option<String>, Vec<NearAiMessage>) {
+///
+/// When `chaining` is true, tool result messages (role=tool) are converted to
+/// `NearAiInputItem::FunctionCallOutput` for the Responses API protocol.
+fn split_messages(
+    messages: Vec<ChatMessage>,
+    chaining: bool,
+) -> (Option<String>, Vec<NearAiInputItem>) {
     let mut instructions: Vec<String> = Vec::new();
-    let mut input: Vec<NearAiMessage> = Vec::new();
+    let mut input: Vec<NearAiInputItem> = Vec::new();
 
     for msg in messages {
         if msg.role == Role::System {
             instructions.push(msg.content);
+        } else if chaining && msg.role == Role::Tool {
+            if let Some(ref call_id) = msg.tool_call_id {
+                input.push(NearAiInputItem::FunctionCallOutput {
+                    item_type: "function_call_output".to_string(),
+                    call_id: call_id.clone(),
+                    output: msg.content,
+                });
+            } else {
+                input.push(NearAiInputItem::Message(msg.into()));
+            }
         } else {
-            input.push(msg.into());
+            input.push(NearAiInputItem::Message(msg.into()));
         }
     }
 
@@ -318,12 +398,14 @@ fn split_messages(messages: Vec<ChatMessage>) -> (Option<String>, Vec<NearAiMess
 #[async_trait]
 impl LlmProvider for NearAiProvider {
     async fn complete(&self, req: CompletionRequest) -> Result<CompletionResponse, LlmError> {
-        let (instructions, input) = split_messages(req.messages);
+        let thread_id = req.metadata.get("thread_id").cloned();
+        let (instructions, input) = split_messages(req.messages, false);
 
         let request = NearAiRequest {
-            model: self.config.model.clone(),
+            model: self.active_model_name(),
             instructions,
             input,
+            previous_response_id: None,
             temperature: req.temperature,
             max_output_tokens: req.max_tokens,
             stream: Some(false),
@@ -350,6 +432,7 @@ impl LlmProvider for NearAiProvider {
                         finish_reason: FinishReason::Stop,
                         input_tokens: usage.input_tokens,
                         output_tokens: usage.output_tokens,
+                        response_id: None,
                     });
                 }
 
@@ -367,6 +450,7 @@ impl LlmProvider for NearAiProvider {
                     finish_reason: FinishReason::Stop,
                     input_tokens: 0,
                     output_tokens: 0,
+                    response_id: None,
                 });
             }
             Err(e) => return Err(e),
@@ -423,11 +507,17 @@ impl LlmProvider for NearAiProvider {
             );
         }
 
+        // Store response ID for chaining
+        if let Some(ref tid) = thread_id {
+            self.store_chain(tid, response.id.clone(), 0);
+        }
+
         Ok(CompletionResponse {
             content: text,
             finish_reason: FinishReason::Stop,
             input_tokens: response.usage.input_tokens,
             output_tokens: response.usage.output_tokens,
+            response_id: Some(response.id),
         })
     }
 
@@ -435,7 +525,33 @@ impl LlmProvider for NearAiProvider {
         &self,
         req: ToolCompletionRequest,
     ) -> Result<ToolCompletionResponse, LlmError> {
-        let (instructions, input) = split_messages(req.messages);
+        let thread_id = req.metadata.get("thread_id").cloned();
+
+        // Look up chaining state for this thread
+        let chain_state = thread_id.as_ref().and_then(|tid| {
+            let chains = self
+                .response_chains
+                .read()
+                .expect("response_chains lock poisoned");
+            chains
+                .get(tid)
+                .map(|c| (c.response_id.clone(), c.input_count))
+        });
+
+        let chaining = chain_state.is_some();
+        let (previous_response_id, prev_input_count) = chain_state
+            .map(|(rid, count)| (Some(rid), count))
+            .unwrap_or((None, 0));
+
+        // When chaining, only send new messages (the delta since last call).
+        // Tool results are converted to function_call_output items.
+        let (instructions, all_input) = split_messages(req.messages, chaining);
+        let input = if chaining && all_input.len() > prev_input_count {
+            all_input[prev_input_count..].to_vec()
+        } else {
+            all_input.clone()
+        };
+        let total_input_count = all_input.len();
 
         let tools: Vec<NearAiTool> = req
             .tools
@@ -449,18 +565,58 @@ impl LlmProvider for NearAiProvider {
             .collect();
 
         let request = NearAiRequest {
-            model: self.config.model.clone(),
-            instructions,
+            model: self.active_model_name(),
+            instructions: if chaining { None } else { instructions.clone() },
             input,
+            previous_response_id: previous_response_id.clone(),
             temperature: req.temperature,
             max_output_tokens: req.max_tokens,
             stream: Some(false),
-            tools: if tools.is_empty() { None } else { Some(tools) },
+            tools: if tools.is_empty() {
+                None
+            } else {
+                Some(tools.clone())
+            },
         };
 
-        // Try to get structured response, fall back to alternative formats
+        // Try to get structured response, fall back to alternative formats.
+        // If chaining fails (bad previous_response_id), retry with full history.
         let response: NearAiResponse = match self.send_request("responses", &request).await {
             Ok(r) => r,
+            Err(ref e) if chaining && is_chain_error(e) => {
+                tracing::warn!(
+                    "Response chaining failed, retrying with full history: {}",
+                    e
+                );
+                if let Some(ref tid) = thread_id {
+                    self.clear_chain(tid);
+                }
+                let (instructions_full, input_full) = split_messages(
+                    // Rebuild from the original input (non-chaining mode)
+                    {
+                        let mut msgs = Vec::new();
+                        if let Some(ref instr) = instructions {
+                            msgs.push(ChatMessage::system(instr.clone()));
+                        }
+                        for item in &all_input {
+                            msgs.push(item.to_chat_message());
+                        }
+                        msgs
+                    },
+                    false,
+                );
+                let retry_request = NearAiRequest {
+                    model: self.active_model_name(),
+                    instructions: instructions_full,
+                    input: input_full,
+                    previous_response_id: None,
+                    temperature: request.temperature,
+                    max_output_tokens: request.max_output_tokens,
+                    stream: Some(false),
+                    tools: request.tools.clone(),
+                };
+                self.send_request("responses", &retry_request).await?
+            }
             Err(LlmError::InvalidResponse { reason, .. }) if reason.contains("Raw: ") => {
                 let raw_text = reason.split("Raw: ").nth(1).unwrap_or("");
 
@@ -490,6 +646,7 @@ impl LlmProvider for NearAiProvider {
                         finish_reason,
                         input_tokens: usage.input_tokens,
                         output_tokens: usage.output_tokens,
+                        response_id: None,
                     });
                 }
 
@@ -507,6 +664,7 @@ impl LlmProvider for NearAiProvider {
                     finish_reason: FinishReason::Stop,
                     input_tokens: 0,
                     output_tokens: 0,
+                    response_id: None,
                 });
             }
             Err(e) => return Err(e),
@@ -560,12 +718,18 @@ impl LlmProvider for NearAiProvider {
             FinishReason::ToolUse
         };
 
+        // Store response ID for chaining on subsequent calls
+        if let Some(ref tid) = thread_id {
+            self.store_chain(tid, response.id.clone(), total_input_count);
+        }
+
         Ok(ToolCompletionResponse {
             content: if text.is_empty() { None } else { Some(text) },
             tool_calls,
             finish_reason,
             input_tokens: response.usage.input_tokens,
             output_tokens: response.usage.output_tokens,
+            response_id: Some(response.id),
         })
     }
 
@@ -584,6 +748,30 @@ impl LlmProvider for NearAiProvider {
         let models = NearAiProvider::list_models(self).await?;
         Ok(models.into_iter().map(|m| m.name).collect())
     }
+
+    fn active_model_name(&self) -> String {
+        self.active_model
+            .read()
+            .expect("active_model lock poisoned")
+            .clone()
+    }
+
+    fn set_model(&self, model: &str) -> Result<(), LlmError> {
+        let mut guard = self
+            .active_model
+            .write()
+            .expect("active_model lock poisoned");
+        *guard = model.to_string();
+        Ok(())
+    }
+
+    fn seed_response_chain(&self, thread_id: &str, response_id: String) {
+        self.seed_response_id(thread_id, response_id);
+    }
+
+    fn get_response_chain_id(&self, thread_id: &str) -> Option<String> {
+        self.get_response_id(thread_id)
+    }
 }
 
 // NEAR AI API types
@@ -597,8 +785,11 @@ struct NearAiRequest {
     /// System instructions (replaces sending system role in input)
     #[serde(skip_serializing_if = "Option::is_none")]
     instructions: Option<String>,
-    /// Input messages (user/assistant/tool only, NOT system)
-    input: Vec<NearAiMessage>,
+    /// Input items: messages and/or function_call_output entries.
+    input: Vec<NearAiInputItem>,
+    /// Chain this request to a previous response (avoids resending full context).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    previous_response_id: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
     temperature: Option<f32>,
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -609,7 +800,7 @@ struct NearAiRequest {
     tools: Option<Vec<NearAiTool>>,
 }
 
-#[derive(Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize, Clone)]
 struct NearAiMessage {
     role: String,
     content: String,
@@ -630,7 +821,68 @@ impl From<ChatMessage> for NearAiMessage {
     }
 }
 
-#[derive(Debug, Serialize)]
+/// Input item for the Responses API. Either a regular message or a
+/// function_call_output (for returning tool results when chaining).
+#[derive(Debug, Serialize, Deserialize, Clone)]
+#[serde(untagged)]
+enum NearAiInputItem {
+    Message(NearAiMessage),
+    FunctionCallOutput {
+        #[serde(rename = "type")]
+        item_type: String,
+        call_id: String,
+        output: String,
+    },
+}
+
+impl NearAiInputItem {
+    /// Convert back to a ChatMessage (used for fallback retry).
+    fn to_chat_message(&self) -> ChatMessage {
+        match self {
+            NearAiInputItem::Message(msg) => {
+                let role = match msg.role.as_str() {
+                    "system" => Role::System,
+                    "user" => Role::User,
+                    "assistant" => Role::Assistant,
+                    "tool" => Role::Tool,
+                    _ => Role::User,
+                };
+                ChatMessage {
+                    role,
+                    content: msg.content.clone(),
+                    tool_call_id: None,
+                    name: None,
+                    tool_calls: None,
+                }
+            }
+            NearAiInputItem::FunctionCallOutput {
+                call_id, output, ..
+            } => ChatMessage {
+                role: Role::Tool,
+                content: output.clone(),
+                tool_call_id: Some(call_id.clone()),
+                name: None,
+                tool_calls: None,
+            },
+        }
+    }
+}
+
+/// Check if an LLM error is likely caused by an invalid previous_response_id.
+fn is_chain_error(err: &LlmError) -> bool {
+    match err {
+        LlmError::RequestFailed { reason, .. } => {
+            let lower = reason.to_lowercase();
+            lower.contains("previous_response_id")
+                || lower.contains("previous response")
+                || lower.contains("not found")
+                || lower.contains("invalid response id")
+        }
+        _ => false,
+    }
+}
+
+#[derive(Debug, Clone, Serialize)]
 struct NearAiTool {
     #[serde(rename = "type")]
     tool_type: String,
@@ -833,14 +1085,17 @@ mod tests {
             ChatMessage::user("Hello"),
             ChatMessage::assistant("Hi there!"),
         ];
-        let (instructions, input) = split_messages(messages);
+        let (instructions, input) = split_messages(messages, false);
         assert_eq!(
             instructions,
             Some("You are a helpful assistant".to_string())
         );
         assert_eq!(input.len(), 2);
-        assert_eq!(input[0].role, "user");
-        assert_eq!(input[1].role, "assistant");
+        // Verify the input items are messages
+        match &input[0] {
+            NearAiInputItem::Message(m) => assert_eq!(m.role, "user"),
+            _ => panic!("expected Message"),
+        }
     }
 
     #[test]
@@ -849,7 +1104,7 @@ mod tests {
             ChatMessage::user("Hello"),
             ChatMessage::assistant("Hi there!"),
         ];
-        let (instructions, input) = split_messages(messages);
+        let (instructions, input) = split_messages(messages, false);
         assert!(instructions.is_none());
         assert_eq!(input.len(), 2);
     }
@@ -861,11 +1116,44 @@ mod tests {
             ChatMessage::system("Second instruction"),
             ChatMessage::user("Hello"),
         ];
-        let (instructions, input) = split_messages(messages);
+        let (instructions, input) = split_messages(messages, false);
         assert_eq!(
             instructions,
             Some("First instruction\n\nSecond instruction".to_string())
         );
         assert_eq!(input.len(), 1);
     }
+
+    #[test]
+    fn test_split_messages_chaining_converts_tool_results() {
+        let messages = vec![
+            ChatMessage::user("Hello"),
+            ChatMessage::tool_result("call_123", "my_tool", "result data"),
+        ];
+        let (_, input) = split_messages(messages, true);
+        assert_eq!(input.len(), 2);
+        match &input[1] {
+            NearAiInputItem::FunctionCallOutput {
+                call_id, output, ..
+            } => {
+                assert_eq!(call_id, "call_123");
+                assert_eq!(output, "result data");
+            }
+            _ => panic!("expected FunctionCallOutput"),
+        }
+    }
+
+    #[test]
+    fn test_split_messages_no_chaining_keeps_tool_as_message() {
+        let messages = vec![
+            ChatMessage::user("Hello"),
+            ChatMessage::tool_result("call_123", "my_tool", "result data"),
+        ];
+        let (_, input) = split_messages(messages, false);
+        assert_eq!(input.len(), 2);
+        match &input[1] {
+            NearAiInputItem::Message(m) => assert_eq!(m.role, "tool"),
+            _ => panic!("expected Message"),
+        }
+    }
 }

src/llm/nearai_chat.rs

@@ -13,14 +13,15 @@ use serde::{Deserialize, Serialize};
 use crate::config::NearAiConfig;
 use crate::error::LlmError;
 use crate::llm::provider::{
-    ChatMessage, CompletionRequest, CompletionResponse, FinishReason, LlmProvider, Role, ToolCall,
-    ToolCompletionRequest, ToolCompletionResponse,
+    ChatMessage, CompletionRequest, CompletionResponse, FinishReason, LlmProvider, ModelMetadata,
+    Role, ToolCall, ToolCompletionRequest, ToolCompletionResponse,
 };
 
 /// NEAR AI Chat Completions API provider.
 pub struct NearAiChatProvider {
     client: Client,
     config: NearAiConfig,
+    active_model: std::sync::RwLock<String>,
 }
 
 impl NearAiChatProvider {
@@ -37,7 +38,12 @@ impl NearAiChatProvider {
             .build()
             .unwrap_or_else(|_| Client::new());
 
-        Ok(Self { client, config })
+        let active_model = std::sync::RwLock::new(config.model.clone());
+        Ok(Self {
+            client,
+            config,
+            active_model,
+        })
     }
 
     fn api_url(&self, path: &str) -> String {
@@ -65,6 +71,11 @@ impl NearAiChatProvider {
 
         tracing::debug!("Sending request to NEAR AI Chat: {}", url);
 
+        // Log the request body for debugging tool call issues
+        if let Ok(json) = serde_json::to_string(body) {
+            tracing::debug!("NEAR AI Chat request body: {}", json);
+        }
+
         let response = self
             .client
             .post(&url)
@@ -111,8 +122,8 @@ impl NearAiChatProvider {
         })
     }
 
-    /// Fetch available models.
-    pub async fn list_models(&self) -> Result<Vec<String>, LlmError> {
+    /// Fetch available models with full metadata from the `/v1/models` endpoint.
+    async fn fetch_models(&self) -> Result<Vec<ApiModelEntry>, LlmError> {
         let url = self.api_url("models");
 
         let response = self
@@ -138,12 +149,7 @@ impl NearAiChatProvider {
 
         #[derive(Deserialize)]
         struct ModelsResponse {
-            data: Vec<ModelEntry>,
-        }
-
-        #[derive(Deserialize)]
-        struct ModelEntry {
-            id: String,
+            data: Vec<ApiModelEntry>,
         }
 
         let resp: ModelsResponse =
@@ -152,10 +158,18 @@ impl NearAiChatProvider {
                 reason: format!("JSON parse error: {}", e),
             })?;
 
-        Ok(resp.data.into_iter().map(|m| m.id).collect())
+        Ok(resp.data)
     }
 }
 
+/// Model entry as returned by the `/v1/models` API.
+#[derive(Debug, Deserialize)]
+struct ApiModelEntry {
+    id: String,
+    #[serde(default)]
+    context_length: Option<u32>,
+}
+
 #[async_trait]
 impl LlmProvider for NearAiChatProvider {
     async fn complete(&self, req: CompletionRequest) -> Result<CompletionResponse, LlmError> {
@@ -163,7 +177,7 @@ impl LlmProvider for NearAiChatProvider {
             req.messages.into_iter().map(|m| m.into()).collect();
 
         let request = ChatCompletionRequest {
-            model: self.config.model.clone(),
+            model: self.active_model_name(),
             messages,
             temperature: req.temperature,
             max_tokens: req.max_tokens,
@@ -197,6 +211,7 @@ impl LlmProvider for NearAiChatProvider {
             finish_reason,
             input_tokens: response.usage.prompt_tokens,
             output_tokens: response.usage.completion_tokens,
+            response_id: None,
         })
     }
 
@@ -221,7 +236,7 @@ impl LlmProvider for NearAiChatProvider {
             .collect();
 
         let request = ChatCompletionRequest {
-            model: self.config.model.clone(),
+            model: self.active_model_name(),
             messages,
             temperature: req.temperature,
             max_tokens: req.max_tokens,
@@ -278,6 +293,7 @@ impl LlmProvider for NearAiChatProvider {
             finish_reason,
             input_tokens: response.usage.prompt_tokens,
             output_tokens: response.usage.completion_tokens,
+            response_id: None,
         })
     }
 
@@ -291,7 +307,34 @@ impl LlmProvider for NearAiChatProvider {
     }
 
     async fn list_models(&self) -> Result<Vec<String>, LlmError> {
-        NearAiChatProvider::list_models(self).await
+        let models = self.fetch_models().await?;
+        Ok(models.into_iter().map(|m| m.id).collect())
+    }
+
+    async fn model_metadata(&self) -> Result<ModelMetadata, LlmError> {
+        let active = self.active_model_name();
+        let models = self.fetch_models().await?;
+        let current = models.iter().find(|m| m.id == active);
+        Ok(ModelMetadata {
+            id: active,
+            context_length: current.and_then(|m| m.context_length),
+        })
+    }
+
+    fn active_model_name(&self) -> String {
+        self.active_model
+            .read()
+            .expect("active_model lock poisoned")
+            .clone()
+    }
+
+    fn set_model(&self, model: &str) -> Result<(), crate::error::LlmError> {
+        let mut guard = self
+            .active_model
+            .write()
+            .expect("active_model lock poisoned");
+        *guard = model.to_string();
+        Ok(())
     }
 }
 
@@ -332,6 +375,7 @@ impl From<ChatMessage> for ChatCompletionMessage {
             Role::Assistant => "assistant",
             Role::Tool => "tool",
         };
+
         let tool_calls = msg.tool_calls.map(|calls| {
             calls
                 .into_iter()
@@ -345,9 +389,16 @@ impl From<ChatMessage> for ChatCompletionMessage {
                 })
                 .collect()
         });
+
+        let content = if role == "assistant" && tool_calls.is_some() && msg.content.is_empty() {
+            None
+        } else {
+            Some(msg.content)
+        };
+
         Self {
             role: role.to_string(),
-            content: Some(msg.content),
+            content,
             tool_call_id: msg.tool_call_id,
             name: msg.name,
             tool_calls,
@@ -454,7 +505,7 @@ mod tests {
             },
         ];
 
-        let msg = ChatMessage::assistant_with_tool_calls("", tool_calls);
+        let msg = ChatMessage::assistant_with_tool_calls(None, tool_calls);
         let chat_msg: ChatCompletionMessage = msg.into();
 
         assert_eq!(chat_msg.role, "assistant");
@@ -484,7 +535,7 @@ mod tests {
             name: "test".to_string(),
             arguments: serde_json::json!({"key": "value"}),
         };
-        let msg = ChatMessage::assistant_with_tool_calls("", vec![tc]);
+        let msg = ChatMessage::assistant_with_tool_calls(None, vec![tc]);
         let chat_msg: ChatCompletionMessage = msg.into();
 
         let calls = chat_msg.tool_calls.unwrap();

src/llm/provider.rs

@@ -27,9 +27,8 @@ pub struct ChatMessage {
     /// Name of the tool for tool results.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub name: Option<String>,
-    /// Tool calls requested by the assistant (for conversation replay).
-    /// OpenAI-compatible APIs require the assistant message to include
-    /// tool_calls when followed by tool result messages.
+    /// Tool calls made by the assistant (OpenAI protocol requires these
+    /// to appear on the assistant message preceding tool result messages).
     #[serde(skip_serializing_if = "Option::is_none")]
     pub tool_calls: Option<Vec<ToolCall>>,
 }
@@ -68,17 +67,14 @@ impl ChatMessage {
         }
     }
 
-    /// Create an assistant message that requested tool calls.
+    /// Create an assistant message that includes tool calls.
     ///
-    /// OpenAI-compatible APIs require the assistant message to carry the
-    /// `tool_calls` array when followed by tool-result messages.
-    pub fn assistant_with_tool_calls(
-        content: impl Into<String>,
-        tool_calls: Vec<ToolCall>,
-    ) -> Self {
+    /// Per the OpenAI protocol, an assistant message with tool_calls must
+    /// precede the corresponding tool result messages in the conversation.
+    pub fn assistant_with_tool_calls(content: Option<String>, tool_calls: Vec<ToolCall>) -> Self {
         Self {
             role: Role::Assistant,
-            content: content.into(),
+            content: content.unwrap_or_default(),
             tool_call_id: None,
             name: None,
             tool_calls: if tool_calls.is_empty() {
@@ -112,6 +108,8 @@ pub struct CompletionRequest {
     pub max_tokens: Option<u32>,
     pub temperature: Option<f32>,
     pub stop_sequences: Option<Vec<String>>,
+    /// Opaque metadata passed through to the provider (e.g. thread_id for chaining).
+    pub metadata: std::collections::HashMap<String, String>,
 }
 
 impl CompletionRequest {
@@ -122,6 +120,7 @@ impl CompletionRequest {
             max_tokens: None,
             temperature: None,
             stop_sequences: None,
+            metadata: std::collections::HashMap::new(),
         }
     }
 
@@ -145,6 +144,8 @@ pub struct CompletionResponse {
     pub input_tokens: u32,
     pub output_tokens: u32,
     pub finish_reason: FinishReason,
+    /// Provider-specific response ID (e.g. for NEAR AI response chaining).
+    pub response_id: Option<String>,
 }
 
 /// Why the completion finished.
@@ -191,6 +192,8 @@ pub struct ToolCompletionRequest {
     pub temperature: Option<f32>,
     /// How to handle tool use: "auto", "required", or "none".
     pub tool_choice: Option<String>,
+    /// Opaque metadata passed through to the provider (e.g. thread_id for chaining).
+    pub metadata: std::collections::HashMap<String, String>,
 }
 
 impl ToolCompletionRequest {
@@ -202,6 +205,7 @@ impl ToolCompletionRequest {
             max_tokens: None,
             temperature: None,
             tool_choice: None,
+            metadata: std::collections::HashMap::new(),
         }
     }
 
@@ -234,6 +238,16 @@ pub struct ToolCompletionResponse {
     pub input_tokens: u32,
     pub output_tokens: u32,
     pub finish_reason: FinishReason,
+    /// Provider-specific response ID (e.g. for NEAR AI response chaining).
+    pub response_id: Option<String>,
+}
+
+/// Metadata about a model returned by the provider's API.
+#[derive(Debug, Clone)]
+pub struct ModelMetadata {
+    pub id: String,
+    /// Total context window size in tokens.
+    pub context_length: Option<u32>,
 }
 
 /// Trait for LLM providers.
@@ -260,6 +274,45 @@ pub trait LlmProvider: Send + Sync {
         Ok(Vec::new())
     }
 
+    /// Fetch metadata for the current model (context length, etc.).
+    /// Default returns the model name with no size info.
+    async fn model_metadata(&self) -> Result<ModelMetadata, LlmError> {
+        Ok(ModelMetadata {
+            id: self.model_name().to_string(),
+            context_length: None,
+        })
+    }
+
+    /// Get the currently active model name.
+    ///
+    /// May differ from `model_name()` if the model was switched at runtime
+    /// via `set_model()`. Default returns `model_name()`.
+    fn active_model_name(&self) -> String {
+        self.model_name().to_string()
+    }
+
+    /// Switch the active model at runtime. Not all providers support this.
+    fn set_model(&self, _model: &str) -> Result<(), LlmError> {
+        Err(LlmError::RequestFailed {
+            provider: "unknown".to_string(),
+            reason: "Runtime model switching not supported by this provider".to_string(),
+        })
+    }
+
+    /// Seed a response chain for a thread (e.g. restoring from DB).
+    ///
+    /// Providers that support response chaining (e.g. NEAR AI `previous_response_id`)
+    /// store this so subsequent calls send only delta messages.
+    fn seed_response_chain(&self, _thread_id: &str, _response_id: String) {}
+
+    /// Get the last response chain ID for a thread.
+    ///
+    /// Returns `None` if the provider doesn't support chaining or has no
+    /// stored state for this thread.
+    fn get_response_chain_id(&self, _thread_id: &str) -> Option<String> {
+        None
+    }
+
     /// Calculate cost for a completion.
     fn calculate_cost(&self, input_tokens: u32, output_tokens: u32) -> Decimal {
         let (input_cost, output_cost) = self.cost_per_token();

src/llm/reasoning.rs

@@ -21,6 +21,8 @@ pub struct ReasoningContext {
     pub job_description: Option<String>,
     /// Current state description.
     pub current_state: Option<String>,
+    /// Opaque metadata forwarded to the LLM provider (e.g. thread_id for chaining).
+    pub metadata: std::collections::HashMap<String, String>,
 }
 
 impl ReasoningContext {
@@ -31,6 +33,7 @@ impl ReasoningContext {
             available_tools: Vec::new(),
             job_description: None,
             current_state: None,
+            metadata: std::collections::HashMap::new(),
         }
     }
 
@@ -57,6 +60,12 @@ impl ReasoningContext {
         self.job_description = Some(description.into());
         self
     }
+
+    /// Set metadata (forwarded to the LLM provider).
+    pub fn with_metadata(mut self, metadata: std::collections::HashMap<String, String>) -> Self {
+        self.metadata = metadata;
+        self
+    }
 }
 
 impl Default for ReasoningContext {
@@ -114,7 +123,12 @@ pub enum RespondResult {
     /// A text response (no tools needed).
     Text(String),
     /// The model wants to call tools. Caller should execute them and call back.
-    ToolCalls(Vec<ToolCall>),
+    /// Includes the optional content from the assistant message (some models
+    /// include explanatory text alongside tool calls).
+    ToolCalls {
+        tool_calls: Vec<ToolCall>,
+        content: Option<String>,
+    },
 }
 
 /// Reasoning engine for the agent.
@@ -192,10 +206,11 @@ impl Reasoning {
             return Ok(vec![]);
         }
 
-        let request =
+        let mut request =
             ToolCompletionRequest::new(context.messages.clone(), context.available_tools.clone())
                 .with_max_tokens(1024)
                 .with_tool_choice("auto");
+        request.metadata = context.metadata.clone();
 
         let response = self.llm.complete_with_tools(request).await?;
 
@@ -271,7 +286,9 @@ Respond in JSON format:
     pub async fn respond(&self, context: &ReasoningContext) -> Result<String, LlmError> {
         match self.respond_with_tools(context).await? {
             RespondResult::Text(text) => Ok(text),
-            RespondResult::ToolCalls(calls) => {
+            RespondResult::ToolCalls {
+                tool_calls: calls, ..
+            } => {
                 // Format tool calls as text (legacy behavior for non-agentic callers)
                 let tool_info: Vec<String> = calls
                     .iter()
@@ -298,28 +315,49 @@ Respond in JSON format:
 
         // If we have tools, use tool completion mode
         if !context.available_tools.is_empty() {
-            let request = ToolCompletionRequest::new(messages, context.available_tools.clone())
+            let mut request = ToolCompletionRequest::new(messages, context.available_tools.clone())
                 .with_max_tokens(4096)
                 .with_temperature(0.7)
                 .with_tool_choice("auto");
+            request.metadata = context.metadata.clone();
 
             let response = self.llm.complete_with_tools(request).await?;
 
