chore(ironclaw): release v0.23.0 (#1658)

Co-authored-by: ironclaw-ci[bot] <266877842+ironclaw-ci[bot]@users.noreply.github.com>

.github/workflows/test.yml

@@ -12,6 +12,7 @@ jobs:
   tests:
     name: Tests (${{ matrix.name }})
     runs-on: ubuntu-latest
+    timeout-minutes: 45
     strategy:
       fail-fast: false
       matrix:
@@ -40,11 +41,14 @@ jobs:
       - name: Build WASM channels (for integration tests)
         run: ./scripts/build-wasm-extensions.sh --channels
       - name: Run Tests
-        run: cargo test ${{ matrix.flags }} -- --nocapture
+        run: |
+          timeout --signal=INT --kill-after=30s 40m \
+            cargo test ${{ matrix.flags }} -- --nocapture
 
   heavy-integration-tests:
     name: Heavy Integration Tests
     runs-on: ubuntu-latest
+    timeout-minutes: 20
     steps:
       - name: Checkout repository
         uses: actions/checkout@v6
@@ -58,9 +62,13 @@ jobs:
       - name: Build Telegram WASM channel
         run: cargo build --manifest-path channels-src/telegram/Cargo.toml --target wasm32-wasip2 --release
       - name: Run thread scheduling integration tests
-        run: cargo test --no-default-features --features libsql,integration --test e2e_thread_scheduling -- --nocapture
+        run: |
+          timeout --signal=INT --kill-after=30s 15m \
+            cargo test --no-default-features --features libsql,integration --test e2e_thread_scheduling -- --nocapture
       - name: Run Telegram thread-scope regression test
-        run: cargo test --features integration --test telegram_auth_integration test_private_messages_use_chat_id_as_thread_scope -- --exact
+        run: |
+          timeout --signal=INT --kill-after=30s 10m \
+            cargo test --features integration --test telegram_auth_integration test_private_messages_use_chat_id_as_thread_scope -- --exact
 
   telegram-tests:
     name: Telegram Channel Tests
@@ -68,6 +76,7 @@ jobs:
       github.event_name != 'pull_request' ||
       github.base_ref != 'staging'
     runs-on: ubuntu-latest
+    timeout-minutes: 15
     steps:
       - name: Checkout repository
         uses: actions/checkout@v6
@@ -75,7 +84,9 @@ jobs:
         uses: dtolnay/rust-toolchain@stable
       - uses: Swatinem/rust-cache@v2
       - name: Run Telegram Channel Tests
-        run: cargo test --manifest-path channels-src/telegram/Cargo.toml -- --nocapture
+        run: |
+          timeout --signal=INT --kill-after=30s 10m \
+            cargo test --manifest-path channels-src/telegram/Cargo.toml -- --nocapture
 
   windows-build:
     name: Windows Build (${{ matrix.name }})
@@ -110,6 +121,7 @@ jobs:
       github.event_name != 'pull_request' ||
       github.base_ref != 'staging'
     runs-on: ubuntu-latest
+    timeout-minutes: 30
     steps:
       - name: Checkout repository
         uses: actions/checkout@v6
@@ -125,7 +137,9 @@ jobs:
       - name: Build all WASM extensions against current WIT
         run: ./scripts/build-wasm-extensions.sh
       - name: Instantiation test (host linker compatibility)
-        run: cargo test --all-features wit_compat -- --nocapture
+        run: |
+          timeout --signal=INT --kill-after=30s 20m \
+            cargo test --all-features wit_compat -- --nocapture
 