             // If there were tool calls, return them for execution
             if !response.tool_calls.is_empty() {
-                return Ok(RespondResult::ToolCalls(response.tool_calls));
+                return Ok(RespondResult::ToolCalls {
+                    tool_calls: response.tool_calls,
+                    content: response.content,
+                });
             }
 
             let content = response
                 .content
                 .unwrap_or_else(|| "I'm not sure how to respond to that.".to_string());
 
+            // Some models (e.g. GLM-4.7) emit tool calls as XML tags in content
+            // instead of using the structured tool_calls field. Try to recover
+            // them before giving up and returning plain text.
+            let recovered = recover_tool_calls_from_content(&content, &context.available_tools);
+            if !recovered.is_empty() {
+                let cleaned = clean_response(&content);
+                return Ok(RespondResult::ToolCalls {
+                    tool_calls: recovered,
+                    content: if cleaned.is_empty() {
+                        None
+                    } else {
+                        Some(cleaned)
+                    },
+                });
+            }
+
             Ok(RespondResult::Text(clean_response(&content)))
         } else {
             // No tools, use simple completion
-            let request = CompletionRequest::new(messages)
+            let mut request = CompletionRequest::new(messages)
                 .with_max_tokens(4096)
                 .with_temperature(0.7);
+            request.metadata = context.metadata.clone();
 
             let response = self.llm.complete(request).await?;
             Ok(RespondResult::Text(clean_response(&response.content)))
@@ -462,47 +500,178 @@ fn extract_json(text: &str) -> Option<&str> {
     }
 }
 
-/// Clean up LLM response by stripping thinking tags and reasoning patterns.
+/// Clean up LLM response by stripping model-internal tags and reasoning patterns.
+///
+/// Some models (GLM-4.7, etc.) emit XML-tagged internal state like
+/// Try to extract tool calls from content text where the model emitted them
+/// as XML tags instead of using the structured tool_calls field.
+///
+/// Handles these formats:
+/// - `<tool_call>tool_name</tool_call>` (bare name)
+/// - `<tool_call>{"name":"x","arguments":{}}</tool_call>` (JSON)
+/// - `<|tool_call|>...<|/tool_call|>` (pipe-delimited variant)
+/// - `<function_call>...</function_call>` (function_call variant)
+///
+/// Only returns calls whose name matches an available tool.
+fn recover_tool_calls_from_content(
+    content: &str,
+    available_tools: &[ToolDefinition],
+) -> Vec<ToolCall> {
+    let tool_names: std::collections::HashSet<&str> =
+        available_tools.iter().map(|t| t.name.as_str()).collect();
+    let mut calls = Vec::new();
+
+    for (open, close) in &[
+        ("<tool_call>", "</tool_call>"),
+        ("<|tool_call|>", "<|/tool_call|>"),
+        ("<function_call>", "</function_call>"),
+        ("<|function_call|>", "<|/function_call|>"),
+    ] {
+        let mut remaining = content;
+        while let Some(start) = remaining.find(open) {
+            let inner_start = start + open.len();
+            let after = &remaining[inner_start..];
+            let Some(end) = after.find(close) else {
+                break;
+            };
+            let inner = after[..end].trim();
+            remaining = &after[end + close.len()..];
+
+            if inner.is_empty() {
+                continue;
+            }
+
+            // Try JSON first: {"name":"x","arguments":{}}
+            if let Ok(parsed) = serde_json::from_str::<serde_json::Value>(inner) {
+                if let Some(name) = parsed.get("name").and_then(|v| v.as_str()) {
+                    if tool_names.contains(name) {
+                        let arguments = parsed
+                            .get("arguments")
+                            .cloned()
+                            .unwrap_or(serde_json::Value::Object(Default::default()));
+                        calls.push(ToolCall {
+                            id: format!("recovered_{}", calls.len()),
+                            name: name.to_string(),
+                            arguments,
+                        });
+                        continue;
+                    }
+                }
+            }
+
+            // Bare tool name (e.g. "<tool_call>tool_list</tool_call>")
+            let name = inner.trim();
+            if tool_names.contains(name) {
+                calls.push(ToolCall {
+                    id: format!("recovered_{}", calls.len()),
+                    name: name.to_string(),
+                    arguments: serde_json::Value::Object(Default::default()),
+                });
+            }
+        }
+    }
+
+    calls
+}
+
+/// `<tool_call>tool_list</tool_call>` or `<|tool_call|>` in the content field
+/// instead of using the standard OpenAI tool_calls array. We strip all of
+/// these before the response reaches channels/users.
 fn clean_response(text: &str) -> String {
-    let text = strip_thinking_tags(text);
+    let text = strip_internal_tags(text);
     strip_reasoning_patterns(&text)
 }
 
-/// Strip `<thinking>...</thinking>` blocks from LLM output.
+/// Tags that are model-internal and should never reach users.
+const INTERNAL_TAGS: &[&str] = &["thinking", "tool_call", "function_call", "tool_calls"];
+
+/// Strip all model-internal XML tags from LLM output.
 ///
-/// Some models (especially Claude with extended thinking) include internal
-/// reasoning in thinking tags. We strip these before showing to users.
-fn strip_thinking_tags(text: &str) -> String {
+/// Handles standard XML tags (`<tag>...</tag>`) and pipe-delimited variants
+/// (`<|tag|>...<|/tag|>`) used by some models (e.g. GLM-4.7).
+fn strip_internal_tags(text: &str) -> String {
+    let mut result = text.to_string();
+    for tag in INTERNAL_TAGS {
+        result = strip_xml_tag(&result, tag);
+        result = strip_pipe_tag(&result, tag);
+    }
+    // Collapse triple+ newlines left behind by removed blocks
+    while result.contains("\n\n\n") {
+        result = result.replace("\n\n\n", "\n\n");
+    }
+    result.trim().to_string()
+}
+
+/// Strip `<tag>...</tag>` and `<tag ...>...</tag>` blocks from text.
+fn strip_xml_tag(text: &str, tag: &str) -> String {
+    let open_exact = format!("<{}>", tag);
+    let open_prefix = format!("<{} ", tag); // for <tag attr="...">
+    let close = format!("</{}>", tag);
+
     let mut result = String::with_capacity(text.len());
     let mut remaining = text;
 
-    while let Some(start) = remaining.find("<thinking>") {
+    loop {
+        // Find the next opening tag (exact or with attributes)
+        let exact_pos = remaining.find(&open_exact);
+        let prefix_pos = remaining.find(&open_prefix);
+        let start = match (exact_pos, prefix_pos) {
+            (Some(a), Some(b)) => a.min(b),
+            (Some(a), None) => a,
+            (None, Some(b)) => b,
+            (None, None) => break,
+        };
+
         // Add everything before the tag
         result.push_str(&remaining[..start]);
 
+        // Find the end of the opening tag (the closing >)
+        let after_open = &remaining[start..];
+        let open_end = match after_open.find('>') {
+            Some(pos) => start + pos + 1,
+            None => break, // malformed, stop
+        };
+
         // Find the closing tag
-        if let Some(end_offset) = remaining[start..].find("</thinking>") {
-            // Skip past the closing tag (start + offset + tag length)
-            let end = start + end_offset + "</thinking>".len();
+        if let Some(close_offset) = remaining[open_end..].find(&close) {
+            let end = open_end + close_offset + close.len();
             remaining = &remaining[end..];
         } else {
-            // No closing tag found, discard everything from here
-            // (malformed, but handle gracefully by not including the unclosed tag)
+            // No closing tag, discard from here (malformed)
             remaining = "";
             break;
         }
     }
 
-    // Add any remaining content after the last thinking block
     result.push_str(remaining);
+    result
+}
 
-    // Clean up any double newlines left behind
-    let mut cleaned = result.trim().to_string();
-    while cleaned.contains("\n\n\n") {
-        cleaned = cleaned.replace("\n\n\n", "\n\n");
+/// Strip `<|tag|>...<|/tag|>` pipe-delimited blocks from text.
+///
+/// Some models (e.g. certain Chinese LLMs) use this format instead of
+/// standard XML tags.
+fn strip_pipe_tag(text: &str, tag: &str) -> String {
+    let open = format!("<|{}|>", tag);
+    let close = format!("<|/{}|>", tag);
+
+    let mut result = String::with_capacity(text.len());
+    let mut remaining = text;
+
+    while let Some(start) = remaining.find(&open) {
+        result.push_str(&remaining[..start]);
+
+        if let Some(close_offset) = remaining[start..].find(&close) {
+            let end = start + close_offset + close.len();
+            remaining = &remaining[end..];
+        } else {
+            remaining = "";
+            break;
+        }
     }
 
-    cleaned
+    result.push_str(remaining);
+    result
 }
 
 /// Strip any remaining reasoning that wasn't in proper <thinking> tags.
@@ -574,7 +743,7 @@ That's my plan."#;
     #[test]
     fn test_strip_thinking_tags_basic() {
         let input = "<thinking>Let me think about this...</thinking>Hello, user!";
-        let output = strip_thinking_tags(input);
+        let output = strip_internal_tags(input);
         assert_eq!(output, "Hello, user!");
     }
 
@@ -582,7 +751,7 @@ That's my plan."#;
     fn test_strip_thinking_tags_multiple() {
         let input =
             "<thinking>First thought</thinking>Hello<thinking>Second thought</thinking> world!";
-        let output = strip_thinking_tags(input);
+        let output = strip_internal_tags(input);
         assert_eq!(output, "Hello world!");
     }
 
@@ -594,14 +763,14 @@ I need to consider:
 2. How to respond
 </thinking>
 Here is my response to your question."#;
-        let output = strip_thinking_tags(input);
+        let output = strip_internal_tags(input);
         assert_eq!(output, "Here is my response to your question.");
     }
 
     #[test]
     fn test_strip_thinking_tags_no_tags() {
         let input = "Just a normal response without thinking tags.";
-        let output = strip_thinking_tags(input);
+        let output = strip_internal_tags(input);
         assert_eq!(output, "Just a normal response without thinking tags.");
     }
 
@@ -609,10 +778,77 @@ Here is my response to your question."#;
     fn test_strip_thinking_tags_unclosed() {
         // Malformed: unclosed tag should strip from there to end
         let input = "Hello <thinking>this never closes";
-        let output = strip_thinking_tags(input);
+        let output = strip_internal_tags(input);
         assert_eq!(output, "Hello");
     }
 
+    #[test]
+    fn test_strip_tool_call_tags() {
+        // GLM-4.7 emits this garbage instead of using the tool_calls array
+        let input = "<tool_call>tool_list</tool_call>";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "");
+    }
+
+    #[test]
+    fn test_strip_tool_call_with_surrounding_text() {
+        let input = "Here is my answer.\n\n<tool_call>\n{\"name\": \"search\", \"arguments\": {}}\n</tool_call>";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "Here is my answer.");
+    }
+
+    #[test]
+    fn test_strip_multiple_internal_tags() {
+        let input = "<thinking>Let me think</thinking>Hello!\n<tool_call>some_tool</tool_call>";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "Hello!");
+    }
+
+    #[test]
+    fn test_strip_function_call_tags() {
+        let input = "Response text<function_call>{\"name\": \"foo\"}</function_call>";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "Response text");
+    }
+
+    #[test]
+    fn test_strip_tool_calls_plural() {
+        let input = "<tool_calls>[{\"id\": \"1\"}]</tool_calls>Actual response.";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "Actual response.");
+    }
+
+    #[test]
+    fn test_strip_pipe_delimited_tags() {
+        let input = "<|tool_call|>{\"name\": \"search\"}<|/tool_call|>Hello!";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "Hello!");
+    }
+
+    #[test]
+    fn test_strip_pipe_delimited_thinking() {
+        let input = "<|thinking|>reasoning here<|/thinking|>The answer is 42.";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "The answer is 42.");
+    }
+
+    #[test]
+    fn test_strip_xml_tag_with_attributes() {
+        let input = "<tool_call type=\"function\">search()</tool_call>Done.";
+        let output = strip_internal_tags(input);
+        assert_eq!(output, "Done.");
+    }
+
+    #[test]
+    fn test_clean_response_preserves_normal_content() {
+        let input = "The function tool_call_handler works great. No tags here!";
+        let output = clean_response(input);
+        assert_eq!(
+            output,
+            "The function tool_call_handler works great. No tags here!"
+        );
+    }
+
     #[test]
     fn test_strip_reasoning_paragraph_break() {
         // Content after paragraph break with "here" marker
@@ -660,4 +896,92 @@ Here is my response to your question."#;
         let output = clean_response(input);
         assert_eq!(output, "Here's the answer.");
     }
+
+    // -- recover_tool_calls_from_content tests --
+
+    fn make_tools(names: &[&str]) -> Vec<ToolDefinition> {
+        names
+            .iter()
+            .map(|n| ToolDefinition {
+                name: n.to_string(),
+                description: String::new(),
+                parameters: serde_json::json!({}),
+            })
+            .collect()
+    }
+
+    #[test]
+    fn test_recover_bare_tool_name() {
+        let tools = make_tools(&["tool_list", "tool_auth"]);
+        let content = "<tool_call>tool_list</tool_call>";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert_eq!(calls.len(), 1);
+        assert_eq!(calls[0].name, "tool_list");
+        assert_eq!(calls[0].arguments, serde_json::json!({}));
+    }
+
+    #[test]
+    fn test_recover_json_tool_call() {
+        let tools = make_tools(&["memory_search"]);
+        let content =
+            r#"<tool_call>{"name": "memory_search", "arguments": {"query": "test"}}</tool_call>"#;
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert_eq!(calls.len(), 1);
+        assert_eq!(calls[0].name, "memory_search");
+        assert_eq!(calls[0].arguments, serde_json::json!({"query": "test"}));
+    }
+
+    #[test]
+    fn test_recover_pipe_delimited() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "<|tool_call|>tool_list<|/tool_call|>";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert_eq!(calls.len(), 1);
+        assert_eq!(calls[0].name, "tool_list");
+    }
+
+    #[test]
+    fn test_recover_unknown_tool_ignored() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "<tool_call>nonexistent_tool</tool_call>";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert!(calls.is_empty());
+    }
+
+    #[test]
+    fn test_recover_no_tags() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "Just a normal response.";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert!(calls.is_empty());
+    }
+
+    #[test]
+    fn test_recover_multiple_tool_calls() {
+        let tools = make_tools(&["tool_list", "tool_auth"]);
+        let content = "<tool_call>tool_list</tool_call>\n<tool_call>tool_auth</tool_call>";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert_eq!(calls.len(), 2);
+        assert_eq!(calls[0].name, "tool_list");
+        assert_eq!(calls[1].name, "tool_auth");
+    }
+
+    #[test]
+    fn test_recover_function_call_variant() {
+        let tools = make_tools(&["shell"]);
+        let content =
+            r#"<function_call>{"name": "shell", "arguments": {"cmd": "ls"}}</function_call>"#;
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert_eq!(calls.len(), 1);
+        assert_eq!(calls[0].name, "shell");
+    }
+
+    #[test]
+    fn test_recover_with_surrounding_text() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "Let me check.\n\n<tool_call>tool_list</tool_call>\n\nDone.";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert_eq!(calls.len(), 1);
+        assert_eq!(calls[0].name, "tool_list");
+    }
 }

src/llm/rig_adapter.rs

@@ -0,0 +1,451 @@
+//! Generic adapter that bridges rig-core's `CompletionModel` trait to IronClaw's `LlmProvider`.
+//!
+//! This lets us use any rig-core provider (OpenAI, Anthropic, Ollama, etc.) as an
+//! `Arc<dyn LlmProvider>` without changing any of the agent, reasoning, or tool code.
+
+use async_trait::async_trait;
+use rig::OneOrMany;
+use rig::completion::{
+    AssistantContent, CompletionModel, CompletionRequest as RigRequest,
+    ToolDefinition as RigToolDefinition, Usage as RigUsage,
+};
+use rig::message::{
+    Message as RigMessage, ToolChoice as RigToolChoice, ToolFunction, ToolResult as RigToolResult,
+    ToolResultContent, UserContent,
+};
+use rust_decimal::Decimal;
+use serde::Serialize;
+use serde::de::DeserializeOwned;
+
+use crate::error::LlmError;
+use crate::llm::costs;
+use crate::llm::provider::{
+    ChatMessage, CompletionRequest, CompletionResponse, FinishReason, LlmProvider,
+    ToolCall as IronToolCall, ToolCompletionRequest, ToolCompletionResponse,
+    ToolDefinition as IronToolDefinition,
+};
+
+/// Adapter that wraps a rig-core `CompletionModel` and implements `LlmProvider`.
+pub struct RigAdapter<M: CompletionModel> {
+    model: M,
+    model_name: String,
+    input_cost: Decimal,
+    output_cost: Decimal,
+}
+
+impl<M: CompletionModel> RigAdapter<M> {
+    /// Create a new adapter wrapping the given rig-core model.
+    pub fn new(model: M, model_name: impl Into<String>) -> Self {
+        let name = model_name.into();
+        let (input_cost, output_cost) =
+            costs::model_cost(&name).unwrap_or_else(costs::default_cost);
+        Self {
+            model,
+            model_name: name,
+            input_cost,
+            output_cost,
+        }
+    }
+}
+
+// -- Type conversion helpers --
+
+/// Convert IronClaw messages to rig-core format.
+///
+/// Returns `(preamble, chat_history)` where preamble is extracted from
+/// any System message and chat_history contains the rest.
+fn convert_messages(messages: &[ChatMessage]) -> (Option<String>, Vec<RigMessage>) {
+    let mut preamble: Option<String> = None;
+    let mut history = Vec::new();
+
+    for msg in messages {
+        match msg.role {
+            crate::llm::Role::System => {
+                // Concatenate system messages into preamble
+                match preamble {
+                    Some(ref mut p) => {
+                        p.push('\n');
+                        p.push_str(&msg.content);
+                    }
+                    None => preamble = Some(msg.content.clone()),
+                }
+            }
+            crate::llm::Role::User => {
+                history.push(RigMessage::user(&msg.content));
+            }
+            crate::llm::Role::Assistant => {
+                if let Some(ref tool_calls) = msg.tool_calls {
+                    // Assistant message with tool calls
+                    let mut contents: Vec<AssistantContent> = Vec::new();
+                    if !msg.content.is_empty() {
+                        contents.push(AssistantContent::text(&msg.content));
+                    }
+                    for tc in tool_calls {
+                        contents.push(AssistantContent::ToolCall(rig::message::ToolCall::new(
+                            tc.id.clone(),
+                            ToolFunction::new(tc.name.clone(), tc.arguments.clone()),
+                        )));
+                    }
+                    if let Ok(many) = OneOrMany::many(contents) {
+                        history.push(RigMessage::Assistant {
+                            id: None,
+                            content: many,
+                        });
+                    } else {
+                        // Shouldn't happen but fall back to text
+                        history.push(RigMessage::assistant(&msg.content));
+                    }
+                } else {
+                    history.push(RigMessage::assistant(&msg.content));
+                }
+            }
+            crate::llm::Role::Tool => {
+                // Tool result message: wrap as User { ToolResult }
+                let tool_id = msg.tool_call_id.clone().unwrap_or_default();
+                history.push(RigMessage::User {
+                    content: OneOrMany::one(UserContent::ToolResult(RigToolResult {
+                        id: tool_id,
+                        call_id: None,
+                        content: OneOrMany::one(ToolResultContent::text(&msg.content)),
+                    })),
+                });
+            }
+        }
+    }
+
+    (preamble, history)
+}
+
+/// Convert IronClaw tool definitions to rig-core format.
+fn convert_tools(tools: &[IronToolDefinition]) -> Vec<RigToolDefinition> {
+    tools
+        .iter()
+        .map(|t| RigToolDefinition {
+            name: t.name.clone(),
+            description: t.description.clone(),
+            parameters: t.parameters.clone(),
+        })
+        .collect()
+}
+
+/// Convert IronClaw tool_choice string to rig-core ToolChoice.
+fn convert_tool_choice(choice: Option<&str>) -> Option<RigToolChoice> {
+    match choice.map(|s| s.to_lowercase()).as_deref() {
+        Some("auto") => Some(RigToolChoice::Auto),
+        Some("required") => Some(RigToolChoice::Required),
+        Some("none") => Some(RigToolChoice::None),
+        _ => None,
+    }
+}
+
+/// Extract text and tool calls from a rig-core completion response.
+fn extract_response(
+    choice: &OneOrMany<AssistantContent>,
+    _usage: &RigUsage,
+) -> (Option<String>, Vec<IronToolCall>, FinishReason) {
+    let mut text_parts: Vec<String> = Vec::new();
+    let mut tool_calls: Vec<IronToolCall> = Vec::new();
+
+    for content in choice.iter() {
+        match content {
+            AssistantContent::Text(t) => {
+                if !t.text.is_empty() {
+                    text_parts.push(t.text.clone());
+                }
+            }
+            AssistantContent::ToolCall(tc) => {
+                tool_calls.push(IronToolCall {
+                    id: tc.id.clone(),
+                    name: tc.function.name.clone(),
+                    arguments: tc.function.arguments.clone(),
+                });
+            }
+            // Reasoning and Image variants are not mapped to IronClaw types
+            _ => {}
+        }
+    }
+
+    let text = if text_parts.is_empty() {
+        None
+    } else {
+        Some(text_parts.join(""))
+    };
+
+    let finish = if !tool_calls.is_empty() {
+        FinishReason::ToolUse
+    } else {
+        FinishReason::Stop
+    };
+
+    (text, tool_calls, finish)
+}
+
+/// Saturate u64 to u32 for token counts.
+fn saturate_u32(val: u64) -> u32 {
+    val.min(u32::MAX as u64) as u32
+}
+
+/// Build a rig-core CompletionRequest from our internal types.
+fn build_rig_request(
+    preamble: Option<String>,
+    mut history: Vec<RigMessage>,
+    tools: Vec<RigToolDefinition>,
+    tool_choice: Option<RigToolChoice>,
+    temperature: Option<f32>,
+    max_tokens: Option<u32>,
+) -> Result<RigRequest, LlmError> {
+    // rig-core requires at least one message in chat_history
+    if history.is_empty() {
+        history.push(RigMessage::user("Hello"));
+    }
+
+    let chat_history = OneOrMany::many(history).map_err(|e| LlmError::RequestFailed {
+        provider: "rig".to_string(),
+        reason: format!("Failed to build chat history: {}", e),
+    })?;
+
+    Ok(RigRequest {
+        preamble,
+        chat_history,
+        documents: Vec::new(),
+        tools,
+        temperature: temperature.map(|t| t as f64),
+        max_tokens: max_tokens.map(|t| t as u64),
+        tool_choice,
+        additional_params: None,
+    })
+}
+
+#[async_trait]
+impl<M> LlmProvider for RigAdapter<M>
+where
+    M: CompletionModel + Send + Sync + 'static,
+    M::Response: Send + Sync + Serialize + DeserializeOwned,
+{
+    fn model_name(&self) -> &str {
+        &self.model_name
+    }
+
+    fn cost_per_token(&self) -> (Decimal, Decimal) {
+        (self.input_cost, self.output_cost)
+    }
+
+    async fn complete(&self, request: CompletionRequest) -> Result<CompletionResponse, LlmError> {
+        let (preamble, history) = convert_messages(&request.messages);
+
+        let rig_req = build_rig_request(
+            preamble,
+            history,
+            Vec::new(),
+            None,
+            request.temperature,
+            request.max_tokens,
+        )?;
+
+        let response =
+            self.model
+                .completion(rig_req)
+                .await
+                .map_err(|e| LlmError::RequestFailed {
+                    provider: self.model_name.clone(),
+                    reason: e.to_string(),
+                })?;
+
+        let (text, _tool_calls, finish) = extract_response(&response.choice, &response.usage);
+
+        Ok(CompletionResponse {
+            content: text.unwrap_or_default(),
+            input_tokens: saturate_u32(response.usage.input_tokens),
+            output_tokens: saturate_u32(response.usage.output_tokens),
+            finish_reason: finish,
+            response_id: None,
+        })
+    }
+
+    async fn complete_with_tools(
+        &self,
+        request: ToolCompletionRequest,
+    ) -> Result<ToolCompletionResponse, LlmError> {
+        let (preamble, history) = convert_messages(&request.messages);
+        let tools = convert_tools(&request.tools);
+        let tool_choice = convert_tool_choice(request.tool_choice.as_deref());
+
+        let rig_req = build_rig_request(
+            preamble,
+            history,
+            tools,
+            tool_choice,
+            request.temperature,
+            request.max_tokens,
+        )?;
+
+        let response =
+            self.model
+                .completion(rig_req)
+                .await
+                .map_err(|e| LlmError::RequestFailed {
+                    provider: self.model_name.clone(),
+                    reason: e.to_string(),
+                })?;
+
+        let (text, tool_calls, finish) = extract_response(&response.choice, &response.usage);
+
+        Ok(ToolCompletionResponse {
+            content: text,
+            tool_calls,
+            input_tokens: saturate_u32(response.usage.input_tokens),
+            output_tokens: saturate_u32(response.usage.output_tokens),
+            finish_reason: finish,
+            response_id: None,
+        })
+    }
+
+    fn active_model_name(&self) -> String {
+        self.model_name.clone()
+    }
+
+    fn set_model(&self, _model: &str) -> Result<(), LlmError> {
+        // rig-core models are baked at construction time.
+        // Switching requires creating a new adapter.
+        Err(LlmError::RequestFailed {
+            provider: self.model_name.clone(),
+            reason: "Runtime model switching not supported for rig-core providers. \
+                     Restart with a different model configured."
+                .to_string(),
+        })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_convert_messages_system_to_preamble() {
+        let messages = vec![
+            ChatMessage::system("You are a helpful assistant."),
+            ChatMessage::user("Hello"),
+        ];
+        let (preamble, history) = convert_messages(&messages);
+        assert_eq!(preamble, Some("You are a helpful assistant.".to_string()));
+        assert_eq!(history.len(), 1);
+    }
+
+    #[test]
+    fn test_convert_messages_multiple_systems_concatenated() {
+        let messages = vec![
+            ChatMessage::system("System 1"),
+            ChatMessage::system("System 2"),
+            ChatMessage::user("Hi"),
+        ];
+        let (preamble, history) = convert_messages(&messages);
+        assert_eq!(preamble, Some("System 1\nSystem 2".to_string()));
+        assert_eq!(history.len(), 1);
+    }
+
+    #[test]
+    fn test_convert_messages_tool_result() {
+        let messages = vec![ChatMessage::tool_result(
+            "call_123",
+            "search",
+            "result text",
+        )];
+        let (preamble, history) = convert_messages(&messages);
+        assert!(preamble.is_none());
+        assert_eq!(history.len(), 1);
+        // Tool results become User messages in rig-core
+        match &history[0] {
+            RigMessage::User { .. } => {}
+            other => panic!("Expected User message, got: {:?}", other),
+        }
+    }
+
+    #[test]
+    fn test_convert_messages_assistant_with_tool_calls() {
+        let tc = IronToolCall {
+            id: "call_1".to_string(),
+            name: "search".to_string(),
+            arguments: serde_json::json!({"query": "test"}),
+        };
+        let msg = ChatMessage::assistant_with_tool_calls(Some("thinking".to_string()), vec![tc]);
+        let messages = vec![msg];
+        let (_preamble, history) = convert_messages(&messages);
+        assert_eq!(history.len(), 1);
+        match &history[0] {
+            RigMessage::Assistant { content, .. } => {
+                // Should have both text and tool call
+                assert!(content.iter().count() >= 2);
+            }
+            other => panic!("Expected Assistant message, got: {:?}", other),
+        }
+    }
+
+    #[test]
+    fn test_convert_tools() {
+        let tools = vec![IronToolDefinition {
+            name: "search".to_string(),
+            description: "Search the web".to_string(),
+            parameters: serde_json::json!({
+                "type": "object",
+                "properties": {
+                    "query": {"type": "string"}
+                }
+            }),
+        }];
+        let rig_tools = convert_tools(&tools);
+        assert_eq!(rig_tools.len(), 1);
+        assert_eq!(rig_tools[0].name, "search");
+        assert_eq!(rig_tools[0].description, "Search the web");
+    }
+
+    #[test]
+    fn test_convert_tool_choice() {
+        assert!(matches!(
+            convert_tool_choice(Some("auto")),
+            Some(RigToolChoice::Auto)
+        ));
+        assert!(matches!(
+            convert_tool_choice(Some("required")),
+            Some(RigToolChoice::Required)
+        ));
+        assert!(matches!(
+            convert_tool_choice(Some("none")),
+            Some(RigToolChoice::None)
+        ));
+        assert!(matches!(
+            convert_tool_choice(Some("AUTO")),
+            Some(RigToolChoice::Auto)
+        ));
+        assert!(convert_tool_choice(None).is_none());
+        assert!(convert_tool_choice(Some("unknown")).is_none());
+    }
+
+    #[test]
+    fn test_extract_response_text_only() {
+        let content = OneOrMany::one(AssistantContent::text("Hello world"));
+        let usage = RigUsage::new();
+        let (text, calls, finish) = extract_response(&content, &usage);
+        assert_eq!(text, Some("Hello world".to_string()));
+        assert!(calls.is_empty());
+        assert_eq!(finish, FinishReason::Stop);
+    }
+
+    #[test]
+    fn test_extract_response_tool_call() {
+        let tc = AssistantContent::tool_call("call_1", "search", serde_json::json!({"q": "test"}));
+        let content = OneOrMany::one(tc);
+        let usage = RigUsage::new();
+        let (text, calls, finish) = extract_response(&content, &usage);
+        assert!(text.is_none());
+        assert_eq!(calls.len(), 1);
+        assert_eq!(calls[0].name, "search");
+        assert_eq!(finish, FinishReason::ToolUse);
+    }
+
+    #[test]
+    fn test_saturate_u32() {
+        assert_eq!(saturate_u32(100), 100);
+        assert_eq!(saturate_u32(u64::MAX), u32::MAX);
+        assert_eq!(saturate_u32(u32::MAX as u64), u32::MAX);
+    }
+}

src/llm/session.rs

@@ -61,6 +61,10 @@ pub struct SessionManager {
     token: RwLock<Option<SecretString>>,
     /// Prevents thundering herd during concurrent 401s.
     renewal_lock: Mutex<()>,
+    /// Optional database store for persisting session to the settings table.
+    store: RwLock<Option<Arc<crate::history::Store>>>,
+    /// User ID for DB settings (default: "default").
+    user_id: RwLock<String>,
 }
 
 impl SessionManager {
@@ -74,6 +78,8 @@ impl SessionManager {
                 .unwrap_or_else(|_| Client::new()),
             token: RwLock::new(None),
             renewal_lock: Mutex::new(()),
+            store: RwLock::new(None),
+            user_id: RwLock::new("default".to_string()),
         };
 