   bench-compile:
     name: Benchmark Compilation

CHANGELOG.md

@@ -7,6 +7,156 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.23.0](https://github.com/nearai/ironclaw/compare/ironclaw-v0.22.0...ironclaw-v0.23.0) - 2026-03-27
+
+### Added
+
+- complete multi-tenant isolation — phases 2–4 ([#1614](https://github.com/nearai/ironclaw/pull/1614))
+
+### Fixed
+
+- *(routines)* recover delete name after failed update fallback ([#1108](https://github.com/nearai/ironclaw/pull/1108))
+- *(mcp)* handle 202 Accepted and wire session manager for Streamable HTTP ([#1437](https://github.com/nearai/ironclaw/pull/1437))
+- *(extensions)* channel-relay auth dead-end, observability, and URL override ([#1681](https://github.com/nearai/ironclaw/pull/1681))
+- *(agent)* discard truncated tool calls when finish_reason == Length ([#1631](https://github.com/nearai/ironclaw/pull/1631)) ([#1632](https://github.com/nearai/ironclaw/pull/1632))
+- *(llm)* filter XML tool-call recovery by context ([#1641](https://github.com/nearai/ironclaw/pull/1641))
+
+### Other
+
+- Support direct hosted OAuth callbacks with proxy auth token ([#1684](https://github.com/nearai/ironclaw/pull/1684))
+
+## [0.22.0](https://github.com/nearai/ironclaw/compare/ironclaw-v0.21.0...ironclaw-v0.22.0) - 2026-03-25
+
+### Added
+
+- *(agent)* thread per-tool reasoning through provider, session, and all surfaces ([#1513](https://github.com/nearai/ironclaw/pull/1513))
+- *(cli)* show credential auth status in tool info ([#1572](https://github.com/nearai/ironclaw/pull/1572))
+- multi-tenant auth with per-user workspace isolation ([#1118](https://github.com/nearai/ironclaw/pull/1118))
+- *(cli)* add ironclaw models subcommands (list/status/set/set-provider) ([#1043](https://github.com/nearai/ironclaw/pull/1043))
+- *(workspace)* multi-scope workspace reads ([#1117](https://github.com/nearai/ironclaw/pull/1117))
+- *(ux)* complete UX overhaul — design system, onboarding, web polish ([#1277](https://github.com/nearai/ironclaw/pull/1277))
+- *(gemini_oauth)* full Gemini CLI OAuth integration with Cloud Code API ([#1356](https://github.com/nearai/ironclaw/pull/1356))
+- *(shell)* add Low/Medium/High risk levels for graduated command approval (closes #172) ([#368](https://github.com/nearai/ironclaw/pull/368))
+- *(agent)* queue and merge messages during active turns ([#1412](https://github.com/nearai/ironclaw/pull/1412))
+- *(cli)* add `ironclaw hooks list` subcommand ([#1023](https://github.com/nearai/ironclaw/pull/1023))
+- *(extensions)* support text setup fields in web configure modal ([#496](https://github.com/nearai/ironclaw/pull/496))
+- *(llm)* add GitHub Copilot as LLM provider ([#1512](https://github.com/nearai/ironclaw/pull/1512))
+- *(workspace)* layered memory with sensitivity-based privacy redirect ([#1112](https://github.com/nearai/ironclaw/pull/1112))
+- *(webhooks)* add public webhook trigger endpoint for routines ([#736](https://github.com/nearai/ironclaw/pull/736))
+- *(llm)* Add OpenAI Codex (ChatGPT subscription) as LLM provider ([#1461](https://github.com/nearai/ironclaw/pull/1461))
+- *(web)* add light theme with dark/light/system toggle ([#1457](https://github.com/nearai/ironclaw/pull/1457))
+- *(agent)* activate stuck_threshold for time-based stuck job detection ([#1234](https://github.com/nearai/ironclaw/pull/1234))
+- chat onboarding and routine advisor ([#927](https://github.com/nearai/ironclaw/pull/927))
+
+### Fixed
+
+- ensure LLM calls always end with user message (closes #763) ([#1259](https://github.com/nearai/ironclaw/pull/1259))
+- restore owner-scoped gateway startup ([#1625](https://github.com/nearai/ironclaw/pull/1625))
+- remove stale stream_token gate from channel-relay activation ([#1623](https://github.com/nearai/ironclaw/pull/1623))
+- *(agent)* case-insensitive channel match and user_id filter for event triggers ([#1211](https://github.com/nearai/ironclaw/pull/1211))
+- *(routines)* normalize status display across web and CLI ([#1469](https://github.com/nearai/ironclaw/pull/1469))
+- *(tunnel)* managed tunnels target wrong port and die from SIGPIPE ([#1093](https://github.com/nearai/ironclaw/pull/1093))
+- *(agent)* persist /model selection to .env, TOML, and DB ([#1581](https://github.com/nearai/ironclaw/pull/1581))
+- post-merge review sweep — 8 fixes across security, perf, and correctness ([#1550](https://github.com/nearai/ironclaw/pull/1550))