         // Try to load existing session synchronously during construction
@@ -103,6 +109,8 @@ impl SessionManager {
                 .unwrap_or_else(|_| Client::new()),
             token: RwLock::new(None),
             renewal_lock: Mutex::new(()),
+            store: RwLock::new(None),
+            user_id: RwLock::new("default".to_string()),
         };
 
         if let Err(e) = manager.load_session().await {
@@ -112,6 +120,21 @@ impl SessionManager {
         manager
     }
 
+    /// Attach a database store for persisting session tokens.
+    ///
+    /// When a store is attached, session tokens are saved to the `settings`
+    /// table (key: `nearai.session_token`) in addition to the disk file.
+    /// On load, DB is preferred over disk.
+    pub async fn attach_store(&self, store: Arc<crate::history::Store>, user_id: &str) {
+        *self.store.write().await = Some(store);
+        *self.user_id.write().await = user_id.to_string();
+
+        // Try to load from DB (may have been saved by a previous run)
+        if let Err(e) = self.load_session_from_db().await {
+            tracing::debug!("No session in DB: {}", e);
+        }
+    }
+
     /// Get the current session token, returning an error if not authenticated.
     pub async fn get_token(&self) -> Result<SecretString, LlmError> {
         let guard = self.token.read().await;
@@ -460,7 +483,7 @@ impl SessionManager {
         Ok(())
     }
 
-    /// Save session data to disk.
+    /// Save session data to disk and (if available) to the database.
     async fn save_session(&self, token: &str, auth_provider: Option<&str>) -> Result<(), LlmError> {
         let session = SessionData {
             session_token: token.to_string(),
@@ -468,7 +491,7 @@ impl SessionManager {
             auth_provider: auth_provider.map(String::from),
         };
 
-        // Ensure parent directory exists
+        // Save to disk (always, as bootstrap fallback)
         if let Some(parent) = self.config.session_path.parent() {
             tokio::fs::create_dir_all(parent).await.map_err(|e| {
                 LlmError::Io(std::io::Error::new(
@@ -498,6 +521,58 @@ impl SessionManager {
             })?;
 
         tracing::debug!("Session saved to {}", self.config.session_path.display());
+
+        // Also save to DB if a store is attached
+        if let Some(ref store) = *self.store.read().await {
+            let user_id = self.user_id.read().await.clone();
+            let session_json = serde_json::to_value(&session)
+                .unwrap_or(serde_json::Value::String(token.to_string()));
+            if let Err(e) = store
+                .set_setting(&user_id, "nearai.session_token", &session_json)
+                .await
+            {
+                tracing::warn!("Failed to save session to DB: {}", e);
+            } else {
+                tracing::debug!("Session also saved to DB settings");
+            }
+        }
+
+        Ok(())
+    }
+
+    /// Try to load session from the database.
+    async fn load_session_from_db(&self) -> Result<(), LlmError> {
+        let store_guard = self.store.read().await;
+        let store = store_guard
+            .as_ref()
+            .ok_or_else(|| LlmError::SessionRenewalFailed {
+                provider: "nearai".to_string(),
+                reason: "No DB store attached".to_string(),
+            })?;
+
+        let user_id = self.user_id.read().await.clone();
+        let value = store
+            .get_setting(&user_id, "nearai.session_token")
+            .await
+            .map_err(|e| LlmError::SessionRenewalFailed {
+                provider: "nearai".to_string(),
+                reason: format!("DB query failed: {}", e),
+            })?
+            .ok_or_else(|| LlmError::SessionRenewalFailed {
+                provider: "nearai".to_string(),
+                reason: "No session in DB".to_string(),
+            })?;
+
+        let session: SessionData =
+            serde_json::from_value(value).map_err(|e| LlmError::SessionRenewalFailed {
+                provider: "nearai".to_string(),
+                reason: format!("Failed to parse DB session: {}", e),
+            })?;
+
+        let mut guard = self.token.write().await;
+        *guard = Some(SecretString::from(session.session_token));
+        tracing::info!("Loaded session from DB settings");
+
         Ok(())
     }
 

src/main.rs

@@ -17,21 +17,26 @@ use ironclaw::{
         web::log_layer::{LogBroadcaster, WebLogLayer},
     },
     cli::{
-        Cli, Command, run_mcp_command, run_memory_command, run_status_command, run_tool_command,
+        Cli, Command, run_mcp_command, run_memory_command, run_pairing_command, run_status_command,
+        run_tool_command,
     },
     config::Config,
     context::ContextManager,
     extensions::ExtensionManager,
     history::Store,
     llm::{SessionConfig, create_llm_provider, create_session_manager},
+    orchestrator::{
+        ContainerJobConfig, ContainerJobManager, OrchestratorApi, TokenStore,
+        api::OrchestratorState,
+    },
+    pairing::PairingStore,
     safety::SafetyLayer,
     secrets::{PostgresSecretsStore, SecretsCrypto, SecretsStore},
-    settings::Settings,
     setup::{SetupConfig, SetupWizard},
     tools::{
         ToolRegistry,
-        mcp::{McpClient, McpSessionManager, config::load_mcp_servers, is_authenticated},
-        wasm::{WasmToolLoader, WasmToolRuntime},
+        mcp::{McpClient, McpSessionManager, config::load_mcp_servers_from_db, is_authenticated},
+        wasm::{WasmToolLoader, WasmToolRuntime, load_dev_tools},
     },
     workspace::{EmbeddingProvider, NearAiEmbeddings, OpenAiEmbeddings, Workspace},
 };
@@ -53,9 +58,14 @@ async fn main() -> anyhow::Result<()> {
             return run_tool_command(tool_cmd.clone()).await;
         }
         Some(Command::Config(config_cmd)) => {
-            // Config commands don't need logging setup
-            return ironclaw::cli::run_config_command(config_cmd.clone())
-                .map_err(|e| anyhow::anyhow!("{}", e));
+            // Config commands need DB access for settings
+            tracing_subscriber::fmt()
+                .with_env_filter(
+                    EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new("warn")),
+                )
+                .init();
+
+            return ironclaw::cli::run_config_command(config_cmd.clone()).await;
         }
         Some(Command::Mcp(mcp_cmd)) => {
             // Simple logging for MCP commands
@@ -76,7 +86,9 @@ async fn main() -> anyhow::Result<()> {
 
             // Memory commands need database (and optionally embeddings)
             let _ = dotenvy::dotenv();
-            let config = Config::from_env().map_err(|e| anyhow::anyhow!("{}", e))?;
+            let config = Config::from_env()
+                .await
+                .map_err(|e| anyhow::anyhow!("{}", e))?;
             let store = ironclaw::history::Store::new(&config.database).await?;
             store.run_migrations().await?;
 
@@ -120,6 +132,15 @@ async fn main() -> anyhow::Result<()> {
 
             return run_memory_command(mem_cmd.clone(), store.pool(), embeddings).await;
         }
+        Some(Command::Pairing(pairing_cmd)) => {
+            tracing_subscriber::fmt()
+                .with_env_filter(
+                    EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new("warn")),
+                )
+                .init();
+
+            return run_pairing_command(pairing_cmd.clone()).map_err(|e| anyhow::anyhow!("{}", e));
+        }
         Some(Command::Status) => {
             let _ = dotenvy::dotenv();
             tracing_subscriber::fmt()
@@ -130,6 +151,83 @@ async fn main() -> anyhow::Result<()> {
 
             return run_status_command().await;
         }
+        Some(Command::Worker {
+            job_id,
+            orchestrator_url,
+            max_iterations,
+        }) => {
+            // Worker mode: runs inside a Docker container.
+            // Simple logging (no TUI, no DB, no channels).
+            tracing_subscriber::fmt()
+                .with_env_filter(
+                    EnvFilter::try_from_default_env()
+                        .unwrap_or_else(|_| EnvFilter::new("ironclaw=info")),
+                )
+                .init();
+
+            tracing::info!(
+                "Starting worker for job {} (orchestrator: {})",
+                job_id,
+                orchestrator_url
+            );
+
+            let config = ironclaw::worker::runtime::WorkerConfig {
+                job_id: *job_id,
+                orchestrator_url: orchestrator_url.clone(),
+                max_iterations: *max_iterations,
+                timeout: std::time::Duration::from_secs(600),
+            };
+
+            let runtime = ironclaw::worker::WorkerRuntime::new(config)
+                .map_err(|e| anyhow::anyhow!("Worker init failed: {}", e))?;
+
+            runtime
+                .run()
+                .await
+                .map_err(|e| anyhow::anyhow!("Worker failed: {}", e))?;
+
+            return Ok(());
+        }
+        Some(Command::ClaudeBridge {
+            job_id,
+            orchestrator_url,
+            max_turns,
+            model,
+        }) => {
+            // Claude Code bridge mode: runs inside a Docker container.
+            // Spawns the `claude` CLI and streams output to the orchestrator.
+            tracing_subscriber::fmt()
+                .with_env_filter(
+                    EnvFilter::try_from_default_env()
+                        .unwrap_or_else(|_| EnvFilter::new("ironclaw=info")),
+                )
+                .init();
+
+            tracing::info!(
+                "Starting Claude Code bridge for job {} (orchestrator: {}, model: {})",
+                job_id,
+                orchestrator_url,
+                model
+            );
+
+            let config = ironclaw::worker::claude_bridge::ClaudeBridgeConfig {
+                job_id: *job_id,
+                orchestrator_url: orchestrator_url.clone(),
+                max_turns: *max_turns,
+                model: model.clone(),
+                timeout: std::time::Duration::from_secs(1800),
+            };
+
+            let runtime = ironclaw::worker::ClaudeBridgeRuntime::new(config)
+                .map_err(|e| anyhow::anyhow!("Claude bridge init failed: {}", e))?;
+
+            runtime
+                .run()
+                .await
+                .map_err(|e| anyhow::anyhow!("Claude bridge failed: {}", e))?;
+
+            return Ok(());
+        }
         Some(Command::Onboard {
             skip_auth,
             channels_only,
@@ -155,7 +253,7 @@ async fn main() -> anyhow::Result<()> {
 
     // Enhanced first-run detection
     if !cli.no_onboard {
-        if let Some(reason) = check_onboard_needed() {
+        if let Some(reason) = check_onboard_needed().await {
             println!("Onboarding needed: {}", reason);
             println!();
             let mut wizard = SetupWizard::new();
@@ -163,8 +261,11 @@ async fn main() -> anyhow::Result<()> {
         }
     }
 
-    // Load configuration (after potential setup)
-    let config = match Config::from_env() {
+    // Load bootstrap config (4 fields that must live on disk)
+    let bootstrap = ironclaw::bootstrap::BootstrapConfig::load();
+
+    // Load initial config from env + disk (before DB is available)
+    let mut config = match Config::from_env().await {
         Ok(c) => c,
         Err(ironclaw::error::ConfigError::MissingRequired { key, hint }) => {
             eprintln!("Configuration error: Missing required setting '{}'", key);
@@ -186,8 +287,10 @@ async fn main() -> anyhow::Result<()> {
     };
     let session = create_session_manager(session_config).await;
 
-    // Ensure we're authenticated before proceeding (may trigger login flow)
-    session.ensure_authenticated().await?;
+    // Ensure we're authenticated before proceeding (only needed for NEAR AI backend)
+    if config.llm.backend == ironclaw::config::LlmBackend::NearAi {
+        session.ensure_authenticated().await?;
+    }
 
     // Initialize tracing
     let env_filter = EnvFilter::try_from_default_env()
@@ -214,7 +317,7 @@ async fn main() -> anyhow::Result<()> {
 
     tracing::info!("Starting IronClaw...");
     tracing::info!("Loaded configuration for agent: {}", config.agent.name);
-    tracing::info!("NEAR AI session authenticated");
+    tracing::info!("LLM backend: {}", config.llm.backend);
 
     // Initialize database store (optional for testing)
     let store = if cli.no_db {
@@ -224,7 +327,38 @@ async fn main() -> anyhow::Result<()> {
         let store = Store::new(&config.database).await?;
         store.run_migrations().await?;
         tracing::info!("Database connected and migrations applied");
-        Some(Arc::new(store))
+
+        // One-time migration: move disk config files into the DB settings table.
+        if let Err(e) = ironclaw::bootstrap::migrate_disk_to_db(&store, "default").await {
+            tracing::warn!("Disk-to-DB settings migration failed: {}", e);
+        }
+
+        // Reload config from DB now that we have a connection.
+        // Priority: env var > DB setting > default.
+        match Config::from_db(&store, "default", &bootstrap).await {
+            Ok(db_config) => {
+                config = db_config;
+                tracing::info!("Configuration reloaded from database");
+            }
+            Err(e) => {
+                tracing::warn!(
+                    "Failed to reload config from DB, keeping env-based config: {}",
+                    e
+                );
+            }
+        }
+
+        let store = Arc::new(store);
+
+        // Attach store to session manager so tokens save to DB too
+        session.attach_store(Arc::clone(&store), "default").await;
+
+        // Mark any jobs left in "running" or "creating" state as "interrupted".
+        if let Err(e) = store.cleanup_stale_sandbox_jobs().await {
+            tracing::warn!("Failed to cleanup stale sandbox jobs: {}", e);
+        }
+
+        Some(store)
     };
 
     // Initialize LLM provider (clone session so we can reuse it for embeddings)
@@ -289,8 +423,11 @@ async fn main() -> anyhow::Result<()> {
         tools.register_memory_tools(workspace);
     }
 
-    // Register builder tool if enabled
-    if config.builder.enabled {
+    // Register builder tool if enabled.
+    // When sandbox is enabled and allow_local_tools is false, skip builder registration
+    // because register_builder_tool also registers dev tools (shell, file ops) that would
+    // bypass the sandbox. The builder runs inside containers instead.
+    if config.builder.enabled && (config.agent.allow_local_tools || !config.sandbox.enabled) {
         tools
             .register_builder_tool(
                 llm.clone(),
@@ -339,6 +476,8 @@ async fn main() -> anyhow::Result<()> {
     let wasm_tools_future = async {
         if let Some(ref runtime) = wasm_tool_runtime {
             let loader = WasmToolLoader::new(Arc::clone(runtime), Arc::clone(&tools));
+
+            // Load installed tools from ~/.ironclaw/tools/
             match loader.load_from_dir(&config.wasm.tools_dir).await {
                 Ok(results) => {
                     if !results.loaded.is_empty() {
@@ -356,12 +495,32 @@ async fn main() -> anyhow::Result<()> {
                     tracing::warn!("Failed to scan WASM tools directory: {}", e);
                 }
             }
+
+            // Load dev tools from build artifacts (overrides installed if newer)
+            match load_dev_tools(&loader, &config.wasm.tools_dir).await {
+                Ok(results) => {
+                    if !results.loaded.is_empty() {
+                        tracing::info!(
+                            "Loaded {} dev WASM tools from build artifacts",
+                            results.loaded.len()
+                        );
+                    }
+                }
+                Err(e) => {
+                    tracing::debug!("No dev WASM tools found: {}", e);
+                }
+            }
         }
     };
 
     let mcp_servers_future = async {
         if let Some(ref secrets) = secrets_store {
-            match load_mcp_servers().await {
+            let servers_result = if let Some(ref s) = store {
+                load_mcp_servers_from_db(s, "default").await
+            } else {
+                ironclaw::tools::mcp::config::load_mcp_servers().await
+            };
+            match servers_result {
                 Ok(servers) => {
                     let enabled: Vec<_> = servers.enabled_servers().cloned().collect();
                     if !enabled.is_empty() {
@@ -470,6 +629,7 @@ async fn main() -> anyhow::Result<()> {
             config.channels.wasm_channels_dir.clone(),
             config.tunnel.public_url.clone(),
             "default".to_string(),
+            store.clone(),
         ));
         tools.register_extension_tools(Arc::clone(&manager));
         tracing::info!("Extension manager initialized with in-chat discovery tools");
@@ -482,6 +642,77 @@ async fn main() -> anyhow::Result<()> {
         None
     };
 
+    // Set up orchestrator for sandboxed job execution
+    // When allow_local_tools is false (default), the LLM uses create_job for FS/shell work.
+    // When allow_local_tools is true, dev tools are also registered directly (current behavior).
+    if config.agent.allow_local_tools {
+        tools.register_dev_tools();
+        tracing::info!(
+            "Local tools enabled (allow_local_tools=true), dev tools registered directly"
+        );
+    }
+
+    // Shared state for job events (used by both orchestrator and web gateway)
+    let job_event_tx: Option<
+        tokio::sync::broadcast::Sender<(uuid::Uuid, ironclaw::channels::web::types::SseEvent)>,
+    > = if config.sandbox.enabled {
+        let (tx, _) = tokio::sync::broadcast::channel(256);
+        Some(tx)
+    } else {
+        None
+    };
+    let prompt_queue = Arc::new(tokio::sync::Mutex::new(std::collections::HashMap::<
+        uuid::Uuid,
+        std::collections::VecDeque<ironclaw::orchestrator::api::PendingPrompt>,
+    >::new()));
+
+    let container_job_manager: Option<Arc<ContainerJobManager>> = if config.sandbox.enabled {
+        let token_store = TokenStore::new();
+        let job_config = ContainerJobConfig {
+            image: config.sandbox.image.clone(),
+            memory_limit_mb: config.sandbox.memory_limit_mb,
+            cpu_shares: config.sandbox.cpu_shares,
+            orchestrator_port: 50051,
+            claude_config_dir: if config.claude_code.enabled {
+                Some(config.claude_code.config_dir.clone())
+            } else {
+                None
+            },
+            claude_code_model: config.claude_code.model.clone(),
+            claude_code_max_turns: config.claude_code.max_turns,
+            claude_code_memory_limit_mb: config.claude_code.memory_limit_mb,
+        };
+        let jm = Arc::new(ContainerJobManager::new(job_config, token_store.clone()));
+
+        // Start the orchestrator internal API in the background
+        let orchestrator_state = OrchestratorState {
+            llm: llm.clone(),
+            job_manager: Arc::clone(&jm),
+            token_store,
+            job_event_tx: job_event_tx.clone(),
+            prompt_queue: Arc::clone(&prompt_queue),
+            store: store.clone(),
+        };
+
+        tokio::spawn(async move {
+            if let Err(e) = OrchestratorApi::start(orchestrator_state, 50051).await {
+                tracing::error!("Orchestrator API failed: {}", e);
+            }
+        });
+
+        tracing::info!("Orchestrator API started on :50051, sandbox delegation enabled");
+        if config.claude_code.enabled {
+            tracing::info!(
+                "Claude Code sandbox mode available (model: {}, max_turns: {})",
+                config.claude_code.model,
+                config.claude_code.max_turns
+            );
+        }
+        Some(jm)
+    } else {
+        None
+    };
+
     tracing::info!(
         "Tool registry initialized with {} total tools",
         tools.count()
@@ -507,7 +738,8 @@ async fn main() -> anyhow::Result<()> {
         match WasmChannelRuntime::new(WasmChannelRuntimeConfig::default()) {
             Ok(runtime) => {
                 let runtime = Arc::new(runtime);
-                let loader = WasmChannelLoader::new(Arc::clone(&runtime));
+                let pairing_store = Arc::new(PairingStore::new());
+                let loader = WasmChannelLoader::new(Arc::clone(&runtime), pairing_store);
 
                 match loader
                     .load_from_dir(&config.channels.wasm_channels_dir)
@@ -563,6 +795,17 @@ async fn main() -> anyhow::Result<()> {
                                     );
                                 }
 
+                                // Inject owner_id for Telegram so the bot only responds
+                                // to the bound user account.
+                                if channel_name == "telegram" {
+                                    if let Some(owner_id) = config.channels.telegram_owner_id {
+                                        config_updates.insert(
+                                            "owner_id".to_string(),
+                                            serde_json::json!(owner_id),
+                                        );
+                                    }
+                                }
+
                                 if !config_updates.is_empty() {
                                     channel_arc.update_config(config_updates).await;
                                     tracing::info!(
@@ -621,7 +864,7 @@ async fn main() -> anyhow::Result<()> {
                             channels.add(Box::new(SharedWasmChannel::new(channel_arc)));
                         }
 
-                        if has_webhook_channels && config.tunnel.public_url.is_some() {
+                        if has_webhook_channels {
                             webhook_routes.push(create_wasm_channel_router(
                                 wasm_router,
                                 extension_manager.as_ref().map(Arc::clone),
@@ -693,6 +936,19 @@ async fn main() -> anyhow::Result<()> {
         Arc::new(ws)
     });
 
+    // Seed workspace with core identity files on first boot
+    if let Some(ref ws) = workspace {
+        match ws.seed_if_empty().await {
+            Ok(count) if count > 0 => {
+                tracing::info!("Workspace seeded with {} core files", count);
+            }
+            Ok(_) => {}
+            Err(e) => {
+                tracing::warn!("Failed to seed workspace: {}", e);
+            }
+        }
+    }
+
     // Backfill embeddings if we just enabled the provider
     if let (Some(ws), Some(_)) = (&workspace, &embeddings) {
         match ws.backfill_embeddings().await {
@@ -712,8 +968,12 @@ async fn main() -> anyhow::Result<()> {
     // Create session manager (shared between agent and web gateway)
     let session_manager = Arc::new(SessionManager::new());
 
-    // Register job tools
-    tools.register_job_tools(Arc::clone(&context_manager));
+    // Register job tools (sandbox deps auto-injected when container_job_manager is available)
+    tools.register_job_tools(
+        Arc::clone(&context_manager),
+        container_job_manager.clone(),
+        store.clone(),
+    );
 
     // Add web gateway channel if configured
     if let Some(ref gw_config) = config.channels.gateway {
@@ -721,19 +981,44 @@ async fn main() -> anyhow::Result<()> {
         if let Some(ref ws) = workspace {
             gw = gw.with_workspace(Arc::clone(ws));
         }
-        gw = gw.with_context_manager(Arc::clone(&context_manager));
         gw = gw.with_session_manager(Arc::clone(&session_manager));
         gw = gw.with_log_broadcaster(Arc::clone(&log_broadcaster));
         gw = gw.with_tool_registry(Arc::clone(&tools));
         if let Some(ref ext_mgr) = extension_manager {
             gw = gw.with_extension_manager(Arc::clone(ext_mgr));
         }
+        if let Some(ref s) = store {
+            gw = gw.with_store(Arc::clone(s));
+        }
+        if let Some(ref jm) = container_job_manager {
+            gw = gw.with_job_manager(Arc::clone(jm));
+        }
+        if config.sandbox.enabled {
+            gw = gw.with_prompt_queue(Arc::clone(&prompt_queue));
+
+            // Spawn a task to forward job events from the broadcast channel to SSE
+            if let Some(ref tx) = job_event_tx {
+                let mut rx = tx.subscribe();
+                let gw_state = Arc::clone(gw.state());
+                tokio::spawn(async move {
+                    while let Ok((_job_id, event)) = rx.recv().await {
+                        gw_state.sse.broadcast(event);
+                    }
+                });
+            }
+        }
 
         tracing::info!(
             "Web gateway enabled on {}:{}",
             gw_config.host,
             gw_config.port
         );
+        tracing::info!(
+            "Web UI: http://{}:{}/?token={}",
+            gw_config.host,
+            gw_config.port,
+            gw.auth_token()
+        );
 
         channels.add(Box::new(gw));
     }
@@ -752,6 +1037,7 @@ async fn main() -> anyhow::Result<()> {
         deps,
         channels,
         Some(config.heartbeat.clone()),
+        Some(config.routines.clone()),
         Some(context_manager),
         Some(session_manager),
     );
@@ -773,18 +1059,18 @@ async fn main() -> anyhow::Result<()> {
 /// Check if onboarding is needed and return the reason.
 ///
 /// Returns `Some(reason)` if onboarding should be triggered, `None` otherwise.
-fn check_onboard_needed() -> Option<&'static str> {
-    let settings = Settings::load();
+async fn check_onboard_needed() -> Option<&'static str> {
+    let bootstrap = ironclaw::bootstrap::BootstrapConfig::load();
 
     // Database not configured (and not in env)
-    if settings.database_url.is_none() && std::env::var("DATABASE_URL").is_err() {
+    if bootstrap.database_url.is_none() && std::env::var("DATABASE_URL").is_err() {
         return Some("Database not configured");
     }
 
     // Secrets not configured (and not in env)
-    if settings.secrets_master_key_source == ironclaw::settings::KeySource::None
+    if bootstrap.secrets_master_key_source == ironclaw::settings::KeySource::None
         && std::env::var("SECRETS_MASTER_KEY").is_err()
-        && !ironclaw::secrets::keychain::has_master_key()
+        && !ironclaw::secrets::keychain::has_master_key().await
     {
         // Only require secrets setup if user hasn't explicitly disabled it
         // For now, we don't require it for first run
@@ -792,7 +1078,7 @@ fn check_onboard_needed() -> Option<&'static str> {
 