+- generate Mistral-compatible 9-char alphanumeric tool call IDs ([#1242](https://github.com/nearai/ironclaw/pull/1242))
+- *(mcp)* handle empty 202 notification acknowledgements ([#1539](https://github.com/nearai/ironclaw/pull/1539))
+- *(tests)* eliminate env mutex poison cascade ([#1558](https://github.com/nearai/ironclaw/pull/1558))
+- *(safety)* escape tool output XML content and remove misleading sanitized attr ([#1067](https://github.com/nearai/ironclaw/pull/1067))
+- *(oauth)* reject malformed ic2.* states in decode_hosted_oauth_state ([#1441](https://github.com/nearai/ironclaw/pull/1441)) ([#1454](https://github.com/nearai/ironclaw/pull/1454))
+- parameter coercion and validation for oneOf/anyOf/allOf schemas ([#1397](https://github.com/nearai/ironclaw/pull/1397))
+- persist startup-loaded MCP clients in ExtensionManager ([#1509](https://github.com/nearai/ironclaw/pull/1509))
+- *(deps)* patch rustls-webpki vulnerability (RUSTSEC-2026-0049)
+- *(routines)* add missing extension_manager field in trigger_manual EngineContext
+- *(ci)* serialize env-mutating OAuth wildcard tests with ENV_MUTEX ([#1280](https://github.com/nearai/ironclaw/pull/1280)) ([#1468](https://github.com/nearai/ironclaw/pull/1468))
+- *(setup)* remove redundant LLM config and API keys from bootstrap .env ([#1448](https://github.com/nearai/ironclaw/pull/1448))
+- resolve wasm broadcast merge conflicts with staging ([#395](https://github.com/nearai/ironclaw/pull/395)) ([#1460](https://github.com/nearai/ironclaw/pull/1460))
+- skip credential validation for Bedrock backend ([#1011](https://github.com/nearai/ironclaw/pull/1011))
+- register sandbox jobs in ContextManager for query tool visibility ([#1426](https://github.com/nearai/ironclaw/pull/1426))
+- prefer execution-local message routing metadata ([#1449](https://github.com/nearai/ironclaw/pull/1449))
+- *(security)* validate embedding base URLs to prevent SSRF ([#1221](https://github.com/nearai/ironclaw/pull/1221))
+- f32→f64 precision artifact in temperature causes provider 400 errors ([#1450](https://github.com/nearai/ironclaw/pull/1450))
+- *(routines)* surface errors when sandbox unavailable for full_job routines ([#769](https://github.com/nearai/ironclaw/pull/769))
+- restore libSQL vector search with dynamic dimensions ([#1393](https://github.com/nearai/ironclaw/pull/1393))
+- staging CI triage — consolidate retry parsing, fix flaky tests, add docs ([#1427](https://github.com/nearai/ironclaw/pull/1427))
+
+### Other
+
+- Merge branch 'main' into staging-promote/455f543b-23329172268
+- Merge pull request #1655 from nearai/codex/fix-staging-promotion-1451-version-bumps
+- Merge pull request #1499 from nearai/staging-promote/9603fefd-23364438978
+- Fix libsql prompt scope regressions ([#1651](https://github.com/nearai/ironclaw/pull/1651))
+- Normalize cron schedules on routine create ([#1648](https://github.com/nearai/ironclaw/pull/1648))
+- Fix MCP lifecycle trace user scope ([#1646](https://github.com/nearai/ironclaw/pull/1646))
+- Fix REPL single-message hang and cap CI test duration ([#1643](https://github.com/nearai/ironclaw/pull/1643))
+- extract AppEvent to crates/ironclaw_common ([#1615](https://github.com/nearai/ironclaw/pull/1615))
+- Fix hosted OAuth refresh via proxy ([#1602](https://github.com/nearai/ironclaw/pull/1602))
+- *(agent)* optimize approval thread resolution (UUID parsing + lock contention) ([#1592](https://github.com/nearai/ironclaw/pull/1592))
+- *(tools)* auto-compact WASM tool schemas, add descriptions, improve credential prompts ([#1525](https://github.com/nearai/ironclaw/pull/1525))
+- Default new lightweight routines to tools-enabled ([#1573](https://github.com/nearai/ironclaw/pull/1573))
+- Google OAuth URL broken when initiated from Telegram channel ([#1165](https://github.com/nearai/ironclaw/pull/1165))
+- add gitcgr code graph badge ([#1563](https://github.com/nearai/ironclaw/pull/1563))
+- Fix owner-scoped message routing fallbacks ([#1574](https://github.com/nearai/ironclaw/pull/1574))
+- *(tools)* remove unconditional params clone in shared execution (fix #893) ([#926](https://github.com/nearai/ironclaw/pull/926))
+- *(llm)* move transcription module into src/llm/ ([#1559](https://github.com/nearai/ironclaw/pull/1559))