     // First run (onboarding never completed and no session)
     let session_path = ironclaw::llm::session::default_session_path();
-    if !settings.onboard_completed && !session_path.exists() {
+    if !bootstrap.onboard_completed && !session_path.exists() {
         return Some("First run");
     }
 

src/orchestrator/api.rs

@@ -0,0 +1,511 @@
+//! Internal HTTP API for worker-to-orchestrator communication.
+//!
+//! This runs on a separate port (default 50051) from the web gateway.
+//! All endpoints are authenticated via per-job bearer tokens.
+
+use std::collections::{HashMap, VecDeque};
+use std::sync::Arc;
+
+use axum::extract::{Path, State};
+use axum::http::StatusCode;
+use axum::routing::{get, post};
+use axum::{Json, Router};
+use serde::{Deserialize, Serialize};
+use tokio::sync::{Mutex, broadcast};
+use uuid::Uuid;
+
+use crate::channels::web::types::SseEvent;
+use crate::history::Store;
+use crate::llm::{CompletionRequest, LlmProvider, ToolCompletionRequest};
+use crate::orchestrator::auth::{TokenStore, worker_auth_middleware};
+use crate::orchestrator::job_manager::ContainerJobManager;
+use crate::worker::api::JobEventPayload;
+use crate::worker::api::{
+    CompletionReport, JobDescription, ProxyCompletionRequest, ProxyCompletionResponse,
+    ProxyToolCompletionRequest, ProxyToolCompletionResponse, StatusUpdate,
+};
+
+/// A follow-up prompt queued for a Claude Code bridge.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PendingPrompt {
+    pub content: String,
+    pub done: bool,
+}
+
+/// Shared state for the orchestrator API.
+#[derive(Clone)]
+pub struct OrchestratorState {
+    pub llm: Arc<dyn LlmProvider>,
+    pub job_manager: Arc<ContainerJobManager>,
+    pub token_store: TokenStore,
+    /// Broadcast channel for job events (consumed by the web gateway SSE).
+    pub job_event_tx: Option<broadcast::Sender<(Uuid, SseEvent)>>,
+    /// Buffered follow-up prompts for sandbox jobs, keyed by job_id.
+    pub prompt_queue: Arc<Mutex<HashMap<Uuid, VecDeque<PendingPrompt>>>>,
+    /// Database handle for persisting job events.
+    pub store: Option<Arc<Store>>,
+}
+
+/// The orchestrator's internal API server.
+pub struct OrchestratorApi;
+
+impl OrchestratorApi {
+    /// Build the axum router for the internal API.
+    pub fn router(state: OrchestratorState) -> Router {
+        Router::new()
+            // Worker routes: authenticated via route_layer middleware.
+            .route("/worker/{job_id}/job", get(get_job))
+            .route("/worker/{job_id}/llm/complete", post(llm_complete))
+            .route(
+                "/worker/{job_id}/llm/complete_with_tools",
+                post(llm_complete_with_tools),
+            )
+            .route("/worker/{job_id}/status", post(report_status))
+            .route("/worker/{job_id}/complete", post(report_complete))
+            .route("/worker/{job_id}/event", post(job_event_handler))
+            .route("/worker/{job_id}/prompt", get(get_prompt_handler))
+            .route_layer(axum::middleware::from_fn_with_state(
+                state.token_store.clone(),
+                worker_auth_middleware,
+            ))
+            // Unauthenticated routes (added after the layer).
+            .route("/health", get(health_check))
+            .with_state(state)
+    }
+
+    /// Start the internal API server on the given port.
+    ///
+    /// On macOS/Windows (Docker Desktop), binds to loopback only because
+    /// Docker Desktop routes `host.docker.internal` through its VM to the
+    /// host's `127.0.0.1`.
+    ///
+    /// On Linux, containers reach the host via the docker bridge gateway
+    /// (`172.17.0.1`), which is NOT loopback. Binding to `127.0.0.1`
+    /// would reject container traffic. We bind to all interfaces instead
+    /// and rely on `worker_auth_middleware` (applied as a route_layer on
+    /// every `/worker/` endpoint) to reject unauthenticated requests.
+    pub async fn start(
+        state: OrchestratorState,
+        port: u16,
+    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let router = Self::router(state);
+        let addr = if cfg!(target_os = "linux") {
+            std::net::SocketAddr::from(([0, 0, 0, 0], port))
+        } else {
+            std::net::SocketAddr::from(([127, 0, 0, 1], port))
+        };
+
+        tracing::info!("Orchestrator internal API listening on {}", addr);
+
+        let listener = tokio::net::TcpListener::bind(addr).await?;
+        axum::serve(listener, router).await?;
+
+        Ok(())
+    }
+}
+
+// -- Handlers --
+//
+// All /worker/ handlers below are behind the worker_auth_middleware route_layer,
+// so they don't need to validate tokens themselves.
+
+async fn health_check() -> &'static str {
+    "ok"
+}
+
+async fn get_job(
+    State(state): State<OrchestratorState>,
+    Path(job_id): Path<Uuid>,
+) -> Result<Json<JobDescription>, StatusCode> {
+    let handle = state
+        .job_manager
+        .get_handle(job_id)
+        .await
+        .ok_or(StatusCode::NOT_FOUND)?;
+
+    Ok(Json(JobDescription {
+        title: format!("Job {}", job_id),
+        description: handle.task_description,
+        project_dir: handle.project_dir.map(|p| p.display().to_string()),
+    }))
+}
+
+async fn llm_complete(
+    State(state): State<OrchestratorState>,
+    Path(job_id): Path<Uuid>,
+    Json(req): Json<ProxyCompletionRequest>,
+) -> Result<Json<ProxyCompletionResponse>, StatusCode> {
+    let completion_req = CompletionRequest {
+        messages: req.messages,
+        max_tokens: req.max_tokens,
+        temperature: req.temperature,
+        stop_sequences: req.stop_sequences,
+        metadata: std::collections::HashMap::new(),
+    };
+
+    let resp = state.llm.complete(completion_req).await.map_err(|e| {
+        tracing::error!("LLM completion failed for job {}: {}", job_id, e);
+        StatusCode::BAD_GATEWAY
+    })?;
+
+    Ok(Json(ProxyCompletionResponse {
+        content: resp.content,
+        input_tokens: resp.input_tokens,
+        output_tokens: resp.output_tokens,
+        finish_reason: format_finish_reason(resp.finish_reason),
+    }))
+}
+
+async fn llm_complete_with_tools(
+    State(state): State<OrchestratorState>,
+    Path(job_id): Path<Uuid>,
+    Json(req): Json<ProxyToolCompletionRequest>,
+) -> Result<Json<ProxyToolCompletionResponse>, StatusCode> {
+    let tool_req = ToolCompletionRequest {
+        messages: req.messages,
+        tools: req.tools,
+        max_tokens: req.max_tokens,
+        temperature: req.temperature,
+        tool_choice: req.tool_choice,
+        metadata: std::collections::HashMap::new(),
+    };
+
+    let resp = state.llm.complete_with_tools(tool_req).await.map_err(|e| {
+        tracing::error!("LLM tool completion failed for job {}: {}", job_id, e);
+        StatusCode::BAD_GATEWAY
+    })?;
+
+    Ok(Json(ProxyToolCompletionResponse {
+        content: resp.content,
+        tool_calls: resp.tool_calls,
+        input_tokens: resp.input_tokens,
+        output_tokens: resp.output_tokens,
+        finish_reason: format_finish_reason(resp.finish_reason),
+    }))
+}
+
+async fn report_status(
+    Path(job_id): Path<Uuid>,
+    Json(update): Json<StatusUpdate>,
+) -> Result<StatusCode, StatusCode> {
+    tracing::debug!(
+        job_id = %job_id,
+        state = %update.state,
+        iteration = update.iteration,
+        "Worker status update"
+    );
+
+    Ok(StatusCode::OK)
+}
+
+async fn report_complete(
+    State(state): State<OrchestratorState>,
+    Path(job_id): Path<Uuid>,
+    Json(report): Json<CompletionReport>,
+) -> Result<StatusCode, StatusCode> {
+    if report.success {
+        tracing::info!(
+            job_id = %job_id,
+            "Worker reported job complete"
+        );
+    } else {
+        tracing::warn!(
+            job_id = %job_id,
+            message = ?report.message,
+            "Worker reported job failure"
+        );
+    }
+
+    // Store the result and clean up the container
+    let result = crate::orchestrator::job_manager::CompletionResult {
+        success: report.success,
+        message: report.message.clone(),
+    };
+    let _ = state.job_manager.complete_job(job_id, result).await;
+
+    Ok(StatusCode::OK)
+}
+
+// -- Sandbox job event handlers --
+
+/// Receive a job event from a worker or Claude Code bridge and broadcast + persist it.
+async fn job_event_handler(
+    State(state): State<OrchestratorState>,
+    Path(job_id): Path<Uuid>,
+    Json(payload): Json<JobEventPayload>,
+) -> Result<StatusCode, StatusCode> {
+    tracing::debug!(
+        job_id = %job_id,
+        event_type = %payload.event_type,
+        "Job event received"
+    );
+
+    // Persist to DB (fire-and-forget)
+    if let Some(ref store) = state.store {
+        let store = Arc::clone(store);
+        let event_type = payload.event_type.clone();
+        let data = payload.data.clone();
+        tokio::spawn(async move {
+            if let Err(e) = store.save_job_event(job_id, &event_type, &data).await {
+                tracing::warn!(job_id = %job_id, "Failed to persist job event: {}", e);
+            }
+        });
+    }
+
+    // Convert to SSE event and broadcast
+    let job_id_str = job_id.to_string();
+    let sse_event = match payload.event_type.as_str() {
+        "message" => SseEvent::JobMessage {
+            job_id: job_id_str,
+            role: payload
+                .data
+                .get("role")
+                .and_then(|v| v.as_str())
+                .unwrap_or("assistant")
+                .to_string(),
+            content: payload
+                .data
+                .get("content")
+                .and_then(|v| v.as_str())
+                .unwrap_or("")
+                .to_string(),
+        },
+        "tool_use" => SseEvent::JobToolUse {
+            job_id: job_id_str,
+            tool_name: payload
+                .data
+                .get("tool_name")
+                .and_then(|v| v.as_str())
+                .unwrap_or("unknown")
+                .to_string(),
+            input: payload
+                .data
+                .get("input")
+                .cloned()
+                .unwrap_or(serde_json::Value::Null),
+        },
+        "tool_result" => SseEvent::JobToolResult {
+            job_id: job_id_str,
+            tool_name: payload
+                .data
+                .get("tool_name")
+                .and_then(|v| v.as_str())
+                .unwrap_or("unknown")
+                .to_string(),
+            output: payload
+                .data
+                .get("output")
+                .and_then(|v| v.as_str())
+                .unwrap_or("")
+                .to_string(),
+        },
+        "result" => SseEvent::JobResult {
+            job_id: job_id_str,
+            status: payload
+                .data
+                .get("status")
+                .and_then(|v| v.as_str())
+                .unwrap_or("unknown")
+                .to_string(),
+            session_id: payload
+                .data
+                .get("session_id")
+                .and_then(|v| v.as_str())
+                .map(|s| s.to_string()),
+        },
+        _ => SseEvent::JobStatus {
+            job_id: job_id_str,
+            message: payload
+                .data
+                .get("message")
+                .and_then(|v| v.as_str())
+                .unwrap_or("")
+                .to_string(),
+        },
+    };
+
+    // Broadcast via the channel (if configured)
+    if let Some(ref tx) = state.job_event_tx {
+        let _ = tx.send((job_id, sse_event));
+    }
+
+    Ok(StatusCode::OK)
+}
+
+/// Return the next queued follow-up prompt for a Claude Code bridge.
+/// Returns 204 No Content if no prompt is available.
+async fn get_prompt_handler(
+    State(state): State<OrchestratorState>,
+    Path(job_id): Path<Uuid>,
+) -> Result<(StatusCode, Json<serde_json::Value>), StatusCode> {
+    let mut queue = state.prompt_queue.lock().await;
+    if let Some(prompts) = queue.get_mut(&job_id) {
+        if let Some(prompt) = prompts.pop_front() {
+            return Ok((
+                StatusCode::OK,
+                Json(serde_json::json!({
+                    "content": prompt.content,
+                    "done": prompt.done,
+                })),
+            ));
+        }
+    }
+
+    // Return 204 with an empty body. The Json wrapper requires some value
+    // but the status code signals "nothing here".
+    Ok((StatusCode::NO_CONTENT, Json(serde_json::Value::Null)))
+}
+
+fn format_finish_reason(reason: crate::llm::FinishReason) -> String {
+    match reason {
+        crate::llm::FinishReason::Stop => "stop".to_string(),
+        crate::llm::FinishReason::Length => "length".to_string(),
+        crate::llm::FinishReason::ToolUse => "tool_use".to_string(),
+        crate::llm::FinishReason::ContentFilter => "content_filter".to_string(),
+        crate::llm::FinishReason::Unknown => "unknown".to_string(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::HashMap;
+
+    use axum::body::Body;
+    use axum::http::Request;
+    use tower::ServiceExt;
+    use uuid::Uuid;
+
+    use crate::error::LlmError;
+    use crate::llm::{
+        CompletionRequest, CompletionResponse, ToolCompletionRequest, ToolCompletionResponse,
+    };
+    use crate::orchestrator::auth::TokenStore;
+    use crate::orchestrator::job_manager::{ContainerJobConfig, ContainerJobManager};
+
+    use super::*;
+
+    /// Stub LLM provider that panics if called (tests only exercise routing/auth).
+    struct StubLlm;
+
+    #[async_trait::async_trait]
+    impl crate::llm::LlmProvider for StubLlm {
+        fn model_name(&self) -> &str {
+            "stub"
+        }
+        fn cost_per_token(&self) -> (rust_decimal::Decimal, rust_decimal::Decimal) {
+            (rust_decimal::Decimal::ZERO, rust_decimal::Decimal::ZERO)
+        }
+        async fn complete(&self, _req: CompletionRequest) -> Result<CompletionResponse, LlmError> {
+            Err(LlmError::RequestFailed {
+                provider: "stub".into(),
+                reason: "not implemented".into(),
+            })
+        }
+        async fn complete_with_tools(
+            &self,
+            _req: ToolCompletionRequest,
+        ) -> Result<ToolCompletionResponse, LlmError> {
+            Err(LlmError::RequestFailed {
+                provider: "stub".into(),
+                reason: "not implemented".into(),
+            })
+        }
+    }
+
+    fn test_state() -> OrchestratorState {
+        let token_store = TokenStore::new();
+        let jm = ContainerJobManager::new(ContainerJobConfig::default(), token_store.clone());
+        OrchestratorState {
+            llm: Arc::new(StubLlm),
+            job_manager: Arc::new(jm),
+            token_store,
+            job_event_tx: None,
+            prompt_queue: Arc::new(Mutex::new(HashMap::new())),
+            store: None,
+        }
+    }
+
+    #[tokio::test]
+    async fn health_requires_no_auth() {
+        let state = test_state();
+        let router = OrchestratorApi::router(state);
+
+        let req = Request::builder()
+            .uri("/health")
+            .body(Body::empty())
+            .unwrap();
+
+        let resp = router.oneshot(req).await.unwrap();
+        assert_eq!(resp.status(), StatusCode::OK);
+    }
+
+    #[tokio::test]
+    async fn worker_route_rejects_missing_token() {
+        let state = test_state();
+        let router = OrchestratorApi::router(state);
+
+        let job_id = Uuid::new_v4();
+        let req = Request::builder()
+            .uri(format!("/worker/{}/job", job_id))
+            .body(Body::empty())
+            .unwrap();
+
+        let resp = router.oneshot(req).await.unwrap();
+        assert_eq!(resp.status(), StatusCode::UNAUTHORIZED);
+    }
+
+    #[tokio::test]
+    async fn worker_route_rejects_wrong_token() {
+        let state = test_state();
+        let router = OrchestratorApi::router(state);
+
+        let job_id = Uuid::new_v4();
+        let req = Request::builder()
+            .uri(format!("/worker/{}/job", job_id))
+            .header("Authorization", "Bearer totally-bogus")
+            .body(Body::empty())
+            .unwrap();
+
+        let resp = router.oneshot(req).await.unwrap();
+        assert_eq!(resp.status(), StatusCode::UNAUTHORIZED);
+    }
+
+    #[tokio::test]
+    async fn worker_route_accepts_valid_token() {
+        let state = test_state();
+        let job_id = Uuid::new_v4();
+        let token = state.token_store.create_token(job_id).await;
+
+        let router = OrchestratorApi::router(state);
+
+        let req = Request::builder()
+            .uri(format!("/worker/{}/job", job_id))
+            .header("Authorization", format!("Bearer {}", token))
+            .body(Body::empty())
+            .unwrap();
+
+        let resp = router.oneshot(req).await.unwrap();
+        // 404 because no container exists for this job_id, but NOT 401.
+        assert_eq!(resp.status(), StatusCode::NOT_FOUND);
+    }
+
+    #[tokio::test]
+    async fn token_for_job_a_rejected_on_job_b() {
+        let state = test_state();
+        let job_a = Uuid::new_v4();
+        let job_b = Uuid::new_v4();
+        let token_a = state.token_store.create_token(job_a).await;
+
+        let router = OrchestratorApi::router(state);
+
+        // Use job_a's token to hit job_b's endpoint
+        let req = Request::builder()
+            .uri(format!("/worker/{}/job", job_b))
+            .header("Authorization", format!("Bearer {}", token_a))
+            .body(Body::empty())
+            .unwrap();
+
+        let resp = router.oneshot(req).await.unwrap();
+        assert_eq!(resp.status(), StatusCode::UNAUTHORIZED);
+    }
+}

src/orchestrator/auth.rs

@@ -0,0 +1,162 @@
+//! Per-job bearer token authentication for worker-to-orchestrator communication.
+//!
+//! Security properties:
+//! - Tokens are cryptographically random (32 bytes, hex-encoded)
+//! - Tokens are scoped to a specific job_id
+//! - Tokens are ephemeral (in-memory only, never persisted)
+//! - A token for Job A cannot access endpoints for Job B
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use axum::extract::{Request, State};
+use axum::http::StatusCode;
+use axum::middleware::Next;
+use axum::response::Response;
+use rand::Rng;
+use tokio::sync::RwLock;
+use uuid::Uuid;
+
+/// In-memory store for per-job authentication tokens.
+#[derive(Clone)]
+pub struct TokenStore {
+    /// Maps job_id -> bearer token. Never logged or persisted.
+    tokens: Arc<RwLock<HashMap<Uuid, String>>>,
+}
+
+impl TokenStore {
+    pub fn new() -> Self {
+        Self {
+            tokens: Arc::new(RwLock::new(HashMap::new())),
+        }
+    }
+
+    /// Generate and store a new token for a job.
+    pub async fn create_token(&self, job_id: Uuid) -> String {
+        let token = generate_token();
+        self.tokens.write().await.insert(job_id, token.clone());
+        token
+    }
+
+    /// Validate a token for a specific job.
+    pub async fn validate(&self, job_id: Uuid, token: &str) -> bool {
+        self.tokens
+            .read()
+            .await
+            .get(&job_id)
+            .map(|stored| stored == token)
+            .unwrap_or(false)
+    }
+
+    /// Remove a token (on container cleanup).
+    pub async fn revoke(&self, job_id: Uuid) {
+        self.tokens.write().await.remove(&job_id);
+    }
+
+    /// Get the number of active tokens (for diagnostics).
+    pub async fn active_count(&self) -> usize {
+        self.tokens.read().await.len()
+    }
+}
+
+impl Default for TokenStore {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+/// Generate a cryptographically random token (32 bytes, hex-encoded = 64 chars).
+fn generate_token() -> String {
+    let mut bytes = [0u8; 32];
+    rand::thread_rng().fill(&mut bytes);
+    hex_encode(&bytes)
+}
+
+fn hex_encode(bytes: &[u8]) -> String {
+    bytes.iter().map(|b| format!("{:02x}", b)).collect()
+}
+
+/// Axum middleware that validates worker bearer tokens.
+///
+/// Extracts the job_id from the path (`/worker/{job_id}/...`) and validates
+/// the `Authorization: Bearer <token>` header against the token store.
+///
+/// Wire up with `axum::middleware::from_fn_with_state(token_store, worker_auth_middleware)`.
+pub async fn worker_auth_middleware(
+    State(token_store): State<TokenStore>,
+    request: Request,
+    next: Next,
+) -> Result<Response, StatusCode> {
+    let path = request.uri().path().to_string();
+    let job_id = extract_job_id_from_path(&path).ok_or(StatusCode::BAD_REQUEST)?;
+
+    let token = request
+        .headers()
+        .get("authorization")
+        .and_then(|v| v.to_str().ok())
+        .and_then(|v| v.strip_prefix("Bearer "))
+        .ok_or(StatusCode::UNAUTHORIZED)?;
+
+    if !token_store.validate(job_id, token).await {
+        return Err(StatusCode::UNAUTHORIZED);
+    }
+
+    Ok(next.run(request).await)
+}
+
+/// Extract job UUID from a path like `/worker/{uuid}/...`
+fn extract_job_id_from_path(path: &str) -> Option<Uuid> {
+    let parts: Vec<&str> = path.trim_start_matches('/').split('/').collect();
+    if parts.len() >= 2 && parts[0] == "worker" {
+        Uuid::parse_str(parts[1]).ok()
+    } else {
+        None
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn test_token_create_and_validate() {
+        let store = TokenStore::new();
+        let job_id = Uuid::new_v4();
+
+        let token = store.create_token(job_id).await;
+        assert_eq!(token.len(), 64); // 32 bytes hex = 64 chars
+
+        assert!(store.validate(job_id, &token).await);
+        assert!(!store.validate(job_id, "wrong-token").await);
+        assert!(!store.validate(Uuid::new_v4(), &token).await);
+    }
+
+    #[tokio::test]
+    async fn test_token_revoke() {
+        let store = TokenStore::new();
+        let job_id = Uuid::new_v4();
+
+        let token = store.create_token(job_id).await;
+        assert!(store.validate(job_id, &token).await);
+
+        store.revoke(job_id).await;
+        assert!(!store.validate(job_id, &token).await);
+    }
+
+    #[test]
+    fn test_extract_job_id() {
+        let id = Uuid::new_v4();
+        let path = format!("/worker/{}/llm/complete", id);
+        assert_eq!(extract_job_id_from_path(&path), Some(id));
+
+        assert_eq!(extract_job_id_from_path("/other/path"), None);
+        assert_eq!(extract_job_id_from_path("/worker/not-a-uuid/foo"), None);
+    }
+
+    #[test]
+    fn test_token_is_random() {
+        let t1 = generate_token();
+        let t2 = generate_token();
+        assert_ne!(t1, t2);
+    }
+}

src/orchestrator/job_manager.rs

@@ -0,0 +1,474 @@
+//! Container lifecycle management for sandboxed jobs.
+//!
+//! Extends the existing `SandboxManager` infrastructure to support persistent
+//! containers with their own agent loops (as opposed to ephemeral per-command containers).
+
+use std::collections::HashMap;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use chrono::{DateTime, Utc};
+use tokio::sync::RwLock;
+use uuid::Uuid;
+
+use crate::error::OrchestratorError;
+use crate::orchestrator::auth::TokenStore;
+use crate::sandbox::connect_docker;
+
+/// Which mode a sandbox container runs in.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum JobMode {
+    /// Standard IronClaw worker with proxied LLM calls.
+    Worker,
+    /// Claude Code bridge that spawns the `claude` CLI directly.
+    ClaudeCode,
+}
+
+impl JobMode {
+    pub fn as_str(&self) -> &'static str {
+        match self {
+            Self::Worker => "worker",
+            Self::ClaudeCode => "claude_code",
+        }
+    }
+}
+
+impl std::fmt::Display for JobMode {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.as_str())
+    }
+}
+
+/// Configuration for the container job manager.
+#[derive(Debug, Clone)]
+pub struct ContainerJobConfig {
+    /// Docker image for worker containers.
+    pub image: String,
+    /// Default memory limit in MB.
+    pub memory_limit_mb: u64,
+    /// Default CPU shares.
+    pub cpu_shares: u32,
+    /// Port the orchestrator internal API listens on.
+    pub orchestrator_port: u16,
+    /// Host directory containing Claude auth config (mounted read-only for ClaudeCode mode).
+    pub claude_config_dir: Option<PathBuf>,
+    /// Claude model to use in ClaudeCode mode.
+    pub claude_code_model: String,
+    /// Maximum turns for Claude Code.
+    pub claude_code_max_turns: u32,
+    /// Memory limit in MB for Claude Code containers (heavier than workers).
+    pub claude_code_memory_limit_mb: u64,
+}
+
+impl Default for ContainerJobConfig {
+    fn default() -> Self {
+        Self {
+            image: "ironclaw-worker:latest".to_string(),
+            memory_limit_mb: 2048,
+            cpu_shares: 1024,
+            orchestrator_port: 50051,
+            claude_config_dir: None,
+            claude_code_model: "sonnet".to_string(),
+            claude_code_max_turns: 50,
+            claude_code_memory_limit_mb: 4096,
+        }
+    }
+}
+
+/// State of a container.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum ContainerState {
+    Creating,
+    Running,
+    Stopped,
+    Failed,
+}
+
+impl std::fmt::Display for ContainerState {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::Creating => write!(f, "creating"),
+            Self::Running => write!(f, "running"),
+            Self::Stopped => write!(f, "stopped"),
+            Self::Failed => write!(f, "failed"),
+        }
+    }
+}
+
+/// Handle to a running container job.
+#[derive(Debug, Clone)]
+pub struct ContainerHandle {
+    pub job_id: Uuid,
+    pub container_id: String,
+    pub state: ContainerState,
+    pub mode: JobMode,
+    pub created_at: DateTime<Utc>,
+    pub project_dir: Option<PathBuf>,
+    pub task_description: String,
+    /// Completion result from the worker (set when the worker reports done).
+    pub completion_result: Option<CompletionResult>,
+    // NOTE: auth_token is intentionally NOT in this struct.
+    // It lives only in the TokenStore (never logged, serialized, or persisted).
+}
+
+/// Result reported by a worker on completion.
+#[derive(Debug, Clone)]
+pub struct CompletionResult {
+    pub success: bool,
+    pub message: Option<String>,
+}
+
+/// Manages the lifecycle of Docker containers for sandboxed job execution.
+pub struct ContainerJobManager {
+    config: ContainerJobConfig,
+    token_store: TokenStore,
+    containers: Arc<RwLock<HashMap<Uuid, ContainerHandle>>>,
+}
+
+impl ContainerJobManager {
+    pub fn new(config: ContainerJobConfig, token_store: TokenStore) -> Self {
+        Self {
+            config,
+            token_store,
+            containers: Arc::new(RwLock::new(HashMap::new())),
+        }
+    }
+
+    /// Create and start a new container for a job.
+    ///
+    /// The caller provides the `job_id` so it can be persisted to the database
+    /// before the container is created. Returns the auth token for the worker.
+    pub async fn create_job(
+        &self,
+        job_id: Uuid,
+        task: &str,
+        project_dir: Option<PathBuf>,
+        mode: JobMode,
+    ) -> Result<String, OrchestratorError> {
+        // Generate auth token (stored in TokenStore, never logged)
+        let token = self.token_store.create_token(job_id).await;
+
+        // Record the handle
+        let handle = ContainerHandle {
+            job_id,
+            container_id: String::new(), // set after container creation
+            state: ContainerState::Creating,
+            mode,
+            created_at: Utc::now(),
+            project_dir: project_dir.clone(),
+            task_description: task.to_string(),
+            completion_result: None,
+        };
+        self.containers.write().await.insert(job_id, handle);
+
+        // Connect to Docker
+        let docker = connect_docker()
+            .await
+            .map_err(|e| OrchestratorError::Docker {
+                reason: e.to_string(),
+            })?;
+
+        // Build container configuration
+        let orchestrator_host = if cfg!(target_os = "linux") {
+            "172.17.0.1"
+        } else {
+            "host.docker.internal"
+        };
+
+        let orchestrator_url = format!(
+            "http://{}:{}",
+            orchestrator_host, self.config.orchestrator_port
+        );
+
+        let mut env_vec = vec![
+            format!("IRONCLAW_WORKER_TOKEN={}", token),
+            format!("IRONCLAW_JOB_ID={}", job_id),
+            format!("IRONCLAW_ORCHESTRATOR_URL={}", orchestrator_url),
+        ];
+
+        // Build volume mounts (validate project_dir stays within ~/.ironclaw/projects/)
+        let mut binds = Vec::new();
+        if let Some(ref dir) = project_dir {
+            let canonical =
+                dir.canonicalize()
+                    .map_err(|e| OrchestratorError::ContainerCreationFailed {
+                        job_id,
+                        reason: format!(
+                            "failed to canonicalize project dir {}: {}",
+                            dir.display(),
+                            e
+                        ),
+                    })?;
+            let projects_base = dirs::home_dir()
+                .unwrap_or_else(|| PathBuf::from("."))
+                .join(".ironclaw")
+                .join("projects");
+            if let Ok(canonical_base) = projects_base.canonicalize() {
+                if !canonical.starts_with(&canonical_base) {
+                    return Err(OrchestratorError::ContainerCreationFailed {
+                        job_id,
+                        reason: format!(
+                            "project directory {} is outside allowed base {}",
+                            canonical.display(),
+                            canonical_base.display()
+                        ),
+                    });
+                }
+            }
+            binds.push(format!("{}:/workspace:rw", canonical.display()));
+            env_vec.push("IRONCLAW_WORKSPACE=/workspace".to_string());
+        }
+
+        // Claude Code mode: mount host ~/.claude read-only for auth
+        if mode == JobMode::ClaudeCode {
+            if let Some(ref claude_dir) = self.config.claude_config_dir {
+                binds.push(format!("{}:/home/sandbox/.claude:ro", claude_dir.display()));
+            }
+        }
+
+        // Memory limit: Claude Code gets more memory
+        let memory_mb = match mode {
+            JobMode::ClaudeCode => self.config.claude_code_memory_limit_mb,
+            JobMode::Worker => self.config.memory_limit_mb,
+        };
+
+        // Create the container
+        use bollard::container::{Config, CreateContainerOptions};
+        use bollard::models::HostConfig;
+
+        let host_config = HostConfig {
+            binds: if binds.is_empty() { None } else { Some(binds) },
+            memory: Some((memory_mb * 1024 * 1024) as i64),
+            cpu_shares: Some(self.config.cpu_shares as i64),
+            network_mode: Some("bridge".to_string()),
+            extra_hosts: Some(vec!["host.docker.internal:host-gateway".to_string()]),
+            cap_drop: Some(vec!["ALL".to_string()]),
+            cap_add: Some(vec![
+                "CHOWN".to_string(),
+                "SETUID".to_string(),
+                "SETGID".to_string(),
+            ]),
+            security_opt: Some(vec!["no-new-privileges:true".to_string()]),
+            tmpfs: Some(
+                [("/tmp".to_string(), "size=512M".to_string())]
+                    .into_iter()
+                    .collect(),
+            ),
+            ..Default::default()
+        };
+
+        // Build CMD based on mode
+        let cmd = match mode {
+            JobMode::Worker => vec![
+                "worker".to_string(),
+                "--job-id".to_string(),
+                job_id.to_string(),
+                "--orchestrator-url".to_string(),
+                orchestrator_url,
+            ],
+            JobMode::ClaudeCode => vec![
+                "claude-bridge".to_string(),
+                "--job-id".to_string(),
+                job_id.to_string(),
+                "--orchestrator-url".to_string(),
+                orchestrator_url,
+                "--max-turns".to_string(),
+                self.config.claude_code_max_turns.to_string(),
+                "--model".to_string(),
+                self.config.claude_code_model.clone(),
+            ],
+        };
+
+        let container_config = Config {
+            image: Some(self.config.image.clone()),
+            cmd: Some(cmd),
+            env: Some(env_vec),
+            host_config: Some(host_config),
+            user: Some("1000:1000".to_string()),
+            working_dir: Some("/workspace".to_string()),
+            ..Default::default()
+        };
+
+        let container_name = match mode {
+            JobMode::Worker => format!("ironclaw-worker-{}", job_id),
+            JobMode::ClaudeCode => format!("ironclaw-claude-{}", job_id),
+        };
+        let options = CreateContainerOptions {
+            name: container_name,
+            ..Default::default()
+        };
+
+        let response = docker
+            .create_container(Some(options), container_config)
+            .await
+            .map_err(|e| OrchestratorError::ContainerCreationFailed {
+                job_id,
+                reason: e.to_string(),
+            })?;
+
+        let container_id = response.id;
+
+        // Start the container
+        docker
+            .start_container::<String>(&container_id, None)
+            .await
+            .map_err(|e| OrchestratorError::ContainerCreationFailed {
+                job_id,
+                reason: format!("failed to start container: {}", e),
+            })?;
+
+        // Update handle with container ID
+        if let Some(handle) = self.containers.write().await.get_mut(&job_id) {
+            handle.container_id = container_id;
+            handle.state = ContainerState::Running;
+        }
+
+        tracing::info!(
+            job_id = %job_id,
+            "Created and started worker container"
+        );
+
+        Ok(token)
+    }
+
+    /// Stop a running container job.
+    pub async fn stop_job(&self, job_id: Uuid) -> Result<(), OrchestratorError> {
+        let container_id = {
+            let containers = self.containers.read().await;
+            containers
+                .get(&job_id)
+                .map(|h| h.container_id.clone())
+                .ok_or(OrchestratorError::ContainerNotFound { job_id })?
+        };
+
+        if container_id.is_empty() {
+            return Err(OrchestratorError::InvalidContainerState {
+                job_id,
+                state: "creating (no container ID yet)".to_string(),
+            });
+        }
+
+        let docker = connect_docker()
+            .await
+            .map_err(|e| OrchestratorError::Docker {
+                reason: e.to_string(),
+            })?;
+
+        // Stop the container (10 second grace period)
+        let _ = docker
+            .stop_container(
+                &container_id,
+                Some(bollard::container::StopContainerOptions { t: 10 }),
+            )
+            .await;
+
+        // Remove the container
+        let _ = docker
+            .remove_container(
+                &container_id,
+                Some(bollard::container::RemoveContainerOptions {
+                    force: true,
+                    ..Default::default()
+                }),
+            )
+            .await;
+
+        // Update state
+        if let Some(handle) = self.containers.write().await.get_mut(&job_id) {
+            handle.state = ContainerState::Stopped;
+        }
+
+        // Revoke the auth token
+        self.token_store.revoke(job_id).await;
+
+        tracing::info!(job_id = %job_id, "Stopped worker container");
+
+        Ok(())
+    }
+
+    /// Mark a job as complete with a result. The container is stopped but the
+    /// handle is kept so `CreateJobTool` can read the completion message.
+    pub async fn complete_job(
+        &self,
+        job_id: Uuid,
+        result: CompletionResult,
+    ) -> Result<(), OrchestratorError> {
+        // Store the result before stopping
+        {
+            let mut containers = self.containers.write().await;
+            if let Some(handle) = containers.get_mut(&job_id) {
+                handle.completion_result = Some(result);
+                handle.state = ContainerState::Stopped;
+            }
+        }
+
+        // Stop container and revoke token (but keep handle in map)
+        let container_id = {
+            let containers = self.containers.read().await;
+            containers.get(&job_id).map(|h| h.container_id.clone())
+        };
+        if let Some(cid) = container_id {
+            if !cid.is_empty() {
+                if let Ok(docker) = connect_docker().await {
+                    let _ = docker
+                        .stop_container(
+                            &cid,
+                            Some(bollard::container::StopContainerOptions { t: 5 }),
+                        )
+                        .await;
+                    let _ = docker
+                        .remove_container(
+                            &cid,
+                            Some(bollard::container::RemoveContainerOptions {
+                                force: true,
+                                ..Default::default()
+                            }),
+                        )
+                        .await;
+                }
+            }
+        }
+        self.token_store.revoke(job_id).await;
+
+        tracing::info!(job_id = %job_id, "Completed worker container");
+        Ok(())
+    }
+
+    /// Remove a completed job handle from memory (called after result is read).
+    pub async fn cleanup_job(&self, job_id: Uuid) {
+        self.containers.write().await.remove(&job_id);
+    }
+
+    /// Get the handle for a job.
+    pub async fn get_handle(&self, job_id: Uuid) -> Option<ContainerHandle> {
+        self.containers.read().await.get(&job_id).cloned()
+    }
+
+    /// List all active container jobs.
+    pub async fn list_jobs(&self) -> Vec<ContainerHandle> {
+        self.containers.read().await.values().cloned().collect()
+    }
+
+    /// Get a reference to the token store.
+    pub fn token_store(&self) -> &TokenStore {
+        &self.token_store
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_container_job_config_default() {
+        let config = ContainerJobConfig::default();
+        assert_eq!(config.orchestrator_port, 50051);
+        assert_eq!(config.memory_limit_mb, 2048);
+    }
+
+    #[test]
+    fn test_container_state_display() {
+        assert_eq!(ContainerState::Running.to_string(), "running");
+        assert_eq!(ContainerState::Stopped.to_string(), "stopped");
+    }
+}

src/orchestrator/mod.rs

@@ -0,0 +1,37 @@
+//! Orchestrator for managing sandboxed worker containers.
+//!
+//! The orchestrator runs in the main agent process and provides:
+//! - An internal HTTP API for worker communication (LLM proxy, status, secrets)
+//! - Per-job bearer token authentication
+//! - Container lifecycle management (create, monitor, stop)
+//!
+//! ```text
+//! ┌───────────────────────────────────────────────┐
+//! │              Orchestrator                       │
+//! │                                                 │
+//! │  Internal API (:50051)                          │
+//! │    POST /worker/{id}/llm/complete               │
+//! │    POST /worker/{id}/llm/complete_with_tools    │
+//! │    GET  /worker/{id}/job                        │
+//! │    POST /worker/{id}/status                     │
+//! │    POST /worker/{id}/complete                   │
+//! │                                                 │
+//! │  ContainerJobManager                            │
+//! │    create_job() -> container + token             │
+//! │    stop_job()                                    │
+//! │    list_jobs()                                   │
+//! │                                                 │
+//! │  TokenStore                                     │
+//! │    per-job bearer tokens (in-memory only)       │
+//! └───────────────────────────────────────────────┘
+//! ```
+
+pub mod api;
+pub mod auth;
+pub mod job_manager;
+
+pub use api::OrchestratorApi;
+pub use auth::TokenStore;
+pub use job_manager::{
+    CompletionResult, ContainerHandle, ContainerJobConfig, ContainerJobManager, JobMode,
+};

src/pairing/mod.rs

@@ -0,0 +1,10 @@
+//! DM pairing for channels.
+//!
+//! Gates DMs from unknown senders. Only approved senders can message the agent.
+//! Unknown senders receive a pairing code and must be approved via `ironclaw pairing approve`.
+//!
+//! OpenClaw reference: src/pairing/pairing-store.ts
+
+mod store;
+
+pub use store::{PairingRequest, PairingStore, PairingStoreError};

src/pairing/store.rs

@@ -0,0 +1,699 @@
+//! Pairing store: pending requests and allowFrom list.
+//!
+//! Stored in ~/.ironclaw/{channel}-pairing.json and {channel}-allowFrom.json.
+
+use std::collections::HashSet;
+use std::fs;
+use std::io::{Seek, SeekFrom, Write};
+use std::path::{Path, PathBuf};
+use std::time::{SystemTime, UNIX_EPOCH};
+
+use fs4::FileExt;
+use rand::Rng;
+use serde::{Deserialize, Serialize};
+
+const PAIRING_CODE_LENGTH: usize = 8;
+const PAIRING_ALPHABET: &[u8] = b"ABCDEFGHJKLMNPQRSTUVWXYZ23456789";
+/// TTL for pending pairing requests (minutes, not hours — reduces brute-force window).
+const PAIRING_PENDING_TTL_SECS: u64 = 15 * 60;
+const PAIRING_PENDING_MAX: usize = 3;
+/// Max failed approve attempts per channel before rate limit kicks in.
+const PAIRING_APPROVE_RATE_LIMIT: usize = 10;
+/// Time window for rate limit (seconds).
+const PAIRING_APPROVE_RATE_WINDOW_SECS: u64 = 5 * 60;
+
+/// Error from pairing store operations.
+#[derive(Debug, thiserror::Error)]
+pub enum PairingStoreError {
+    #[error("Invalid channel: {0}")]
+    InvalidChannel(String),
+
+    #[error("IO error: {0}")]
+    Io(#[from] std::io::Error),
+
+    #[error("JSON error: {0}")]
+    Json(#[from] serde_json::Error),
+
+    #[error("Rate limit: too many failed approve attempts; try again later")]
+    ApproveRateLimited,
+}
+
+/// Result of upserting a pairing request.
+#[derive(Debug)]
+pub struct UpsertResult {
+    pub code: String,
+    pub created: bool,
+}
+
+/// A pending pairing request.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PairingRequest {
+    pub id: String,
+    pub code: String,
+    pub created_at: String,
+    pub last_seen_at: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub meta: Option<serde_json::Value>,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+struct PairingStoreFile {
+    version: u8,
+    requests: Vec<PairingRequest>,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+struct AllowFromStoreFile {
+    version: u8,
+    #[serde(rename = "allowFrom")]
+    allow_from: Vec<String>,
+}
+
+fn default_pairing_dir() -> PathBuf {
+    dirs::home_dir()
+        .unwrap_or_else(|| PathBuf::from("."))
+        .join(".ironclaw")
+}
+
+fn safe_channel_key(channel: &str) -> Result<String, PairingStoreError> {
+    let raw = channel.trim().to_lowercase();
+    if raw.is_empty() {
+        return Err(PairingStoreError::InvalidChannel("empty".to_string()));
+    }
+    let safe = raw
+        .chars()
+        .map(|c| match c {
+            '\\' | '/' | ':' | '*' | '?' | '"' | '<' | '>' | '|' => '_',
+            _ => c,
+        })
+        .collect::<String>()
+        .replace("..", "_");
+    if safe.is_empty() || safe == "_" {
+        return Err(PairingStoreError::InvalidChannel(channel.to_string()));
+    }
+    Ok(safe)
+}
+
+fn pairing_path(base_dir: &Path, channel: &str) -> Result<PathBuf, PairingStoreError> {
+    let key = safe_channel_key(channel)?;
+    Ok(base_dir.join(format!("{}-pairing.json", key)))
+}
+
+fn allow_from_path(base_dir: &Path, channel: &str) -> Result<PathBuf, PairingStoreError> {
+    let key = safe_channel_key(channel)?;
+    Ok(base_dir.join(format!("{}-allowFrom.json", key)))
+}
+
+fn approve_attempts_path(base_dir: &Path, channel: &str) -> Result<PathBuf, PairingStoreError> {
+    let key = safe_channel_key(channel)?;
+    Ok(base_dir.join(format!("{}-approve-attempts.json", key)))
+}
+
+#[derive(Debug, Default, Serialize, Deserialize)]
+struct ApproveAttemptsFile {
+    failed_at: Vec<u64>,
+}
+
+fn now_iso() -> String {
+    let now = SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .unwrap_or_default();
+    #[allow(clippy::cast_possible_wrap)]
+    chrono::DateTime::from_timestamp(now.as_secs() as i64, 0)
+        .map(|dt| dt.to_rfc3339())
+        .unwrap_or_else(|| now.as_secs().to_string())
+}
+
+fn now_secs() -> u64 {
+    SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .map(|d| d.as_secs())
+        .unwrap_or(0)
+}
+
+fn parse_timestamp(value: &str) -> Option<u64> {
+    chrono::DateTime::parse_from_rfc3339(value)
+        .ok()
+        .map(|dt| dt.timestamp() as u64)
+        .or_else(|| value.parse::<u64>().ok())
+}
+
+fn is_expired(req: &PairingRequest, now_secs: u64) -> bool {
+    let created = parse_timestamp(&req.created_at).unwrap_or(0);
+    now_secs.saturating_sub(created) > PAIRING_PENDING_TTL_SECS
+}
+
+fn random_code() -> String {
+    let mut rng = rand::thread_rng();
+    (0..PAIRING_CODE_LENGTH)
+        .map(|_| {
+            let idx = rng.gen_range(0..PAIRING_ALPHABET.len());
+            PAIRING_ALPHABET[idx] as char
+        })
+        .collect()
+}
+
+fn generate_unique_code(existing: &HashSet<String>) -> String {
+    let mut rng = rand::thread_rng();
+    for _ in 0..500 {
+        let code = random_code();
+        if !existing.contains(&code) {
+            return code;
+        }
+    }
+    // Fallback: add suffix
+    format!("{}{:04}", random_code(), rng.gen_range(0..10000))
+}
+
+/// Pairing store for a channel.
+#[derive(Debug, Clone)]
+pub struct PairingStore {
+    base_dir: PathBuf,
+}
+
+impl PairingStore {
+    /// Create a new pairing store using default directory (~/.ironclaw).
+    pub fn new() -> Self {
+        Self {
+            base_dir: default_pairing_dir(),
+        }
+    }
+
+    /// Create a pairing store with a custom base directory (for testing).
+    pub fn with_base_dir(base_dir: PathBuf) -> Self {
+        Self { base_dir }
+    }
+
+    /// List pending pairing requests for a channel.
+    pub fn list_pending(&self, channel: &str) -> Result<Vec<PairingRequest>, PairingStoreError> {
+        let path = pairing_path(&self.base_dir, channel)?;
+        let content = match fs::read_to_string(&path) {
+            Ok(c) => c,
+            Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
+                return Ok(Vec::new());
+            }
+            Err(e) => return Err(e.into()),
+        };
+
+        let file: PairingStoreFile = serde_json::from_str(&content).unwrap_or(PairingStoreFile {
+            version: 1,
+            requests: Vec::new(),
+        });
+
+        let now = now_secs();
+        let original_len = file.requests.len();
+        let mut requests: Vec<_> = file
+            .requests
+            .into_iter()
+            .filter(|r| !is_expired(r, now))
+            .collect();
+
+        if requests.len() != original_len {
+            self.write_pairing_file(channel, &requests)?;
+        }
+
+        requests.sort_by(|a, b| a.created_at.cmp(&b.created_at));
+        Ok(requests)
+    }
+
+    /// Upsert a pairing request. Returns (code, created).
+    pub fn upsert_request(
+        &self,
+        channel: &str,
+        id: &str,
+        meta: Option<serde_json::Value>,
+    ) -> Result<UpsertResult, PairingStoreError> {
+        let path = pairing_path(&self.base_dir, channel)?;
+        fs::create_dir_all(path.parent().unwrap())?;
+
+        let mut file = fs::OpenOptions::new()
+            .read(true)
+            .write(true)
+            .create(true)
+            .truncate(false)
+            .open(&path)?;
+
+        file.lock_exclusive()?;
+
+        let content = fs::read_to_string(&path).unwrap_or_default();
+        let mut store: PairingStoreFile =
+            serde_json::from_str(&content).unwrap_or(PairingStoreFile {
+                version: 1,
+                requests: Vec::new(),
+            });
+
+        let now = now_iso();
+        let now_secs = now_secs();
+        let id = id.trim().to_string();
+        if id.is_empty() {
+            fs4::FileExt::unlock(&file)?;
+            return Err(PairingStoreError::InvalidChannel("empty id".to_string()));
+        }
+
+        store.requests.retain(|r| !is_expired(r, now_secs));
+        let existing_codes: HashSet<String> = store
+            .requests
+            .iter()
+            .map(|r| r.code.to_uppercase())
+            .collect();
+
+        if let Some(idx) = store.requests.iter().position(|r| r.id == id) {
+            let req = &mut store.requests[idx];
+            let code = if req.code.is_empty() {
+                generate_unique_code(&existing_codes)
+            } else {
+                req.code.clone()
+            };
+            req.last_seen_at = now.clone();
+            req.code = code.clone();
+            if let Some(m) = meta {
+                req.meta = Some(m);
+            }
+            self.write_pairing_file_locked(&mut file, channel, &store.requests)?;
+            fs4::FileExt::unlock(&file)?;
+            return Ok(UpsertResult {
+                code,
+                created: false,
+            });
+        }
+
+        if store.requests.len() >= PAIRING_PENDING_MAX {
+            fs4::FileExt::unlock(&file)?;
+            return Ok(UpsertResult {
+                code: String::new(),
+                created: false,
+            });
+        }
+
+        let code = generate_unique_code(&existing_codes);
+        store.requests.push(PairingRequest {
+            id: id.clone(),
+            code: code.clone(),
+            created_at: now.clone(),
+            last_seen_at: now,
+            meta,
+        });
+
+        self.write_pairing_file_locked(&mut file, channel, &store.requests)?;
+        fs4::FileExt::unlock(&file)?;
+
+        Ok(UpsertResult {
+            code,
+            created: true,
+        })
+    }
+
+    fn is_approve_rate_limited(&self, channel: &str) -> Result<bool, PairingStoreError> {
+        let path = approve_attempts_path(&self.base_dir, channel)?;
+        let content = match fs::read_to_string(&path) {
+            Ok(c) => c,
+            Err(e) if e.kind() == std::io::ErrorKind::NotFound => return Ok(false),
+            Err(e) => return Err(e.into()),
+        };
+        let mut data: ApproveAttemptsFile = serde_json::from_str(&content).unwrap_or_default();
+        let now = now_secs();
+        let cutoff = now.saturating_sub(PAIRING_APPROVE_RATE_WINDOW_SECS);
+        data.failed_at.retain(|&t| t >= cutoff);
+        Ok(data.failed_at.len() >= PAIRING_APPROVE_RATE_LIMIT)
+    }
+
+    fn record_failed_approve(&self, channel: &str) -> Result<(), PairingStoreError> {
+        let path = approve_attempts_path(&self.base_dir, channel)?;
+        fs::create_dir_all(path.parent().unwrap())?;
+        let file = fs::OpenOptions::new()
+            .read(true)
+            .write(true)
+            .create(true)
+            .truncate(true)
+            .open(&path)?;
+        file.lock_exclusive()?;
+        let content = fs::read_to_string(&path).unwrap_or_default();
+        let mut data: ApproveAttemptsFile = serde_json::from_str(&content).unwrap_or_default();
+        let now = now_secs();
+        data.failed_at.push(now);
+        let cutoff = now.saturating_sub(PAIRING_APPROVE_RATE_WINDOW_SECS);
+        data.failed_at.retain(|&t| t >= cutoff);
+        let json = serde_json::to_string_pretty(&data)?;
+        fs::write(&path, json)?;
+        fs4::FileExt::unlock(&file)?;
+        Ok(())
+    }
+
+    /// Approve a pairing code and add the sender to allowFrom.
+    pub fn approve(
+        &self,
+        channel: &str,
+        code: &str,
+    ) -> Result<Option<PairingRequest>, PairingStoreError> {
+        let code = code.trim().to_uppercase();
+        if code.is_empty() {
+            return Ok(None);
+        }
+
+        if self.is_approve_rate_limited(channel)? {
+            return Err(PairingStoreError::ApproveRateLimited);
+        }
+
+        let path = pairing_path(&self.base_dir, channel)?;
+        let mut file = fs::OpenOptions::new()
+            .read(true)
+            .write(true)
+            .create(false)
+            .open(&path)
+            .map_err(|e| {
+                if e.kind() == std::io::ErrorKind::NotFound {
+                    PairingStoreError::InvalidChannel("no pairing file".to_string())
+                } else {
+                    PairingStoreError::Io(e)
+                }
+            })?;
+
+        file.lock_exclusive()?;
+
+        let content = fs::read_to_string(&path).unwrap_or_default();
+        let mut store: PairingStoreFile =
+            serde_json::from_str(&content).unwrap_or(PairingStoreFile {
+                version: 1,
+                requests: Vec::new(),
+            });
+
+        let now_secs = now_secs();
+        store.requests.retain(|r| !is_expired(r, now_secs));
+
+        let idx = store
+            .requests
+            .iter()
+            .position(|r| r.code.to_uppercase() == code);
+
+        let entry = match idx {
+            Some(i) => store.requests.remove(i),
+            None => {
+                fs4::FileExt::unlock(&file)?;
+                self.record_failed_approve(channel)?;
+                return Ok(None);
+            }
+        };
+
+        self.write_pairing_file_locked(&mut file, channel, &store.requests)?;
+        fs4::FileExt::unlock(&file)?;
+
+        self.add_allow_from(channel, &entry.id)?;
+
+        Ok(Some(entry))
+    }
+
+    /// Read the allowFrom list for a channel.
+    pub fn read_allow_from(&self, channel: &str) -> Result<Vec<String>, PairingStoreError> {
+        let path = allow_from_path(&self.base_dir, channel)?;
+        let content = match fs::read_to_string(&path) {
+            Ok(c) => c,
+            Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
+                return Ok(Vec::new());
+            }
+            Err(e) => return Err(e.into()),
+        };
+
+        let file: AllowFromStoreFile =
+            serde_json::from_str(&content).unwrap_or(AllowFromStoreFile {
+                version: 1,
+                allow_from: Vec::new(),
+            });
+
+        Ok(file.allow_from)
+    }
+
+    /// Check if a sender is allowed (by id or username).
+    pub fn is_sender_allowed(
+        &self,
+        channel: &str,
+        id: &str,
+        username: Option<&str>,
+    ) -> Result<bool, PairingStoreError> {
+        let allow = self.read_allow_from(channel)?;
+        let id = id.trim();
+        let id_ok = allow.iter().any(|e| e.trim() == id);
+        if id_ok {
+            return Ok(true);
+        }
+        if let Some(u) = username {
+            let u = u.trim().to_lowercase();
+            let u_norm = u.strip_prefix('@').unwrap_or(&u);
+            if allow.iter().any(|e| {
+                e.trim().to_lowercase() == u || e.trim().to_lowercase() == format!("@{}", u_norm)
+            }) {
+                return Ok(true);
+            }
+        }
+        Ok(false)
+    }
+
+    fn add_allow_from(&self, channel: &str, entry: &str) -> Result<(), PairingStoreError> {
+        let entry = entry.trim().to_string();
+        if entry.is_empty() {
+            return Ok(());
+        }
+
+        let path = allow_from_path(&self.base_dir, channel)?;
+        fs::create_dir_all(path.parent().unwrap())?;
+
+        let file = fs::OpenOptions::new()
+            .read(true)
+            .write(true)
+            .create(true)
+            .truncate(true)
+            .open(&path)?;
+
+        file.lock_exclusive()?;
+
+        let content = fs::read_to_string(&path).unwrap_or_default();
+        let mut store: AllowFromStoreFile =
+            serde_json::from_str(&content).unwrap_or(AllowFromStoreFile {
+                version: 1,
+                allow_from: Vec::new(),
+            });
+
+        let normalized = entry.to_lowercase();
+        if store
+            .allow_from
+            .iter()
+            .any(|e| e.to_lowercase() == normalized)
+        {
+            fs4::FileExt::unlock(&file)?;
+            return Ok(());
+        }
+
+        store.allow_from.push(entry);
+        let json = serde_json::to_string_pretty(&store)?;
+        fs::write(&path, json)?;
+
+        fs4::FileExt::unlock(&file)?;
+        Ok(())
+    }
+
+    fn write_pairing_file(
+        &self,
+        channel: &str,
+        requests: &[PairingRequest],
+    ) -> Result<(), PairingStoreError> {
+        let path = pairing_path(&self.base_dir, channel)?;
+        let mut file = fs::OpenOptions::new()
+            .write(true)
+            .create(true)
+            .truncate(true)
+            .open(&path)?;
+        file.lock_exclusive()?;
+        self.write_pairing_file_locked(&mut file, channel, requests)?;
+        fs4::FileExt::unlock(&file)?;
+        Ok(())
+    }
+
+    fn write_pairing_file_locked(
+        &self,
+        file: &mut fs::File,
+        _channel: &str,
+        requests: &[PairingRequest],
+    ) -> Result<(), PairingStoreError> {
+        let store = PairingStoreFile {
+            version: 1,
+            requests: requests.to_vec(),
+        };
+        let json = serde_json::to_string_pretty(&store)?;
+        file.set_len(0)?;
+        file.seek(SeekFrom::Start(0))?;
+        file.write_all(json.as_bytes())?;
+        file.sync_all()?;
+        Ok(())
+    }
+}
+
+impl Default for PairingStore {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::TempDir;
+
+    #[test]
+    fn test_safe_channel_key() {
+        assert_eq!(safe_channel_key("telegram").unwrap(), "telegram");
+        assert_eq!(safe_channel_key("Telegram").unwrap(), "telegram");
+        safe_channel_key("").unwrap_err();
+    }
+
+    #[test]
+    fn test_random_code() {
+        let c = random_code();
+        assert_eq!(c.len(), PAIRING_CODE_LENGTH);
+        assert!(c.chars().all(|c| PAIRING_ALPHABET.contains(&(c as u8))));
+    }
+
+    fn test_store() -> (PairingStore, TempDir) {
+        let dir = TempDir::new().unwrap();
+        let store = PairingStore::with_base_dir(dir.path().to_path_buf());
+        (store, dir)
+    }
+
+    #[test]
+    fn test_list_pending_empty() {
+        let (store, _) = test_store();
+        let requests = store.list_pending("telegram").unwrap();
+        assert!(requests.is_empty());
+    }
+
+    #[test]
+    fn test_upsert_request_creates_new() {
+        let (store, _) = test_store();
+        let result = store
+            .upsert_request(
+                "telegram",
+                "user123",
+                Some(serde_json::json!({"chat_id": 456})),
+            )
+            .unwrap();
+        assert!(result.created);
+        assert_eq!(result.code.len(), PAIRING_CODE_LENGTH);
+        assert!(
+            result
+                .code
+                .chars()
+                .all(|c| PAIRING_ALPHABET.contains(&(c as u8)))
+        );
+    }
+
+    #[test]
+    fn test_upsert_request_updates_existing() {
+        let (store, _) = test_store();
+        let r1 = store.upsert_request("telegram", "user123", None).unwrap();
+        assert!(r1.created);
+        let r2 = store
+            .upsert_request("telegram", "user123", Some(serde_json::json!({"x": 1})))
+            .unwrap();
+        assert!(!r2.created);
+        assert_eq!(r1.code, r2.code);
+
+        let pending = store.list_pending("telegram").unwrap();
+        assert_eq!(pending.len(), 1);
+        assert_eq!(pending[0].id, "user123");
+        assert_eq!(pending[0].meta, Some(serde_json::json!({"x": 1})));
+    }
+
+    #[test]
+    fn test_approve_adds_to_allow_from() {
+        let (store, _) = test_store();
+        let r = store.upsert_request("telegram", "user456", None).unwrap();
+        assert!(r.created);
+
+        let approved = store.approve("telegram", &r.code).unwrap();
+        assert!(approved.is_some());
+        assert_eq!(approved.unwrap().id, "user456");
+
+        let allow = store.read_allow_from("telegram").unwrap();
+        assert_eq!(allow, vec!["user456"]);
+    }
+
+    #[test]
+    fn test_approve_case_insensitive_code() {
+        let (store, _) = test_store();
+        let r = store.upsert_request("telegram", "user789", None).unwrap();
+        let code_lower = r.code.to_lowercase();
+        let approved = store.approve("telegram", &code_lower).unwrap();
+        assert!(approved.is_some());
+    }
+
+    #[test]
+    fn test_approve_invalid_code_returns_none() {
+        let (store, _) = test_store();
+        store.upsert_request("telegram", "user123", None).unwrap();
+        let approved = store.approve("telegram", "BADCODE1").unwrap();
+        assert!(approved.is_none());
+    }
+
+    #[test]
+    fn test_approve_rate_limited_after_many_failures() {
+        let (store, _) = test_store();
+        store.upsert_request("telegram", "user123", None).unwrap();
+        for _ in 0..PAIRING_APPROVE_RATE_LIMIT {
+            let _ = store.approve("telegram", "WRONG01");
+        }
+        let err = store.approve("telegram", "WRONG02").unwrap_err();
+        assert!(matches!(err, PairingStoreError::ApproveRateLimited));
+    }
+
+    #[test]
+    fn test_is_sender_allowed_by_id() {
+        let (store, _) = test_store();
+        let r = store.upsert_request("telegram", "user999", None).unwrap();
+        store.approve("telegram", &r.code).unwrap();
+
+        assert!(
+            store
+                .is_sender_allowed("telegram", "user999", None)
+                .unwrap()
+        );
+        assert!(!store.is_sender_allowed("telegram", "other", None).unwrap());
+    }
+
+    #[test]
+    fn test_is_sender_allowed_by_username() {
+        let (store, _) = test_store();
+        store
+            .upsert_request(
+                "telegram",
+                "alice",
+                Some(serde_json::json!({"username": "alice"})),
+            )
+            .unwrap();
+        let pending = store.list_pending("telegram").unwrap();
+        store.approve("telegram", &pending[0].code).unwrap();
+
+        // approve adds id to allow_from. For username we need to add it manually.
+        // Actually approve adds entry.id which is "alice". So is_sender_allowed("telegram", "alice", None) would work.
+        assert!(store.is_sender_allowed("telegram", "alice", None).unwrap());
+        assert!(
+            store
+                .is_sender_allowed("telegram", "alice", Some("alice"))
+                .unwrap()
+        );
+    }
+
+    #[test]
+    fn test_channel_normalization() {
+        let (store, _) = test_store();
+        store.upsert_request("Telegram", "u1", None).unwrap();
+        let pending = store.list_pending("telegram").unwrap();
+        assert_eq!(pending.len(), 1);
+        assert_eq!(pending[0].id, "u1");
+    }
+
+    #[test]
+    fn test_invalid_channel_rejected() {
+        let (store, _) = test_store();
+        store.upsert_request("telegram", "u1", None).unwrap();
+        store.list_pending("").unwrap_err();
+        store.upsert_request("", "u1", None).unwrap_err();
+    }
+}