+- *(agent)* avoid preview allocations for non-truncated strings (fix #894) ([#924](https://github.com/nearai/ironclaw/pull/924))
+- Expand AGENTS.md with coding agents guidance ([#1392](https://github.com/nearai/ironclaw/pull/1392))
+- Fix CI approval flows and stale fixtures ([#1478](https://github.com/nearai/ironclaw/pull/1478))
+- Use live owner tool scope for autonomous routines and jobs ([#1453](https://github.com/nearai/ironclaw/pull/1453))
+- use Arc in embedding cache to avoid clones on miss path ([#1438](https://github.com/nearai/ironclaw/pull/1438))
+- Add owner-scoped permissions for full-job routines ([#1440](https://github.com/nearai/ironclaw/pull/1440))
+
+## [0.21.0](https://github.com/nearai/ironclaw/compare/v0.20.0...v0.21.0) - 2026-03-20
+
+### Added
+
+- structured fallback deliverables for failed/stuck jobs ([#236](https://github.com/nearai/ironclaw/pull/236))
+- LRU embedding cache for workspace search ([#1423](https://github.com/nearai/ironclaw/pull/1423))
+- receive relay events via webhook callbacks ([#1254](https://github.com/nearai/ironclaw/pull/1254))
+
+### Fixed
+
+- bump Feishu channel version for promotion
+- *(approval)* make "always" auto-approve work for credentialed HTTP requests ([#1257](https://github.com/nearai/ironclaw/pull/1257))
+- skip NEAR AI session check when backend is not nearai ([#1413](https://github.com/nearai/ironclaw/pull/1413))
+
+### Other
+
+- Make hosted OAuth and MCP auth generic ([#1375](https://github.com/nearai/ironclaw/pull/1375))
+
+## [0.20.0](https://github.com/nearai/ironclaw/compare/v0.19.0...v0.20.0) - 2026-03-19
+
+### Added
+
+- *(self-repair)* wire stuck_threshold, store, and builder ([#712](https://github.com/nearai/ironclaw/pull/712))
+- *(testing)* add FaultInjector framework for StubLlm ([#1233](https://github.com/nearai/ironclaw/pull/1233))
+- *(gateway)* unified settings page with subtabs ([#1191](https://github.com/nearai/ironclaw/pull/1191))
+- upgrade MiniMax default model to M2.7 ([#1357](https://github.com/nearai/ironclaw/pull/1357))
+
+### Fixed
+
+- navigate telegram E2E tests to channels subtab ([#1408](https://github.com/nearai/ironclaw/pull/1408))
+- add missing `builder` field and update E2E extensions tab navigation ([#1400](https://github.com/nearai/ironclaw/pull/1400))
+- remove debug_assert guards that panic on valid error paths ([#1385](https://github.com/nearai/ironclaw/pull/1385))
+- address valid review comments from PR #1359 ([#1380](https://github.com/nearai/ironclaw/pull/1380))
+- full_job routine runs stay running until linked job completion ([#1374](https://github.com/nearai/ironclaw/pull/1374))
+- full_job routine concurrency tracks linked job lifetime ([#1372](https://github.com/nearai/ironclaw/pull/1372))
+- remove -x from coverage pytest to prevent suite-blocking failures ([#1360](https://github.com/nearai/ironclaw/pull/1360))
+- add debug_assert invariant guards to critical code paths ([#1312](https://github.com/nearai/ironclaw/pull/1312))
+- *(mcp)* retry after missing session id errors ([#1355](https://github.com/nearai/ironclaw/pull/1355))
+- *(telegram)* preserve polling after secret-blocked updates ([#1353](https://github.com/nearai/ironclaw/pull/1353))
+- *(llm)* cap retry-after delays ([#1351](https://github.com/nearai/ironclaw/pull/1351))
+- *(setup)* remove nonexistent webhook secret command hint ([#1349](https://github.com/nearai/ironclaw/pull/1349))
+- Rate limiter returns retry after None instead of a duration ([#1269](https://github.com/nearai/ironclaw/pull/1269))
+
+### Other
+
+- bump telegram channel version to 0.2.5 ([#1410](https://github.com/nearai/ironclaw/pull/1410))
+- *(ci)* enforce test requirement for state machine and resilience changes ([#1230](https://github.com/nearai/ironclaw/pull/1230)) ([#1304](https://github.com/nearai/ironclaw/pull/1304))
+- Fix duplicate LLM responses for matched event routines ([#1275](https://github.com/nearai/ironclaw/pull/1275))
+- add Japanese README ([#1306](https://github.com/nearai/ironclaw/pull/1306))
+- *(ci)* add coverage gates via codecov.yml ([#1228](https://github.com/nearai/ironclaw/pull/1228)) ([#1291](https://github.com/nearai/ironclaw/pull/1291))
+- Redesign routine create requests for LLMs ([#1147](https://github.com/nearai/ironclaw/pull/1147))
+
 ## [0.19.0](https://github.com/nearai/ironclaw/compare/v0.18.0...v0.19.0) - 2026-03-17
 