src/sandbox/container.rs

@@ -491,9 +491,36 @@ impl ContainerRunner {
 }
 
 /// Connect to the Docker daemon.
+///
+/// Tries these locations in order:
+/// 1. `DOCKER_HOST` env var (bollard default)
+/// 2. `/var/run/docker.sock` (Linux default)
+/// 3. `~/.docker/run/docker.sock` (Docker Desktop on macOS)
 pub async fn connect_docker() -> Result<Docker> {
-    Docker::connect_with_local_defaults().map_err(|e| SandboxError::DockerNotAvailable {
-        reason: e.to_string(),
+    // First try bollard defaults (checks DOCKER_HOST, then /var/run/docker.sock)
+    if let Ok(docker) = Docker::connect_with_local_defaults() {
+        if docker.ping().await.is_ok() {
+            return Ok(docker);
+        }
+    }
+
+    // Try Docker Desktop socket (macOS)
+    if let Some(home) = std::env::var_os("HOME") {
+        let desktop_sock = std::path::Path::new(&home).join(".docker/run/docker.sock");
+        if desktop_sock.exists() {
+            let sock_str = desktop_sock.to_string_lossy();
+            if let Ok(docker) =
+                Docker::connect_with_socket(&sock_str, 120, bollard::API_DEFAULT_VERSION)
+            {
+                if docker.ping().await.is_ok() {
+                    return Ok(docker);
+                }
+            }
+        }
+    }
+
+    Err(SandboxError::DockerNotAvailable {
+        reason: "Socket not found: /var/run/docker.sock".to_string(),
     })
 }
 

src/secrets/keychain.rs

@@ -52,7 +52,7 @@ mod platform {
     use super::*;
 
     /// Store the master key in the macOS Keychain.
-    pub fn store_master_key(key: &[u8]) -> Result<(), SecretError> {
+    pub async fn store_master_key(key: &[u8]) -> Result<(), SecretError> {
         // Convert to hex for storage (keychain prefers strings)
         let key_hex: String = key.iter().map(|b| format!("{:02x}", b)).collect();
 
@@ -61,7 +61,7 @@ mod platform {
     }
 
     /// Retrieve the master key from the macOS Keychain.
-    pub fn get_master_key() -> Result<Vec<u8>, SecretError> {
+    pub async fn get_master_key() -> Result<Vec<u8>, SecretError> {
         let password = get_generic_password(SERVICE_NAME, MASTER_KEY_ACCOUNT).map_err(|e| {
             SecretError::KeychainError(format!("Failed to get from keychain: {}", e))
         })?;
@@ -74,14 +74,14 @@ mod platform {
     }
 
     /// Delete the master key from the macOS Keychain.
-    pub fn delete_master_key() -> Result<(), SecretError> {
+    pub async fn delete_master_key() -> Result<(), SecretError> {
         delete_generic_password(SERVICE_NAME, MASTER_KEY_ACCOUNT).map_err(|e| {
             SecretError::KeychainError(format!("Failed to delete from keychain: {}", e))
         })
     }
 
     /// Check if a master key exists in the keychain.
-    pub fn has_master_key() -> bool {
+    pub async fn has_master_key() -> bool {
         get_generic_password(SERVICE_NAME, MASTER_KEY_ACCOUNT).is_ok()
     }
 }
@@ -97,163 +97,131 @@ mod platform {
     use super::*;
 
     /// Store the master key in the Linux secret service (GNOME Keyring, KWallet).
-    pub fn store_master_key(key: &[u8]) -> Result<(), SecretError> {
-        let rt = tokio::runtime::Handle::try_current()
-            .map_err(|_| SecretError::KeychainError("No tokio runtime available".to_string()))?;
-
-        rt.block_on(async {
-            let ss = SecretService::connect(EncryptionType::Dh)
-                .await
-                .map_err(|e| {
-                    SecretError::KeychainError(format!(
-                        "Failed to connect to secret service: {}",
-                        e
-                    ))
-                })?;
-
-            let collection = ss.get_default_collection().await.map_err(|e| {
-                SecretError::KeychainError(format!("Failed to get collection: {}", e))
+    pub async fn store_master_key(key: &[u8]) -> Result<(), SecretError> {
+        let ss = SecretService::connect(EncryptionType::Dh)
+            .await
+            .map_err(|e| {
+                SecretError::KeychainError(format!("Failed to connect to secret service: {}", e))
             })?;
 
-            // Unlock if needed
-            if collection.is_locked().await.unwrap_or(true) {
-                collection.unlock().await.map_err(|e| {
-                    SecretError::KeychainError(format!("Failed to unlock collection: {}", e))
-                })?;
-            }
+        let collection = ss
+            .get_default_collection()
+            .await
+            .map_err(|e| SecretError::KeychainError(format!("Failed to get collection: {}", e)))?;
 
-            // Convert to hex for storage
-            let key_hex: String = key.iter().map(|b| format!("{:02x}", b)).collect();
+        // Unlock if needed
+        if collection.is_locked().await.unwrap_or(true) {
+            collection.unlock().await.map_err(|e| {
+                SecretError::KeychainError(format!("Failed to unlock collection: {}", e))
+            })?;
+        }
 
-            collection
-                .create_item(
-                    &format!("{} master key", SERVICE_NAME),
-                    [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
-                        .into_iter()
-                        .collect(),
-                    key_hex.as_bytes(),
-                    true, // Replace if exists
-                    "text/plain",
-                )
-                .await
-                .map_err(|e| {
-                    SecretError::KeychainError(format!("Failed to create secret: {}", e))
-                })?;
+        // Convert to hex for storage
+        let key_hex: String = key.iter().map(|b| format!("{:02x}", b)).collect();
 
-            Ok(())
-        })
+        collection
+            .create_item(
+                &format!("{} master key", SERVICE_NAME),
+                [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
+                    .into_iter()
+                    .collect(),
+                key_hex.as_bytes(),
+                true, // Replace if exists
+                "text/plain",
+            )
+            .await
+            .map_err(|e| SecretError::KeychainError(format!("Failed to create secret: {}", e)))?;
+
+        Ok(())
     }
 
     /// Retrieve the master key from the Linux secret service.
-    pub fn get_master_key() -> Result<Vec<u8>, SecretError> {
-        let rt = tokio::runtime::Handle::try_current()
-            .map_err(|_| SecretError::KeychainError("No tokio runtime available".to_string()))?;
+    pub async fn get_master_key() -> Result<Vec<u8>, SecretError> {
+        let ss = SecretService::connect(EncryptionType::Dh)
+            .await
+            .map_err(|e| {
+                SecretError::KeychainError(format!("Failed to connect to secret service: {}", e))
+            })?;
 
-        rt.block_on(async {
-            let ss = SecretService::connect(EncryptionType::Dh)
+        let items = ss
+            .search_items(
+                [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
+                    .into_iter()
+                    .collect(),
+            )
+            .await
+            .map_err(|e| SecretError::KeychainError(format!("Failed to search: {}", e)))?;
+
+        let item = items
+            .unlocked
+            .first()
+            .or(items.locked.first())
+            .ok_or_else(|| SecretError::KeychainError("Master key not found".to_string()))?;
+
+        // Unlock if needed
+        if item.is_locked().await.unwrap_or(true) {
+            item.unlock()
                 .await
-                .map_err(|e| {
-                    SecretError::KeychainError(format!(
-                        "Failed to connect to secret service: {}",
-                        e
-                    ))
-                })?;
+                .map_err(|e| SecretError::KeychainError(format!("Failed to unlock: {}", e)))?;
+        }
 
-            let items = ss
-                .search_items(
-                    [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
-                        .into_iter()
-                        .collect(),
-                )
-                .await
-                .map_err(|e| SecretError::KeychainError(format!("Failed to search: {}", e)))?;
+        let secret = item
+            .get_secret()
+            .await
+            .map_err(|e| SecretError::KeychainError(format!("Failed to get secret: {}", e)))?;
 
-            let item = items
-                .unlocked
-                .first()
-                .or(items.locked.first())
-                .ok_or_else(|| SecretError::KeychainError("Master key not found".to_string()))?;
+        let hex_str = String::from_utf8(secret)
+            .map_err(|_| SecretError::KeychainError("Invalid UTF-8 in secret".to_string()))?;
 
-            // Unlock if needed
-            if item.is_locked().await.unwrap_or(true) {
-                item.unlock()
-                    .await
-                    .map_err(|e| SecretError::KeychainError(format!("Failed to unlock: {}", e)))?;
-            }
-
-            let secret = item
-                .get_secret()
-                .await
-                .map_err(|e| SecretError::KeychainError(format!("Failed to get secret: {}", e)))?;
-
-            let hex_str = String::from_utf8(secret)
-                .map_err(|_| SecretError::KeychainError("Invalid UTF-8 in secret".to_string()))?;
-
-            hex_to_bytes(&hex_str)
-        })
+        hex_to_bytes(&hex_str)
     }
 
     /// Delete the master key from the Linux secret service.
-    pub fn delete_master_key() -> Result<(), SecretError> {
-        let rt = tokio::runtime::Handle::try_current()
-            .map_err(|_| SecretError::KeychainError("No tokio runtime available".to_string()))?;
+    pub async fn delete_master_key() -> Result<(), SecretError> {
+        let ss = SecretService::connect(EncryptionType::Dh)
+            .await
+            .map_err(|e| {
+                SecretError::KeychainError(format!("Failed to connect to secret service: {}", e))
+            })?;
 
-        rt.block_on(async {
-            let ss = SecretService::connect(EncryptionType::Dh)
+        let items = ss
+            .search_items(
+                [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
+                    .into_iter()
+                    .collect(),
+            )
+            .await
+            .map_err(|e| SecretError::KeychainError(format!("Failed to search: {}", e)))?;
+
+        for item in items.unlocked.iter().chain(items.locked.iter()) {
+            item.delete()
                 .await
-                .map_err(|e| {
-                    SecretError::KeychainError(format!(
-                        "Failed to connect to secret service: {}",
-                        e
-                    ))
-                })?;
+                .map_err(|e| SecretError::KeychainError(format!("Failed to delete: {}", e)))?;
+        }
 
-            let items = ss
-                .search_items(
-                    [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
-                        .into_iter()
-                        .collect(),
-                )
-                .await
-                .map_err(|e| SecretError::KeychainError(format!("Failed to search: {}", e)))?;
-
-            for item in items.unlocked.iter().chain(items.locked.iter()) {
-                item.delete()
-                    .await
-                    .map_err(|e| SecretError::KeychainError(format!("Failed to delete: {}", e)))?;
-            }
-
-            Ok(())
-        })
+        Ok(())
     }
 
     /// Check if a master key exists in the secret service.
-    pub fn has_master_key() -> bool {
-        let rt = match tokio::runtime::Handle::try_current() {
-            Ok(rt) => rt,
+    pub async fn has_master_key() -> bool {
+        let ss = match SecretService::connect(EncryptionType::Dh).await {
+            Ok(ss) => ss,
             Err(_) => return false,
         };
 
-        rt.block_on(async {
-            let ss = match SecretService::connect(EncryptionType::Dh).await {
-                Ok(ss) => ss,
-                Err(_) => return false,
-            };
+        let items = match ss
+            .search_items(
+                [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
+                    .into_iter()
+                    .collect(),
+            )
+            .await
+        {
+            Ok(items) => items,
+            Err(_) => return false,
+        };
 
-            let items = match ss
-                .search_items(
-                    [("service", SERVICE_NAME), ("account", MASTER_KEY_ACCOUNT)]
-                        .into_iter()
-                        .collect(),
-                )
-                .await
-            {
-                Ok(items) => items,
-                Err(_) => return false,
-            };
-
-            !items.unlocked.is_empty() || !items.locked.is_empty()
-        })
+        !items.unlocked.is_empty() || !items.locked.is_empty()
     }
 }
 
@@ -265,25 +233,25 @@ mod platform {
 mod platform {
     use super::*;
 
-    pub fn store_master_key(_key: &[u8]) -> Result<(), SecretError> {
+    pub async fn store_master_key(_key: &[u8]) -> Result<(), SecretError> {
         Err(SecretError::KeychainError(
             "Keychain not supported on this platform. Use SECRETS_MASTER_KEY env var.".to_string(),
         ))
     }
 
-    pub fn get_master_key() -> Result<Vec<u8>, SecretError> {
+    pub async fn get_master_key() -> Result<Vec<u8>, SecretError> {
         Err(SecretError::KeychainError(
             "Keychain not supported on this platform. Use SECRETS_MASTER_KEY env var.".to_string(),
         ))
     }
 
-    pub fn delete_master_key() -> Result<(), SecretError> {
+    pub async fn delete_master_key() -> Result<(), SecretError> {
         Err(SecretError::KeychainError(
             "Keychain not supported on this platform".to_string(),
         ))
     }
 
-    pub fn has_master_key() -> bool {
+    pub async fn has_master_key() -> bool {
         false
     }
 }

src/secrets/types.rs

@@ -192,9 +192,10 @@ impl CreateSecretParams {
 }
 
 /// Where a credential should be injected in an HTTP request.
-#[derive(Debug, Clone, Serialize, Deserialize)]
+#[derive(Debug, Clone, Default, Serialize, Deserialize)]
 pub enum CredentialLocation {
     /// Inject as Authorization header (e.g., "Bearer {secret}")
+    #[default]
     AuthorizationBearer,
     /// Inject as Authorization header with Basic auth
     AuthorizationBasic { username: String },
@@ -209,12 +210,6 @@ pub enum CredentialLocation {
     UrlPath { placeholder: String },
 }
 
-impl Default for CredentialLocation {
-    fn default() -> Self {
-        Self::AuthorizationBearer
-    }
-}
-
 /// Mapping from a secret name to where it should be injected.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct CredentialMapping {

src/settings.rs

@@ -149,6 +149,11 @@ pub struct ChannelSettings {
     #[serde(default)]
     pub http_host: Option<String>,
 
+    /// Telegram owner user ID. When set, the bot only responds to this user.
+    /// Captured during setup by having the user message the bot.
+    #[serde(default)]
+    pub telegram_owner_id: Option<i64>,
+
     /// Enabled WASM channels by name.
     /// Channels not in this list but present in the channels directory will still load.
     /// This is primarily used by the setup wizard to track which channels were configured.
@@ -490,6 +495,51 @@ impl Settings {
             .join("settings.json")
     }
 
+    /// Reconstruct Settings from a flat key-value map (as stored in the DB).
+    ///
+    /// Each key is a dotted path (e.g., "agent.name"), value is a JSONB value.
+    /// Missing keys get their default value.
+    pub fn from_db_map(map: &std::collections::HashMap<String, serde_json::Value>) -> Self {
+        // Start with defaults, then overlay each DB setting
+        let mut settings = Self::default();
+
+        for (key, value) in map {
+            // Convert the JSONB value to a string for the existing set() method
+            let value_str = match value {
+                serde_json::Value::String(s) => s.clone(),
+                serde_json::Value::Bool(b) => b.to_string(),
+                serde_json::Value::Number(n) => n.to_string(),
+                serde_json::Value::Null => "null".to_string(),
+                other => other.to_string(),
+            };
+
+            if let Err(e) = settings.set(key, &value_str) {
+                tracing::warn!(
+                    "Failed to apply DB setting '{}' = '{}': {}",
+                    key,
+                    value_str,
+                    e
+                );
+            }
+        }
+
+        settings
+    }
+
+    /// Flatten Settings into a key-value map suitable for DB storage.
+    ///
+    /// Each entry is a (dotted_path, JSONB value) pair.
+    pub fn to_db_map(&self) -> std::collections::HashMap<String, serde_json::Value> {
+        let json = match serde_json::to_value(self) {
+            Ok(v) => v,
+            Err(_) => return std::collections::HashMap::new(),
+        };
+
+        let mut map = std::collections::HashMap::new();
+        collect_settings_json(&json, String::new(), &mut map);
+        map
+    }
+
     /// Load settings from disk, returning default if not found.
     pub fn load() -> Self {
         Self::load_from(&Self::default_path())
@@ -656,6 +706,29 @@ impl Settings {
     }
 }
 
+/// Recursively collect settings paths with their JSON values (for DB storage).
+fn collect_settings_json(
+    value: &serde_json::Value,
+    prefix: String,
+    results: &mut std::collections::HashMap<String, serde_json::Value>,
+) {
+    match value {
+        serde_json::Value::Object(obj) => {
+            for (key, val) in obj {
+                let path = if prefix.is_empty() {
+                    key.clone()
+                } else {
+                    format!("{}.{}", prefix, key)
+                };
+                collect_settings_json(val, path, results);
+            }
+        }
+        other => {
+            results.insert(prefix, other.clone());
+        }
+    }
+}
+
 /// Recursively collect settings paths and values.
 fn collect_settings(
     value: &serde_json::Value,
@@ -799,4 +872,32 @@ mod tests {
         assert_eq!(settings.embeddings.provider, "nearai");
         assert_eq!(settings.embeddings.model, "text-embedding-3-small");
     }
+
+    #[test]
+    fn test_telegram_owner_id_round_trip() {
+        let dir = tempdir().unwrap();
+        let path = dir.path().join("settings.json");
+
+        let mut settings = Settings::default();
+        settings.channels.telegram_owner_id = Some(123456789);
+        settings.save_to(&path).unwrap();
+
+        let loaded = Settings::load_from(&path);
+        assert_eq!(loaded.channels.telegram_owner_id, Some(123456789));
+    }
+
+    #[test]
+    fn test_telegram_owner_id_default_none() {
+        let settings = Settings::default();
+        assert_eq!(settings.channels.telegram_owner_id, None);
+    }
+
+    #[test]
+    fn test_telegram_owner_id_via_set() {
+        let mut settings = Settings::default();
+        settings
+            .set("channels.telegram_owner_id", "987654321")
+            .unwrap();
+        assert_eq!(settings.channels.telegram_owner_id, Some(987654321));
+    }
 }

src/setup/channels.rs

@@ -52,6 +52,16 @@ impl SecretsContext {
             .await
             .unwrap_or(false)
     }
+
+    /// Read a secret from the database (decrypted).
+    pub async fn get_secret(&self, name: &str) -> Result<SecretString, String> {
+        let decrypted = self
+            .store
+            .get_decrypted(&self.user_id, name)
+            .await
+            .map_err(|e| format!("Failed to read secret: {}", e))?;
+        Ok(SecretString::from(decrypted.expose().to_string()))
+    }
 }
 
 /// Result of Telegram setup.
@@ -60,6 +70,7 @@ pub struct TelegramSetupResult {
     pub enabled: bool,
     pub bot_username: Option<String>,
     pub webhook_secret: Option<String>,
+    pub owner_id: Option<i64>,
 }
 
 /// Telegram Bot API response for getMe.
@@ -76,6 +87,32 @@ struct TelegramUser {
     first_name: String,
 }
 
+/// Telegram Bot API response for getUpdates.
+#[derive(Debug, Deserialize)]
+struct TelegramGetUpdatesResponse {
+    ok: bool,
+    result: Vec<TelegramUpdate>,
+}
+
+#[derive(Debug, Deserialize)]
+struct TelegramUpdate {
+    #[allow(dead_code)]
+    update_id: i64,
+    message: Option<TelegramUpdateMessage>,
+}
+
+#[derive(Debug, Deserialize)]
+struct TelegramUpdateMessage {
+    from: Option<TelegramUpdateUser>,
+}
+
+#[derive(Debug, Deserialize)]
+struct TelegramUpdateUser {
+    id: i64,
+    first_name: String,
+    username: Option<String>,
+}
+
 /// Set up Telegram bot channel.
 ///
 /// Guides the user through:
@@ -96,12 +133,14 @@ pub async fn setup_telegram(secrets: &SecretsContext) -> Result<TelegramSetupRes
     if secrets.secret_exists("telegram_bot_token").await {
         print_info("Existing Telegram token found in database.");
         if !confirm("Replace existing token?", false).map_err(|e| e.to_string())? {
-            // Still offer to configure webhook secret if not already done
+            // Still offer to configure webhook secret and owner binding
             let webhook_secret = setup_telegram_webhook_secret(secrets).await?;
+            let owner_id = bind_telegram_owner_flow(secrets).await?;
             return Ok(TelegramSetupResult {
                 enabled: true,
                 bot_username: None,
                 webhook_secret,
+                owner_id,
             });
         }
     }
@@ -122,6 +161,9 @@ pub async fn setup_telegram(secrets: &SecretsContext) -> Result<TelegramSetupRes
             secrets.save_secret("telegram_bot_token", &token).await?;
             print_success("Token saved to database");
 
+            // Bind bot to owner's Telegram account
+            let owner_id = bind_telegram_owner(&token).await?;
+
             // Offer webhook secret configuration
             let webhook_secret = setup_telegram_webhook_secret(secrets).await?;
 
@@ -129,6 +171,7 @@ pub async fn setup_telegram(secrets: &SecretsContext) -> Result<TelegramSetupRes
                 enabled: true,
                 bot_username: username,
                 webhook_secret,
+                owner_id,
             })
         }
         Err(e) => {
@@ -141,12 +184,128 @@ pub async fn setup_telegram(secrets: &SecretsContext) -> Result<TelegramSetupRes
                     enabled: false,
                     bot_username: None,
                     webhook_secret: None,
+                    owner_id: None,
                 })
             }
         }
     }
 }
 
+/// Bind the bot to the owner's Telegram account by having them send a message.
+///
+/// Polls `getUpdates` until a message arrives, then captures the sender's user ID.
+/// Returns `None` if the user declines or the flow times out.
+async fn bind_telegram_owner(token: &SecretString) -> Result<Option<i64>, String> {
+    println!();
+    print_info("Account Binding (recommended):");
+    print_info("Binding restricts the bot so only YOU can use it.");
+    print_info("Without this, anyone who finds your bot can send it messages.");
+    println!();
+
+    if !confirm("Bind bot to your Telegram account?", true).map_err(|e| e.to_string())? {
+        print_info("Skipping account binding. Bot will accept messages from all users.");
+        return Ok(None);
+    }
+
+    print_info("Send any message (e.g. /start) to your bot in Telegram.");
+    print_info("Waiting for your message (up to 120 seconds)...");
+
+    let client = Client::builder()
+        .timeout(std::time::Duration::from_secs(35))
+        .build()
+        .map_err(|e| format!("Failed to create HTTP client: {}", e))?;
+
+    // Clear any existing webhook so getUpdates works
+    let delete_url = format!(
+        "https://api.telegram.org/bot{}/deleteWebhook",
+        token.expose_secret()
+    );
+    let _ = client.post(&delete_url).send().await;
+
+    let updates_url = format!(
+        "https://api.telegram.org/bot{}/getUpdates",
+        token.expose_secret()
+    );
+
+    let deadline = std::time::Instant::now() + std::time::Duration::from_secs(120);
+
+    while std::time::Instant::now() < deadline {
+        let response = client
+            .get(&updates_url)
+            .query(&[("timeout", "30"), ("allowed_updates", "[\"message\"]")])
+            .send()
+            .await
+            .map_err(|e| format!("getUpdates request failed: {}", e))?;
+
+        if !response.status().is_success() {
+            return Err(format!("getUpdates returned status {}", response.status()));
+        }
+
+        let body: TelegramGetUpdatesResponse = response
+            .json()
+            .await
+            .map_err(|e| format!("Failed to parse getUpdates response: {}", e))?;
+
+        if !body.ok {
+            return Err("Telegram API returned error for getUpdates".to_string());
+        }
+
+        // Find the first message with a sender
+        for update in &body.result {
+            if let Some(ref msg) = update.message {
+                if let Some(ref from) = msg.from {
+                    let display_name = from
+                        .username
+                        .as_ref()
+                        .map(|u| format!("@{}", u))
+                        .unwrap_or_else(|| from.first_name.clone());
+
+                    print_success(&format!(
+                        "Received message from {} (ID: {})",
+                        display_name, from.id
+                    ));
+
+                    // Acknowledge the update so it doesn't pile up
+                    let ack_url = format!(
+                        "https://api.telegram.org/bot{}/getUpdates",
+                        token.expose_secret()
+                    );
+                    let _ = client
+                        .get(&ack_url)
+                        .query(&[("offset", &(update.update_id + 1).to_string())])
+                        .send()
+                        .await;
+
+                    return Ok(Some(from.id));
+                }
+            }
+        }
+    }
+
+    print_error("Timed out waiting for a message. You can re-run setup to try again.");
+    print_info("Bot will accept messages from all users until owner is bound.");
+    Ok(None)
+}
+
+/// Bind flow when the token already exists (reads from secrets store).
+///
+/// Retrieves the saved bot token and delegates to `bind_telegram_owner`.
+async fn bind_telegram_owner_flow(secrets: &SecretsContext) -> Result<Option<i64>, String> {
+    // Check current settings first
+    let settings = Settings::load();
+    if settings.channels.telegram_owner_id.is_some() {
+        print_info("Bot is already bound to a Telegram account.");
+        if !confirm("Re-bind to a different account?", false).map_err(|e| e.to_string())? {
+            return Ok(settings.channels.telegram_owner_id);
+        }
+    }
+
+    // We need the token to poll getUpdates
+    let token = secrets.get_secret("telegram_bot_token").await?;
+
+    bind_telegram_owner(&token).await
+}
+
 /// Set up a tunnel for exposing the agent to the internet.
 ///
 /// This is shared across all channels that need webhook endpoints.