 ### Added

Cargo.lock

@@ -2323,7 +2323,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "39cab71617ae0d63f51a36d69f866391735b51691dbda63cf6f96d042b63efeb"
 dependencies = [
  "libc",
- "windows-sys 0.52.0",
+ "windows-sys 0.59.0",
 ]
 
 [[package]]
@@ -3390,7 +3390,7 @@ dependencies = [
 
 [[package]]
 name = "ironclaw"
-version = "0.19.0"
+version = "0.23.0"
 dependencies = [
  "aes-gcm",
  "aho-corasick",
@@ -3428,6 +3428,7 @@ dependencies = [
  "hyper-util",
  "iana-time-zone",
  "insta",
+ "ironclaw_common",
  "ironclaw_safety",
  "json5",
  "libsql",
@@ -3486,8 +3487,16 @@ dependencies = [
 ]
 
 [[package]]
-name = "ironclaw_safety"
+name = "ironclaw_common"
 version = "0.1.0"
+dependencies = [
+ "serde",
+ "serde_json",
+]
+
+[[package]]
+name = "ironclaw_safety"
+version = "0.2.0"
 dependencies = [
  "aho-corasick",
  "regex",
@@ -5472,7 +5481,7 @@ dependencies = [
  "errno",
  "libc",
  "linux-raw-sys 0.12.1",
- "windows-sys 0.52.0",
+ "windows-sys 0.59.0",
 ]
 
 [[package]]
@@ -6379,7 +6388,7 @@ dependencies = [
  "getrandom 0.4.2",
  "once_cell",
  "rustix 1.1.4",
- "windows-sys 0.52.0",
+ "windows-sys 0.59.0",
 ]
 
 [[package]]

Cargo.toml

@@ -1,5 +1,5 @@
 [workspace]
-members = [".", "crates/ironclaw_safety"]
+members = [".", "crates/ironclaw_common", "crates/ironclaw_safety"]
 exclude = [
     "channels-src/discord",
     "channels-src/telegram",
@@ -20,7 +20,7 @@ exclude = [
 