src/setup/wizard.rs

@@ -17,7 +17,7 @@ use secrecy::SecretString;
 use tokio_postgres::NoTls;
 
 use crate::channels::wasm::{
-    ChannelCapabilitiesFile, bundled_channel_names, install_bundled_channel,
+    ChannelCapabilitiesFile, available_channel_names, install_bundled_channel,
 };
 use crate::llm::{SessionConfig, SessionManager};
 use crate::secrets::SecretsCrypto;
@@ -266,7 +266,7 @@ impl SetupWizard {
     async fn step_security(&mut self) -> Result<(), SetupError> {
         // Check current configuration
         let env_key_exists = std::env::var("SECRETS_MASTER_KEY").is_ok();
-        let keychain_key_exists = crate::secrets::keychain::has_master_key();
+        let keychain_key_exists = crate::secrets::keychain::has_master_key().await;
 
         if env_key_exists {
             print_info("Secrets master key found in SECRETS_MASTER_KEY environment variable.");
@@ -304,9 +304,11 @@ impl SetupWizard {
                 print_info("Generating master key...");
                 let key = crate::secrets::keychain::generate_master_key();
 
-                crate::secrets::keychain::store_master_key(&key).map_err(|e| {
-                    SetupError::Config(format!("Failed to store in keychain: {}", e))
-                })?;
+                crate::secrets::keychain::store_master_key(&key)
+                    .await
+                    .map_err(|e| {
+                        SetupError::Config(format!("Failed to store in keychain: {}", e))
+                    })?;
 
                 // Also create crypto instance
                 let key_hex: String = key.iter().map(|b| format!("{:02x}", b)).collect();
@@ -453,6 +455,7 @@ impl SetupWizard {
             .unwrap_or_else(|_| "https://private.near.ai".to_string());
 
         let config = LlmConfig {
+            backend: crate::config::LlmBackend::NearAi,
             nearai: crate::config::NearAiConfig {
                 model: "dummy".to_string(),
                 base_url,
@@ -461,6 +464,10 @@ impl SetupWizard {
                 api_mode: crate::config::NearAiApiMode::Responses,
                 api_key: None,
             },
+            openai: None,
+            anthropic: None,
+            ollama: None,
+            openai_compatible: None,
         };
 
         match create_llm_provider(&config, Arc::clone(session)) {
@@ -550,7 +557,7 @@ impl SetupWizard {
             // Try to load master key from keychain or env
             let key = if let Ok(env_key) = std::env::var("SECRETS_MASTER_KEY") {
                 env_key
-            } else if let Ok(keychain_key) = crate::secrets::keychain::get_master_key() {
+            } else if let Ok(keychain_key) = crate::secrets::keychain::get_master_key().await {
                 keychain_key.iter().map(|b| format!("{:02x}", b)).collect()
             } else {
                 return Err(SetupError::Config(
@@ -689,6 +696,9 @@ impl SetupWizard {
                     } else if channel_name == "telegram" {
                         let telegram_result =
                             setup_telegram(ctx).await.map_err(SetupError::Channel)?;
+                        if let Some(owner_id) = telegram_result.owner_id {
+                            self.settings.channels.telegram_owner_id = Some(owner_id);
+                        }
                         crate::setup::channels::WasmChannelSetupResult {
                             enabled: telegram_result.enabled,
                             channel_name: "telegram".to_string(),
@@ -772,12 +782,9 @@ impl SetupWizard {
     fn save_and_summarize(&mut self) -> Result<(), SetupError> {
         self.settings.onboard_completed = true;
 
-        self.settings.save().map_err(|e| {
-            SetupError::Io(std::io::Error::new(
-                std::io::ErrorKind::Other,
-                format!("Failed to save settings: {}", e),
-            ))
-        })?;
+        self.settings
+            .save()
+            .map_err(|e| std::io::Error::other(format!("Failed to save settings: {}", e)))?;
 
         println!();
         print_success("Configuration saved to ~/.ironclaw/");
@@ -978,7 +985,7 @@ async fn install_missing_bundled_channels(
 ) -> Result<Vec<String>, SetupError> {
     let mut installed = Vec::new();
 
-    for name in bundled_channel_names().iter().copied() {
+    for name in available_channel_names().iter().copied() {
         if already_installed.contains(name) {
             continue;
         }
@@ -995,7 +1002,7 @@ async fn install_missing_bundled_channels(
 fn wasm_channel_option_names(discovered: &[(String, ChannelCapabilitiesFile)]) -> Vec<String> {
     let mut names: Vec<String> = discovered.iter().map(|(name, _)| name.clone()).collect();
 
-    for bundled in bundled_channel_names().iter().copied() {
+    for bundled in available_channel_names().iter().copied() {
         if !names.iter().any(|name| name == bundled) {
             names.push(bundled.to_string());
         }
@@ -1009,7 +1016,7 @@ async fn install_selected_bundled_channels(
     selected_channels: &[String],
     already_installed: &HashSet<String>,
 ) -> Result<Option<Vec<String>>, SetupError> {
-    let bundled: HashSet<&str> = bundled_channel_names().iter().copied().collect();
+    let bundled: HashSet<&str> = available_channel_names().iter().copied().collect();
     let selected_missing: HashSet<String> = selected_channels
         .iter()
         .filter(|name| bundled.contains(name.as_str()) && !already_installed.contains(*name))
@@ -1092,16 +1099,29 @@ mod tests {
     }
 
     #[test]
-    fn test_wasm_channel_option_names_includes_bundled_when_missing() {
+    fn test_wasm_channel_option_names_includes_available_when_missing() {
         let discovered = Vec::new();
         let options = wasm_channel_option_names(&discovered);
-        assert_eq!(options, vec!["telegram".to_string()]);
+        let available = available_channel_names();
+        // All available (built) channels should appear
+        for name in &available {
+            assert!(
+                options.contains(&name.to_string()),
+                "expected '{}' in options",
+                name
+            );
+        }
     }
 
     #[test]
-    fn test_wasm_channel_option_names_dedupes_bundled() {
+    fn test_wasm_channel_option_names_dedupes_available() {
         let discovered = vec![(String::from("telegram"), ChannelCapabilitiesFile::default())];
         let options = wasm_channel_option_names(&discovered);
-        assert_eq!(options, vec!["telegram".to_string()]);
+        // telegram should appear exactly once despite being both discovered and available
+        assert_eq!(
+            options.iter().filter(|n| *n == "telegram").count(),
+            1,
+            "telegram should not be duplicated"
+        );
     }
 }

src/tools/builder/core.rs

@@ -689,9 +689,20 @@ Create alongside the .wasm file to grant capabilities:
                         .messages
                         .push(ChatMessage::user("Continue with the next step."));
                 }
-                RespondResult::ToolCalls(tool_calls) => {
+                RespondResult::ToolCalls {
+                    tool_calls,
+                    content,
+                } => {
                     tools_executed = true;
 
+                    // Add assistant message with tool_calls (OpenAI protocol)
+                    reason_ctx
+                        .messages
+                        .push(ChatMessage::assistant_with_tool_calls(
+                            content,
+                            tool_calls.clone(),
+                        ));
+
                     // Execute each tool call
                     for tc in tool_calls {
                         logs.push(BuildLog {

src/tools/builtin/extension_tools.rs

@@ -185,8 +185,9 @@ impl Tool for ToolAuthTool {
     }
 
     fn description(&self) -> &str {
-        "Authenticate an installed extension. For MCP servers, starts OAuth flow. \
-         For WASM tools with manual auth, returns instructions; call again with token param to complete."
+        "Initiate authentication for an extension. For OAuth, returns a URL. \
+         For manual auth, returns instructions. The user provides their token \
+         through a secure channel, never through this tool."
     }
 
     fn parameters_schema(&self) -> serde_json::Value {
@@ -196,10 +197,6 @@ impl Tool for ToolAuthTool {
                 "name": {
                     "type": "string",
                     "description": "Extension name to authenticate"
-                },
-                "token": {
-                    "type": "string",
-                    "description": "API token/key for manual auth (WASM tools). Provide after user gives you the token."
                 }
             },
             "required": ["name"]
@@ -218,11 +215,9 @@ impl Tool for ToolAuthTool {
             .and_then(|v| v.as_str())
             .ok_or_else(|| ToolError::InvalidParameters("name is required".to_string()))?;
 
-        let token = params.get("token").and_then(|v| v.as_str());
-
         let result = self
             .manager
-            .auth(name, token)
+            .auth(name, None)
             .await
             .map_err(|e| ToolError::ExecutionFailed(e.to_string()))?;
 
@@ -316,16 +311,53 @@ impl Tool for ToolActivateTool {
             .and_then(|v| v.as_str())
             .ok_or_else(|| ToolError::InvalidParameters("name is required".to_string()))?;
 
-        let result = self
-            .manager
-            .activate(name)
-            .await
-            .map_err(|e| ToolError::ExecutionFailed(e.to_string()))?;
+        match self.manager.activate(name).await {
+            Ok(result) => {
+                let output = serde_json::to_value(&result)
+                    .unwrap_or_else(|_| serde_json::json!({"error": "serialization failed"}));
+                Ok(ToolOutput::success(output, start.elapsed()))
+            }
+            Err(activate_err) => {
+                let err_str = activate_err.to_string();
+                let needs_auth = err_str.contains("authentication")
+                    || err_str.contains("401")
+                    || err_str.contains("Unauthorized")
+                    || err_str.contains("not authenticated");
 
-        let output = serde_json::to_value(&result)
-            .unwrap_or_else(|_| serde_json::json!({"error": "serialization failed"}));
+                if !needs_auth {
+                    return Err(ToolError::ExecutionFailed(err_str));
+                }
 
-        Ok(ToolOutput::success(output, start.elapsed()))
+                // Activation failed due to missing auth; initiate auth flow
+                // so the agent loop can show the auth card.
+                match self.manager.auth(name, None).await {
+                    Ok(auth_result) if auth_result.status == "authenticated" => {
+                        // Auth succeeded (e.g. env var was set); retry activation.
+                        let result = self
+                            .manager
+                            .activate(name)
+                            .await
+                            .map_err(|e| ToolError::ExecutionFailed(e.to_string()))?;
+                        let output = serde_json::to_value(&result).unwrap_or_else(
+                            |_| serde_json::json!({"error": "serialization failed"}),
+                        );
+                        Ok(ToolOutput::success(output, start.elapsed()))
+                    }
+                    Ok(auth_result) => {
+                        // Auth needs user input (awaiting_token). Return the auth
+                        // result so detect_auth_awaiting picks it up.
+                        let output = serde_json::to_value(&auth_result).unwrap_or_else(
+                            |_| serde_json::json!({"error": "serialization failed"}),
+                        );
+                        Ok(ToolOutput::success(output, start.elapsed()))
+                    }
+                    Err(auth_err) => Err(ToolError::ExecutionFailed(format!(
+                        "Activation failed ({}), and authentication also failed: {}",
+                        err_str, auth_err
+                    ))),
+                }
+            }
+        }
     }
 }
 
@@ -499,7 +531,11 @@ mod tests {
         assert!(tool.requires_approval());
         let schema = tool.parameters_schema();
         assert!(schema["properties"].get("name").is_some());
-        assert!(schema["properties"].get("token").is_some());
+        // token param must NOT be in schema (security: tokens never go through LLM)
+        assert!(
+            schema["properties"].get("token").is_none(),
+            "tool_auth must not have a token parameter"
+        );
     }
 
     #[test]
@@ -550,6 +586,7 @@ mod tests {
             std::path::PathBuf::from("/tmp/ironclaw-test-channels"),
             None,
             "test".to_string(),
+            None,
         ))
     }
 }

src/tools/builtin/file.rs

@@ -11,7 +11,7 @@ use async_trait::async_trait;
 use tokio::fs;
 
 use crate::context::JobContext;
-use crate::tools::tool::{Tool, ToolError, ToolOutput};
+use crate::tools::tool::{Tool, ToolDomain, ToolError, ToolOutput};
 use crate::workspace::paths as ws_paths;
 
 /// Well-known workspace filenames that must go through memory_write, not write_file.
@@ -36,7 +36,7 @@ fn is_workspace_path(path: &str) -> bool {
         .and_then(|f| f.to_str())
         .unwrap_or(path);
 
-    WORKSPACE_FILES.iter().any(|ws| *ws == filename)
+    WORKSPACE_FILES.contains(&filename)
         || path.starts_with("daily/")
         || path.starts_with("context/")
 }
@@ -246,6 +246,10 @@ impl Tool for ReadFileTool {
     fn requires_approval(&self) -> bool {
         true // Reading local files should require approval
     }
+
+    fn domain(&self) -> ToolDomain {
+        ToolDomain::Container
+    }
 }
 
 /// Write file contents tool.
@@ -359,6 +363,10 @@ impl Tool for WriteFileTool {
     fn requires_sanitization(&self) -> bool {
         false // We're writing, not reading external data
     }
+
+    fn domain(&self) -> ToolDomain {
+        ToolDomain::Container
+    }
 }
 
 /// List directory contents tool.
@@ -467,6 +475,10 @@ impl Tool for ListDirTool {
     fn requires_approval(&self) -> bool {
         true // Directory listings can leak filesystem structure
     }
+
+    fn domain(&self) -> ToolDomain {
+        ToolDomain::Container
+    }
 }
 
 /// Recursively list directory contents.
@@ -685,6 +697,10 @@ impl Tool for ApplyPatchTool {
     fn requires_sanitization(&self) -> bool {
         false // We're writing, not reading external data
     }
+
+    fn domain(&self) -> ToolDomain {
+        ToolDomain::Container
+    }
 }
 
 #[cfg(test)]

src/tools/builtin/job.rs

@@ -1,77 +1,108 @@
 //! Job management tools.
 //!
 //! These tools allow the LLM to manage jobs:
-//! - Create new jobs/tasks
+//! - Create new jobs/tasks (with optional sandbox delegation)
 //! - List existing jobs
 //! - Check job status
 //! - Cancel running jobs
 
+use std::path::PathBuf;
 use std::sync::Arc;
+use std::time::Duration;
 
 use async_trait::async_trait;
+use chrono::Utc;
 use uuid::Uuid;
 
 use crate::context::{ContextManager, JobContext, JobState};
+use crate::history::{SandboxJobRecord, Store};
+use crate::orchestrator::job_manager::{ContainerJobManager, JobMode};
 use crate::tools::tool::{Tool, ToolError, ToolOutput};
 
 /// Tool for creating a new job.
+///
+/// When sandbox deps are injected (via `with_sandbox`), the tool automatically
+/// delegates execution to a Docker container. Otherwise it creates an in-memory
+/// job via the ContextManager. The LLM never needs to know the difference.
 pub struct CreateJobTool {
     context_manager: Arc<ContextManager>,
+    job_manager: Option<Arc<ContainerJobManager>>,
+    store: Option<Arc<Store>>,
 }
 
 impl CreateJobTool {
     pub fn new(context_manager: Arc<ContextManager>) -> Self {
-        Self { context_manager }
-    }
-}
-
-#[async_trait]
-impl Tool for CreateJobTool {
-    fn name(&self) -> &str {
-        "create_job"
+        Self {
+            context_manager,
+            job_manager: None,
+            store: None,
+        }
     }
 
-    fn description(&self) -> &str {
-        "Create a new job or task for the agent to work on. Use this when the user wants \
-         you to do something substantial that should be tracked as a separate job."
+    /// Inject sandbox dependencies so `create_job` delegates to Docker containers.
+    pub fn with_sandbox(
+        mut self,
+        job_manager: Arc<ContainerJobManager>,
+        store: Option<Arc<Store>>,
+    ) -> Self {
+        self.job_manager = Some(job_manager);
+        self.store = store;
+        self
     }
 
-    fn parameters_schema(&self) -> serde_json::Value {
-        serde_json::json!({
-            "type": "object",
-            "properties": {
-                "title": {
-                    "type": "string",
-                    "description": "A short title for the job (max 100 chars)"
-                },
-                "description": {
-                    "type": "string",
-                    "description": "Full description of what needs to be done"
+    fn sandbox_enabled(&self) -> bool {
+        self.job_manager.is_some()
+    }
+
+    /// Persist a sandbox job record (fire-and-forget).
+    fn persist_job(&self, record: SandboxJobRecord) {
+        if let Some(store) = self.store.clone() {
+            tokio::spawn(async move {
+                if let Err(e) = store.save_sandbox_job(&record).await {
+                    tracing::warn!(job_id = %record.id, "Failed to persist sandbox job: {}", e);
                 }
-            },
-            "required": ["title", "description"]
-        })
+            });
+        }
     }
 
-    async fn execute(
+    /// Update sandbox job status in DB (fire-and-forget).
+    fn update_status(
         &self,
-        params: serde_json::Value,
+        job_id: Uuid,
+        status: &str,
+        success: Option<bool>,
+        message: Option<String>,
+        started_at: Option<chrono::DateTime<Utc>>,
+        completed_at: Option<chrono::DateTime<Utc>>,
+    ) {
+        if let Some(store) = self.store.clone() {
+            let status = status.to_string();
+            tokio::spawn(async move {
+                if let Err(e) = store
+                    .update_sandbox_job_status(
+                        job_id,
+                        &status,
+                        success,
+                        message.as_deref(),
+                        started_at,
+                        completed_at,
+                    )
+                    .await
+                {
+                    tracing::warn!(job_id = %job_id, "Failed to update sandbox job status: {}", e);
+                }
+            });
+        }
+    }
+
+    /// Execute via in-memory ContextManager (no sandbox).
+    async fn execute_local(
+        &self,
+        title: &str,
+        description: &str,
         ctx: &JobContext,
     ) -> Result<ToolOutput, ToolError> {
         let start = std::time::Instant::now();
-
-        let title = params
-            .get("title")
-            .and_then(|v| v.as_str())
-            .ok_or_else(|| ToolError::InvalidParameters("missing 'title' parameter".into()))?;
-
-        let description = params
-            .get("description")
-            .and_then(|v| v.as_str())
-            .ok_or_else(|| {
-                ToolError::InvalidParameters("missing 'description' parameter".into())
-            })?;
-
         match self
             .context_manager
             .create_job_for_user(&ctx.user_id, title, description)
@@ -95,6 +126,374 @@ impl Tool for CreateJobTool {
         }
     }
 
+    /// Execute via sandboxed Docker container.
+    async fn execute_sandbox(
+        &self,
+        task: &str,
+        explicit_dir: Option<PathBuf>,
+        wait: bool,
+        mode: JobMode,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let start = std::time::Instant::now();
+        let jm = self.job_manager.as_ref().expect("sandbox deps required");
+
+        let job_id = Uuid::new_v4();
+        let (project_dir, browse_id) = resolve_project_dir(explicit_dir, job_id)?;
+        let project_dir_str = project_dir.display().to_string();
+
+        // Persist the job to DB before creating the container.
+        self.persist_job(SandboxJobRecord {
+            id: job_id,
+            task: task.to_string(),
+            status: "creating".to_string(),
+            user_id: ctx.user_id.clone(),
+            project_dir: project_dir_str.clone(),
+            success: None,
+            failure_reason: None,
+            created_at: Utc::now(),
+            started_at: None,
+            completed_at: None,
+        });
+
+        // Persist the job mode to DB
+        if mode == JobMode::ClaudeCode {
+            if let Some(store) = self.store.clone() {
+                let job_id_copy = job_id;
+                tokio::spawn(async move {
+                    if let Err(e) = store
+                        .update_sandbox_job_mode(job_id_copy, "claude_code")
+                        .await
+                    {
+                        tracing::warn!(job_id = %job_id_copy, "Failed to set job mode: {}", e);
+                    }
+                });
+            }
+        }
+
+        // Create the container job with the pre-determined job_id.
+        let _token = jm
+            .create_job(job_id, task, Some(project_dir), mode)
+            .await
+            .map_err(|e| {
+                self.update_status(
+                    job_id,
+                    "failed",
+                    Some(false),
+                    Some(e.to_string()),
+                    None,
+                    Some(Utc::now()),
+                );
+                ToolError::ExecutionFailed(format!("failed to create container: {}", e))
+            })?;
+
+        // Container started successfully.
+        let now = Utc::now();
+        self.update_status(job_id, "running", None, None, Some(now), None);
+
+        if !wait {
+            let result = serde_json::json!({
+                "job_id": job_id.to_string(),
+                "status": "started",
+                "message": "Container started. Use job tools to check status.",
+                "project_dir": project_dir_str,
+                "browse_url": format!("/projects/{}", browse_id),
+            });
+            return Ok(ToolOutput::success(result, start.elapsed()));
+        }
+
+        // Wait for completion by polling the container state.
+        let timeout = Duration::from_secs(600);
+        let poll_interval = Duration::from_secs(2);
+        let deadline = tokio::time::Instant::now() + timeout;
+
+        loop {
+            if tokio::time::Instant::now() > deadline {
+                let _ = jm.stop_job(job_id).await;
+                jm.cleanup_job(job_id).await;
+                self.update_status(
+                    job_id,
+                    "failed",
+                    Some(false),
+                    Some("Timed out (10 minutes)".to_string()),
+                    None,
+                    Some(Utc::now()),
+                );
+                return Err(ToolError::ExecutionFailed(
+                    "container execution timed out (10 minutes)".to_string(),
+                ));
+            }
+
+            match jm.get_handle(job_id).await {
+                Some(handle) => match handle.state {
+                    crate::orchestrator::job_manager::ContainerState::Running
+                    | crate::orchestrator::job_manager::ContainerState::Creating => {
+                        tokio::time::sleep(poll_interval).await;
+                    }
+                    crate::orchestrator::job_manager::ContainerState::Stopped => {
+                        let message = handle
+                            .completion_result
+                            .as_ref()
+                            .and_then(|r| r.message.clone())
+                            .unwrap_or_else(|| "Container job completed".to_string());
+                        let success = handle
+                            .completion_result
+                            .as_ref()
+                            .map(|r| r.success)
+                            .unwrap_or(true);
+                        jm.cleanup_job(job_id).await;
+
+                        let finished_at = Utc::now();
+                        if success {
+                            self.update_status(
+                                job_id,
+                                "completed",
+                                Some(true),
+                                None,
+                                None,
+                                Some(finished_at),
+                            );
+                            let result = serde_json::json!({
+                                "job_id": job_id.to_string(),
+                                "status": "completed",
+                                "output": message,
+                                "project_dir": project_dir_str,
+                                "browse_url": format!("/projects/{}", browse_id),
+                            });
+                            return Ok(ToolOutput::success(result, start.elapsed()));
+                        } else {
+                            self.update_status(
+                                job_id,
+                                "failed",
+                                Some(false),
+                                Some(message.clone()),
+                                None,
+                                Some(finished_at),
+                            );
+                            return Err(ToolError::ExecutionFailed(format!(
+                                "container job failed: {}",
+                                message
+                            )));
+                        }
+                    }
+                    crate::orchestrator::job_manager::ContainerState::Failed => {
+                        let message = handle
+                            .completion_result
+                            .as_ref()
+                            .and_then(|r| r.message.clone())
+                            .unwrap_or_else(|| "unknown failure".to_string());
+                        jm.cleanup_job(job_id).await;
+                        self.update_status(
+                            job_id,
+                            "failed",
+                            Some(false),
+                            Some(message.clone()),
+                            None,
+                            Some(Utc::now()),
+                        );
+                        return Err(ToolError::ExecutionFailed(format!(
+                            "container job failed: {}",
+                            message
+                        )));
+                    }
+                },
+                None => {
+                    self.update_status(
+                        job_id,
+                        "completed",
+                        Some(true),
+                        None,
+                        None,
+                        Some(Utc::now()),
+                    );
+                    let result = serde_json::json!({
+                        "job_id": job_id.to_string(),
+                        "status": "completed",
+                        "output": "Container job completed",
+                        "project_dir": project_dir_str,
+                        "browse_url": format!("/projects/{}", browse_id),
+                    });
+                    return Ok(ToolOutput::success(result, start.elapsed()));
+                }
+            }
+        }
+    }
+}
+
+/// The base directory where all project directories must live.
+fn projects_base() -> PathBuf {
+    dirs::home_dir()
+        .unwrap_or_else(|| PathBuf::from("."))
+        .join(".ironclaw")
+        .join("projects")
+}
+
+/// Resolve the project directory, creating it if it doesn't exist.
+///
+/// Auto-creates `~/.ironclaw/projects/{project_id}/` so every sandbox job has a
+/// persistent bind mount that survives container teardown.
+///
+/// When an explicit path is provided (e.g. job restarts reusing the old dir),
+/// it is validated to fall within `~/.ironclaw/projects/` after canonicalization.
+fn resolve_project_dir(
+    explicit: Option<PathBuf>,
+    project_id: Uuid,
+) -> Result<(PathBuf, String), ToolError> {
+    let base = projects_base();
+    std::fs::create_dir_all(&base).map_err(|e| {
+        ToolError::ExecutionFailed(format!(
+            "failed to create projects base {}: {}",
+            base.display(),
+            e
+        ))
+    })?;
+    let canonical_base = base.canonicalize().map_err(|e| {
+        ToolError::ExecutionFailed(format!("failed to canonicalize projects base: {}", e))
+    })?;
+
+    let dir = match explicit {
+        Some(d) => d,
+        None => canonical_base.join(project_id.to_string()),
+    };
+
+    std::fs::create_dir_all(&dir).map_err(|e| {
+        ToolError::ExecutionFailed(format!(
+            "failed to create project dir {}: {}",
+            dir.display(),
+            e
+        ))
+    })?;
+
+    // Canonicalize resolves symlinks, `..`, etc. so we can do a reliable prefix check.
+    let canonical_dir = dir.canonicalize().map_err(|e| {
+        ToolError::ExecutionFailed(format!(
+            "failed to canonicalize project dir {}: {}",
+            dir.display(),
+            e
+        ))
+    })?;
+
+    if !canonical_dir.starts_with(&canonical_base) {
+        return Err(ToolError::InvalidParameters(format!(
+            "project directory must be under {}",
+            canonical_base.display()
+        )));
+    }
+
+    let browse_id = canonical_dir
+        .file_name()
+        .map(|n| n.to_string_lossy().to_string())
+        .unwrap_or_else(|| project_id.to_string());
+    Ok((canonical_dir, browse_id))
+}
+
+#[async_trait]
+impl Tool for CreateJobTool {
+    fn name(&self) -> &str {
+        "create_job"
+    }
+
+    fn description(&self) -> &str {
+        if self.sandbox_enabled() {
+            "Create and execute a job. The job runs in a sandboxed Docker container with its own \
+             sub-agent that has shell, file read/write, list_dir, and apply_patch tools. Use this \
+             whenever the user asks you to build, create, or work on something. The task \
+             description should be detailed enough for the sub-agent to work independently. \
+             Set wait=false to start immediately while continuing the conversation. Set mode \
+             to 'claude_code' for complex software engineering tasks."
+        } else {
+            "Create a new job or task for the agent to work on. Use this when the user wants \
+             you to do something substantial that should be tracked as a separate job."
+        }
+    }
+
+    fn parameters_schema(&self) -> serde_json::Value {
+        if self.sandbox_enabled() {
+            serde_json::json!({
+                "type": "object",
+                "properties": {
+                    "title": {
+                        "type": "string",
+                        "description": "Clear description of what to accomplish"
+                    },
+                    "description": {
+                        "type": "string",
+                        "description": "Full description of what needs to be done"
+                    },
+                    "wait": {
+                        "type": "boolean",
+                        "description": "If true (default), wait for the container to complete and return results. \
+                                        If false, start the container and return the job_id immediately."
+                    },
+                    "mode": {
+                        "type": "string",
+                        "enum": ["worker", "claude_code"],
+                        "description": "Execution mode. 'worker' (default) uses the IronClaw sub-agent. \
+                                        'claude_code' uses Claude Code CLI for full agentic software engineering."
+                    }
+                },
+                "required": ["title", "description"]
+            })
+        } else {
+            serde_json::json!({
+                "type": "object",
+                "properties": {
+                    "title": {
+                        "type": "string",
+                        "description": "A short title for the job (max 100 chars)"
+                    },
+                    "description": {
+                        "type": "string",
+                        "description": "Full description of what needs to be done"
+                    }
+                },
+                "required": ["title", "description"]
+            })
+        }
+    }
+
+    fn execution_timeout(&self) -> Duration {
+        if self.sandbox_enabled() {
+            // Sandbox polls for up to 10 min internally; give an extra 60s buffer.
+            Duration::from_secs(660)
+        } else {
+            Duration::from_secs(30)
+        }
+    }
+
+    async fn execute(
+        &self,
+        params: serde_json::Value,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let title = params
+            .get("title")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'title' parameter".into()))?;
+
+        let description = params
+            .get("description")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| {
+                ToolError::InvalidParameters("missing 'description' parameter".into())
+            })?;
+
+        if self.sandbox_enabled() {
+            let wait = params.get("wait").and_then(|v| v.as_bool()).unwrap_or(true);
+
+            let mode = match params.get("mode").and_then(|v| v.as_str()) {
+                Some("claude_code") => JobMode::ClaudeCode,
+                _ => JobMode::Worker,
+            };
+
+            // Combine title and description into the task prompt for the sub-agent.
+            let task = format!("{}\n\n{}", title, description);
+            self.execute_sandbox(&task, None, wait, mode, ctx).await
+        } else {
+            self.execute_local(title, description, ctx).await
+        }
+    }
+
     fn requires_sanitization(&self) -> bool {
         false
     }
@@ -377,10 +776,13 @@ mod tests {
     use super::*;
 
     #[tokio::test]
-    async fn test_create_job_tool() {
+    async fn test_create_job_tool_local() {
         let manager = Arc::new(ContextManager::new(5));
         let tool = CreateJobTool::new(manager.clone());
 
+        // Without sandbox deps, it should use the local path
+        assert!(!tool.sandbox_enabled());
+
         let params = serde_json::json!({
             "title": "Test Job",
             "description": "A test job description"
@@ -391,6 +793,37 @@ mod tests {
 
         let job_id = result.result.get("job_id").unwrap().as_str().unwrap();
         assert!(!job_id.is_empty());
+        assert_eq!(
+            result.result.get("status").unwrap().as_str().unwrap(),
+            "pending"
+        );
+    }
+
+    #[test]
+    fn test_schema_changes_with_sandbox() {
+        let manager = Arc::new(ContextManager::new(5));
+
+        // Without sandbox
+        let tool = CreateJobTool::new(Arc::clone(&manager));
+        let schema = tool.parameters_schema();
+        let props = schema.get("properties").unwrap().as_object().unwrap();
+        assert!(props.contains_key("title"));
+        assert!(props.contains_key("description"));
+        assert!(
+            !props.contains_key("project_dir"),
+            "project_dir must not be exposed to the LLM"
+        );
+        assert!(!props.contains_key("wait"));
+        assert!(!props.contains_key("mode"));
+    }
+
+    #[test]
+    fn test_execution_timeout_sandbox() {
+        let manager = Arc::new(ContextManager::new(5));
+
+        // Without sandbox: default timeout
+        let tool = CreateJobTool::new(Arc::clone(&manager));
+        assert_eq!(tool.execution_timeout(), Duration::from_secs(30));
     }
 
     #[tokio::test]
@@ -429,4 +862,67 @@ mod tests {
             "Test Job"
         );
     }
+
+    #[test]
+    fn test_resolve_project_dir_auto() {
+        let project_id = Uuid::new_v4();
+        let (dir, browse_id) = resolve_project_dir(None, project_id).unwrap();
+        assert!(dir.exists());
+        assert!(dir.ends_with(project_id.to_string()));
+        assert_eq!(browse_id, project_id.to_string());
+
+        // Must be under the projects base
+        let base = projects_base().canonicalize().unwrap();
+        assert!(dir.starts_with(&base));
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
+    #[test]
+    fn test_resolve_project_dir_explicit_under_base() {
+        let base = projects_base();
+        std::fs::create_dir_all(&base).unwrap();
+        let explicit = base.join("test_explicit_project");
+        let project_id = Uuid::new_v4();
+
+        let (dir, browse_id) = resolve_project_dir(Some(explicit.clone()), project_id).unwrap();
+        assert!(dir.exists());
+        assert_eq!(browse_id, "test_explicit_project");
+
+        let canonical_base = base.canonicalize().unwrap();
+        assert!(dir.starts_with(&canonical_base));
+
+        let _ = std::fs::remove_dir_all(&explicit);
+    }
+
+    #[test]
+    fn test_resolve_project_dir_rejects_outside_base() {
+        let tmp = tempfile::tempdir().unwrap();
+        let escape_attempt = tmp.path().join("evil_project");
+
+        let result = resolve_project_dir(Some(escape_attempt), Uuid::new_v4());
+        assert!(result.is_err());
+        let err = result.unwrap_err().to_string();
+        assert!(
+            err.contains("must be under"),
+            "expected 'must be under' error, got: {}",
+            err
+        );
+    }
+
+    #[test]
+    fn test_resolve_project_dir_rejects_traversal() {
+        // Attempt to escape via `..` components
+        let base = projects_base();
+        let traversal = base.join("legit").join("..").join("..").join(".ssh");
+
+        let result = resolve_project_dir(Some(traversal), Uuid::new_v4());
+        assert!(result.is_err());
+        let err = result.unwrap_err().to_string();
+        assert!(
+            err.contains("must be under"),
+            "expected 'must be under' error, got: {}",
+            err
+        );
+    }
 }

src/tools/builtin/mod.rs

@@ -10,6 +10,7 @@ mod json;
 mod marketplace;
 mod memory;
 mod restaurant;
+pub mod routine;
 mod shell;
 mod taskrabbit;
 mod time;
@@ -26,6 +27,9 @@ pub use json::JsonTool;
 pub use marketplace::MarketplaceTool;
 pub use memory::{MemoryReadTool, MemorySearchTool, MemoryTreeTool, MemoryWriteTool};
 pub use restaurant::RestaurantTool;
+pub use routine::{
+    RoutineCreateTool, RoutineDeleteTool, RoutineHistoryTool, RoutineListTool, RoutineUpdateTool,
+};
 pub use shell::ShellTool;
 pub use taskrabbit::TaskRabbitTool;
 pub use time::TimeTool;

src/tools/builtin/routine.rs

@@ -0,0 +1,654 @@
+//! LLM-facing tools for managing routines.
+//!
+//! Five tools let the agent manage routines conversationally:
+//! - `routine_create` - Create a new routine
+//! - `routine_list` - List all routines with status
+//! - `routine_update` - Modify or toggle a routine
+//! - `routine_delete` - Remove a routine
+//! - `routine_history` - View past runs
+
+use std::sync::Arc;
+use std::time::Duration;
+
+use async_trait::async_trait;
+use chrono::Utc;
+use uuid::Uuid;
+
+use crate::agent::routine::{
+    NotifyConfig, Routine, RoutineAction, RoutineGuardrails, Trigger, next_cron_fire,
+};
+use crate::agent::routine_engine::RoutineEngine;
+use crate::context::JobContext;
+use crate::history::Store;
+use crate::tools::tool::{Tool, ToolError, ToolOutput};
+
+// ==================== routine_create ====================
+
+pub struct RoutineCreateTool {
+    store: Arc<Store>,
+    engine: Arc<RoutineEngine>,
+}
+
+impl RoutineCreateTool {
+    pub fn new(store: Arc<Store>, engine: Arc<RoutineEngine>) -> Self {
+        Self { store, engine }
+    }
+}
+
+#[async_trait]
+impl Tool for RoutineCreateTool {
+    fn name(&self) -> &str {
+        "routine_create"
+    }
+
+    fn description(&self) -> &str {
+        "Create a new routine (scheduled or event-driven task). \
+         Supports cron schedules, event pattern matching, webhooks, and manual triggers. \
+         Use this when the user wants something to happen periodically or reactively."
+    }
+
+    fn parameters_schema(&self) -> serde_json::Value {
+        serde_json::json!({
+            "type": "object",
+            "properties": {
+                "name": {
+                    "type": "string",
+                    "description": "Unique name for the routine (e.g. 'daily-pr-review')"
+                },
+                "description": {
+                    "type": "string",
+                    "description": "What this routine does"
+                },
+                "trigger_type": {
+                    "type": "string",
+                    "enum": ["cron", "event", "webhook", "manual"],
+                    "description": "When the routine fires"
+                },
+                "schedule": {
+                    "type": "string",
+                    "description": "Cron expression (for cron trigger). E.g. '0 9 * * MON-FRI' for weekdays at 9am. Uses 6-field cron (sec min hour day month weekday)."
+                },
+                "event_pattern": {
+                    "type": "string",
+                    "description": "Regex pattern to match messages (for event trigger)"
+                },
+                "event_channel": {
+                    "type": "string",
+                    "description": "Optional channel filter for event trigger (e.g. 'telegram')"
+                },
+                "prompt": {
+                    "type": "string",
+                    "description": "The prompt/instructions for the routine"
+                },
+                "context_paths": {
+                    "type": "array",
+                    "items": { "type": "string" },
+                    "description": "Workspace paths to load as context (e.g. ['context/priorities.md'])"
+                },
+                "action_type": {
+                    "type": "string",
+                    "enum": ["lightweight", "full_job"],
+                    "description": "Execution mode: 'lightweight' (single LLM call, default) or 'full_job' (multi-turn with tools)"
+                },
+                "cooldown_secs": {
+                    "type": "integer",
+                    "description": "Minimum seconds between fires (default: 300)"
+                }
+            },
+            "required": ["name", "trigger_type", "prompt"]
+        })
+    }
+
+    async fn execute(
+        &self,
+        params: serde_json::Value,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let start = std::time::Instant::now();
+
+        let name = params
+            .get("name")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'name'".to_string()))?;
+
+        let description = params
+            .get("description")
+            .and_then(|v| v.as_str())
+            .unwrap_or("");
+
+        let trigger_type = params
+            .get("trigger_type")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'trigger_type'".to_string()))?;
+
+        let prompt = params
+            .get("prompt")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'prompt'".to_string()))?;
+
+        // Build trigger
+        let trigger = match trigger_type {
+            "cron" => {
+                let schedule =
+                    params
+                        .get("schedule")
+                        .and_then(|v| v.as_str())
+                        .ok_or_else(|| {
+                            ToolError::InvalidParameters(
+                                "cron trigger requires 'schedule'".to_string(),
+                            )
+                        })?;
+                // Validate cron expression
+                next_cron_fire(schedule).map_err(|e| {
+                    ToolError::InvalidParameters(format!("invalid cron schedule: {e}"))
+                })?;
+                Trigger::Cron {
+                    schedule: schedule.to_string(),
+                }
+            }
+            "event" => {
+                let pattern = params
+                    .get("event_pattern")
+                    .and_then(|v| v.as_str())
+                    .ok_or_else(|| {
+                        ToolError::InvalidParameters(
+                            "event trigger requires 'event_pattern'".to_string(),
+                        )
+                    })?;
+                // Validate regex
+                regex::Regex::new(pattern)
+                    .map_err(|e| ToolError::InvalidParameters(format!("invalid regex: {e}")))?;
+                let channel = params
+                    .get("event_channel")
+                    .and_then(|v| v.as_str())
+                    .map(String::from);
+                Trigger::Event {
+                    channel,
+                    pattern: pattern.to_string(),
+                }
+            }
+            "webhook" => Trigger::Webhook {
+                path: None,
+                secret: None,
+            },
+            "manual" => Trigger::Manual,
+            other => {
+                return Err(ToolError::InvalidParameters(format!(
+                    "unknown trigger_type: {other}"
+                )));
+            }
+        };
+
+        // Build action
+        let action_type = params
+            .get("action_type")
+            .and_then(|v| v.as_str())
+            .unwrap_or("lightweight");
+
+        let context_paths: Vec<String> = params
+            .get("context_paths")
+            .and_then(|v| v.as_array())
+            .map(|arr| {
+                arr.iter()
+                    .filter_map(|v| v.as_str().map(String::from))
+                    .collect()
+            })
+            .unwrap_or_default();
+
+        let action = match action_type {
+            "lightweight" => RoutineAction::Lightweight {
+                prompt: prompt.to_string(),
+                context_paths,
+                max_tokens: 4096,
+            },
+            "full_job" => RoutineAction::FullJob {
+                title: name.to_string(),
+                description: prompt.to_string(),
+                max_iterations: 10,
+            },
+            other => {
+                return Err(ToolError::InvalidParameters(format!(
+                    "unknown action_type: {other}"
+                )));
+            }
+        };
+
+        let cooldown_secs = params
+            .get("cooldown_secs")
+            .and_then(|v| v.as_u64())
+            .unwrap_or(300);
+
+        // Compute next fire time for cron
+        let next_fire = if let Trigger::Cron { ref schedule } = trigger {
+            next_cron_fire(schedule).unwrap_or(None)
+        } else {
+            None
+        };
+
+        let routine = Routine {
+            id: Uuid::new_v4(),
+            name: name.to_string(),
+            description: description.to_string(),
+            user_id: ctx.user_id.clone(),
+            enabled: true,
+            trigger,
+            action,
+            guardrails: RoutineGuardrails {
+                cooldown: Duration::from_secs(cooldown_secs),
+                max_concurrent: 1,
+                dedup_window: None,
+            },
+            notify: NotifyConfig::default(),
+            last_run_at: None,
+            next_fire_at: next_fire,
+            run_count: 0,
+            consecutive_failures: 0,
+            state: serde_json::json!({}),
+            created_at: Utc::now(),
+            updated_at: Utc::now(),
+        };
+
+        self.store
+            .create_routine(&routine)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("failed to create routine: {e}")))?;
+
+        // Refresh event cache if this is an event trigger
+        if routine.trigger.type_tag() == "event" {
+            self.engine.refresh_event_cache().await;
+        }
+
+        let result = serde_json::json!({
+            "id": routine.id.to_string(),
+            "name": routine.name,
+            "trigger_type": routine.trigger.type_tag(),
+            "next_fire_at": routine.next_fire_at.map(|t| t.to_rfc3339()),
+            "status": "created",
+        });
+
+        Ok(ToolOutput::success(result, start.elapsed()))
+    }
+
+    fn requires_sanitization(&self) -> bool {
+        false
+    }
+}
+
+// ==================== routine_list ====================
+
+pub struct RoutineListTool {
+    store: Arc<Store>,
+}
+
+impl RoutineListTool {
+    pub fn new(store: Arc<Store>) -> Self {
+        Self { store }
+    }
+}
+
+#[async_trait]
+impl Tool for RoutineListTool {
+    fn name(&self) -> &str {
+        "routine_list"
+    }
+
+    fn description(&self) -> &str {
+        "List all routines with their status, trigger info, and next fire time."
+    }
+
+    fn parameters_schema(&self) -> serde_json::Value {
+        serde_json::json!({
+            "type": "object",
+            "properties": {},
+            "required": []
+        })
+    }
+
+    async fn execute(
+        &self,
+        _params: serde_json::Value,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let start = std::time::Instant::now();
+
+        let routines = self
+            .store
+            .list_routines(&ctx.user_id)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("failed to list routines: {e}")))?;
+
+        let list: Vec<serde_json::Value> = routines
+            .iter()
+            .map(|r| {
+                serde_json::json!({
+                    "id": r.id.to_string(),
+                    "name": r.name,
+                    "description": r.description,
+                    "enabled": r.enabled,
+                    "trigger_type": r.trigger.type_tag(),
+                    "action_type": r.action.type_tag(),
+                    "last_run_at": r.last_run_at.map(|t| t.to_rfc3339()),
+                    "next_fire_at": r.next_fire_at.map(|t| t.to_rfc3339()),
+                    "run_count": r.run_count,
+                    "consecutive_failures": r.consecutive_failures,
+                })
+            })
+            .collect();
+
+        let result = serde_json::json!({
+            "count": list.len(),
+            "routines": list,
+        });
+
+        Ok(ToolOutput::success(result, start.elapsed()))
+    }
+
+    fn requires_sanitization(&self) -> bool {
+        false
+    }
+}
+
+// ==================== routine_update ====================
+
+pub struct RoutineUpdateTool {
+    store: Arc<Store>,
+    engine: Arc<RoutineEngine>,
+}
+
+impl RoutineUpdateTool {
+    pub fn new(store: Arc<Store>, engine: Arc<RoutineEngine>) -> Self {
+        Self { store, engine }
+    }
+}
+
+#[async_trait]
+impl Tool for RoutineUpdateTool {
+    fn name(&self) -> &str {
+        "routine_update"
+    }
+
+    fn description(&self) -> &str {
+        "Update an existing routine. Can modify trigger, prompt, schedule, or toggle enabled state. \
+         Pass the routine name and only the fields you want to change."
+    }
+
+    fn parameters_schema(&self) -> serde_json::Value {
+        serde_json::json!({
+            "type": "object",
+            "properties": {
+                "name": {
+                    "type": "string",
+                    "description": "Name of the routine to update"
+                },
+                "enabled": {
+                    "type": "boolean",
+                    "description": "Enable or disable the routine"
+                },
+                "prompt": {
+                    "type": "string",
+                    "description": "New prompt/instructions"
+                },
+                "schedule": {
+                    "type": "string",
+                    "description": "New cron schedule (for cron triggers)"
+                },
+                "description": {
+                    "type": "string",
+                    "description": "New description"
+                }
+            },
+            "required": ["name"]
+        })
+    }
+
+    async fn execute(
+        &self,
+        params: serde_json::Value,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let start = std::time::Instant::now();
+
+        let name = params
+            .get("name")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'name'".to_string()))?;
+
+        let mut routine = self
+            .store
+            .get_routine_by_name(&ctx.user_id, name)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("DB error: {e}")))?
+            .ok_or_else(|| ToolError::ExecutionFailed(format!("routine '{}' not found", name)))?;
+
+        // Apply updates
+        if let Some(enabled) = params.get("enabled").and_then(|v| v.as_bool()) {
+            routine.enabled = enabled;
+        }
+
+        if let Some(desc) = params.get("description").and_then(|v| v.as_str()) {
+            routine.description = desc.to_string();
+        }
+
+        if let Some(prompt) = params.get("prompt").and_then(|v| v.as_str()) {
+            match &mut routine.action {
+                RoutineAction::Lightweight { prompt: p, .. } => *p = prompt.to_string(),
+                RoutineAction::FullJob { description: d, .. } => *d = prompt.to_string(),
+            }
+        }
+
+        if let Some(schedule) = params.get("schedule").and_then(|v| v.as_str()) {
+            // Validate
+            next_cron_fire(schedule)
+                .map_err(|e| ToolError::InvalidParameters(format!("invalid cron schedule: {e}")))?;
+
+            routine.trigger = Trigger::Cron {
+                schedule: schedule.to_string(),
+            };
+            routine.next_fire_at = next_cron_fire(schedule).unwrap_or(None);
+        }
+
+        self.store
+            .update_routine(&routine)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("failed to update: {e}")))?;
+
+        // Refresh event cache in case trigger changed
+        self.engine.refresh_event_cache().await;
+
+        let result = serde_json::json!({
+            "name": routine.name,
+            "enabled": routine.enabled,
+            "trigger_type": routine.trigger.type_tag(),
+            "next_fire_at": routine.next_fire_at.map(|t| t.to_rfc3339()),
+            "status": "updated",
+        });
+
+        Ok(ToolOutput::success(result, start.elapsed()))
+    }
+
+    fn requires_sanitization(&self) -> bool {
+        false
+    }
+}
+
+// ==================== routine_delete ====================
+
+pub struct RoutineDeleteTool {
+    store: Arc<Store>,
+    engine: Arc<RoutineEngine>,
+}
+
+impl RoutineDeleteTool {
+    pub fn new(store: Arc<Store>, engine: Arc<RoutineEngine>) -> Self {
+        Self { store, engine }
+    }
+}
+
+#[async_trait]
+impl Tool for RoutineDeleteTool {
+    fn name(&self) -> &str {
+        "routine_delete"
+    }
+
+    fn description(&self) -> &str {
+        "Delete a routine permanently. This also removes all run history."
+    }
+
+    fn parameters_schema(&self) -> serde_json::Value {
+        serde_json::json!({
+            "type": "object",
+            "properties": {
+                "name": {
+                    "type": "string",
+                    "description": "Name of the routine to delete"
+                }
+            },
+            "required": ["name"]
+        })
+    }
+
+    async fn execute(
+        &self,
+        params: serde_json::Value,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let start = std::time::Instant::now();
+
+        let name = params
+            .get("name")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'name'".to_string()))?;
+
+        let routine = self
+            .store
+            .get_routine_by_name(&ctx.user_id, name)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("DB error: {e}")))?
+            .ok_or_else(|| ToolError::ExecutionFailed(format!("routine '{}' not found", name)))?;
+
+        let deleted = self
+            .store
+            .delete_routine(routine.id)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("failed to delete: {e}")))?;
+
+        // Refresh event cache
+        self.engine.refresh_event_cache().await;
+
+        let result = serde_json::json!({
+            "name": name,
+            "deleted": deleted,
+        });
+
+        Ok(ToolOutput::success(result, start.elapsed()))
+    }
+
+    fn requires_sanitization(&self) -> bool {
+        false
+    }
+}
+
+// ==================== routine_history ====================
+
+pub struct RoutineHistoryTool {
+    store: Arc<Store>,
+}
+
+impl RoutineHistoryTool {
+    pub fn new(store: Arc<Store>) -> Self {
+        Self { store }
+    }
+}
+
+#[async_trait]
+impl Tool for RoutineHistoryTool {
+    fn name(&self) -> &str {
+        "routine_history"
+    }
+
+    fn description(&self) -> &str {
+        "View the execution history of a routine. Shows recent runs with status, duration, and results."
+    }
+
+    fn parameters_schema(&self) -> serde_json::Value {
+        serde_json::json!({
+            "type": "object",
+            "properties": {
+                "name": {
+                    "type": "string",
+                    "description": "Name of the routine"
+                },
+                "limit": {
+                    "type": "integer",
+                    "description": "Max runs to return (default: 10)",
+                    "default": 10
+                }
+            },
+            "required": ["name"]
+        })
+    }
+
+    async fn execute(
+        &self,
+        params: serde_json::Value,
+        ctx: &JobContext,
+    ) -> Result<ToolOutput, ToolError> {
+        let start = std::time::Instant::now();
+
+        let name = params
+            .get("name")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| ToolError::InvalidParameters("missing 'name'".to_string()))?;
+
+        let limit = params
+            .get("limit")
+            .and_then(|v| v.as_i64())
+            .unwrap_or(10)
+            .min(50);
+
+        let routine = self
+            .store
+            .get_routine_by_name(&ctx.user_id, name)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("DB error: {e}")))?
+            .ok_or_else(|| ToolError::ExecutionFailed(format!("routine '{}' not found", name)))?;
+
+        let runs = self
+            .store
+            .list_routine_runs(routine.id, limit)
+            .await
+            .map_err(|e| ToolError::ExecutionFailed(format!("failed to list runs: {e}")))?;
+
+        let run_list: Vec<serde_json::Value> = runs
+            .iter()
+            .map(|r| {
+                let duration_secs = r
+                    .completed_at
+                    .map(|c| c.signed_duration_since(r.started_at).num_seconds());
+                serde_json::json!({
+                    "id": r.id.to_string(),
+                    "trigger_type": r.trigger_type,
+                    "trigger_detail": r.trigger_detail,
+                    "started_at": r.started_at.to_rfc3339(),
+                    "completed_at": r.completed_at.map(|t| t.to_rfc3339()),
+                    "duration_secs": duration_secs,
+                    "status": r.status.to_string(),
+                    "result_summary": r.result_summary,
+                    "tokens_used": r.tokens_used,
+                })
+            })
+            .collect();
+
+        let result = serde_json::json!({
+            "routine": name,
+            "total_runs": routine.run_count,
+            "runs": run_list,
+        });
+
+        Ok(ToolOutput::success(result, start.elapsed()))
+    }
+
+    fn requires_sanitization(&self) -> bool {
+        false
+    }
+}

src/tools/builtin/shell.rs

@@ -30,7 +30,7 @@ use tokio::process::Command;
 
 use crate::context::JobContext;
 use crate::sandbox::{SandboxManager, SandboxPolicy};
-use crate::tools::tool::{Tool, ToolError, ToolOutput};
+use crate::tools::tool::{Tool, ToolDomain, ToolError, ToolOutput};
 
 /// Maximum output size before truncation (64KB).
 const MAX_OUTPUT_SIZE: usize = 64 * 1024;
@@ -386,6 +386,10 @@ impl Tool for ShellTool {
     fn requires_sanitization(&self) -> bool {
         true // Shell output could contain anything
     }
+
+    fn domain(&self) -> ToolDomain {
+        ToolDomain::Container
+    }
 }
 
 /// Truncate output to fit within limits.

src/tools/mcp/config.rs

@@ -327,6 +327,81 @@ pub async fn get_mcp_server(name: &str) -> Result<McpServerConfig, ConfigError>
         })
 }
 
+// ==================== Database-backed MCP server config ====================
+
+/// Load MCP server configurations from the database settings table.
+///
+/// Falls back to the disk file if DB has no entry.
+pub async fn load_mcp_servers_from_db(
+    store: &crate::history::Store,
+    user_id: &str,
+) -> Result<McpServersFile, ConfigError> {
+    match store.get_setting(user_id, "mcp_servers").await {
+        Ok(Some(value)) => {
+            let config: McpServersFile = serde_json::from_value(value)?;
+            Ok(config)
+        }
+        Ok(None) => {
+            // No entry in DB, fall back to disk
+            load_mcp_servers().await
+        }
+        Err(e) => {
+            tracing::warn!(
+                "Failed to load MCP servers from DB: {}, falling back to disk",
+                e
+            );
+            load_mcp_servers().await
+        }
+    }
+}
+
+/// Save MCP server configurations to the database settings table.
+pub async fn save_mcp_servers_to_db(
+    store: &crate::history::Store,
+    user_id: &str,
+    config: &McpServersFile,
+) -> Result<(), ConfigError> {
+    let value = serde_json::to_value(config)?;
+    store
+        .set_setting(user_id, "mcp_servers", &value)
+        .await
+        .map_err(std::io::Error::other)?;
+    Ok(())
+}
+
+/// Add a new MCP server configuration (DB-backed).
+pub async fn add_mcp_server_db(
+    store: &crate::history::Store,
+    user_id: &str,
+    config: McpServerConfig,
+) -> Result<(), ConfigError> {
+    config.validate()?;
+
+    let mut servers = load_mcp_servers_from_db(store, user_id).await?;
+    servers.upsert(config);
+    save_mcp_servers_to_db(store, user_id, &servers).await?;
+
+    Ok(())
+}
+
+/// Remove an MCP server by name (DB-backed).
+pub async fn remove_mcp_server_db(
+    store: &crate::history::Store,
+    user_id: &str,
+    name: &str,
+) -> Result<(), ConfigError> {
+    let mut servers = load_mcp_servers_from_db(store, user_id).await?;
+
+    if !servers.remove(name) {
+        return Err(ConfigError::ServerNotFound {
+            name: name.to_string(),
+        });
+    }
+
+    save_mcp_servers_to_db(store, user_id, &servers).await?;
+    Ok(())
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

src/tools/mod.rs

@@ -23,4 +23,4 @@ pub use builder::{
 };
 pub use registry::ToolRegistry;
 pub use sandbox::ToolSandbox;
-pub use tool::{Tool, ToolError, ToolOutput};
+pub use tool::{Tool, ToolDomain, ToolError, ToolOutput};