 [package]
 name = "ironclaw"
-version = "0.19.0"
+version = "0.23.0"
 edition = "2024"
 rust-version = "1.92"
 description = "Secure personal AI assistant that protects your data and expands its capabilities on the fly"
@@ -100,8 +100,11 @@ tower-http = { version = "0.6", features = ["trace", "cors", "set-header"] }
 # Cron scheduling for routines
 cron = "0.13"
 
+# Shared types
+ironclaw_common = { path = "crates/ironclaw_common", version = "0.1.0" }
+
 # Safety/sanitization
-ironclaw_safety = { path = "crates/ironclaw_safety", version = "0.1.0" }
+ironclaw_safety = { path = "crates/ironclaw_safety", version = "0.2.0" }
 regex = "1"
 aho-corasick = "1"
 

channels-src/feishu/feishu.capabilities.json

@@ -21,7 +21,7 @@
       },
       {
         "name": "feishu_app_secret",
-        "prompt": "Enter your Feishu/Lark App Secret",
+        "prompt": "Enter your Feishu/Lark App Secret (from your app settings at open.feishu.cn)",
         "optional": false
       },
       {

crates/ironclaw_common/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "ironclaw_common"
+version = "0.1.0"
+edition = "2024"
+rust-version = "1.92"
+description = "Shared types and utilities for the IronClaw workspace"
+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.dist]
+dist = false
+
+[dependencies]
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"

crates/ironclaw_common/src/event.rs

@@ -0,0 +1,393 @@
+//! Application-wide event types.
+//!
+//! `AppEvent` is the real-time event protocol used across the entire
+//! application.  The web gateway serialises these to SSE / WebSocket
+//! frames, but other subsystems (agent loop, orchestrator, extensions)
+//! produce and consume them too.
+
+use serde::{Deserialize, Serialize};
+
+/// A single tool decision in a reasoning update (SSE DTO).
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ToolDecisionDto {
+    pub tool_name: String,
+    pub rationale: String,
+}
+
+impl ToolDecisionDto {
+    /// Parse a list of tool decisions from a JSON array value.
+    pub fn from_json_array(value: &serde_json::Value) -> Vec<Self> {
+        value
+            .as_array()
+            .map(|arr| {
+                arr.iter()
+                    .filter_map(|d| {
+                        Some(Self {
+                            tool_name: d.get("tool_name")?.as_str()?.to_string(),
+                            rationale: d.get("rationale")?.as_str()?.to_string(),
+                        })
+                    })
+                    .collect()
+            })
+            .unwrap_or_default()
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(tag = "type")]
+pub enum AppEvent {
+    #[serde(rename = "response")]
+    Response { content: String, thread_id: String },
+    #[serde(rename = "thinking")]
+    Thinking {
+        message: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+    #[serde(rename = "tool_started")]
+    ToolStarted {
+        name: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+    #[serde(rename = "tool_completed")]
+    ToolCompleted {
+        name: String,
+        success: bool,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        error: Option<String>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        parameters: Option<String>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+    #[serde(rename = "tool_result")]
+    ToolResult {
+        name: String,
+        preview: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+    #[serde(rename = "stream_chunk")]
+    StreamChunk {
+        content: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+    #[serde(rename = "status")]
+    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,
+        tool_name: String,
+        description: String,
+        parameters: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+        /// Whether the "always" auto-approve option should be shown.
+        allow_always: bool,
+    },
+    #[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,
+        #[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>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        fallback_deliverable: Option<serde_json::Value>,
+    },
+
+    /// An image was generated by a tool.
+    #[serde(rename = "image_generated")]
+    ImageGenerated {
+        data_url: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        path: Option<String>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+
+    /// Suggested follow-up messages for the user.
+    #[serde(rename = "suggestions")]
+    Suggestions {
+        suggestions: Vec<String>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+
+    /// Per-turn token usage and cost summary.
+    #[serde(rename = "turn_cost")]
+    TurnCost {
+        input_tokens: u64,
+        output_tokens: u64,
+        cost_usd: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+
+    /// Extension activation status change (WASM channels).
+    #[serde(rename = "extension_status")]
+    ExtensionStatus {
+        extension_name: String,
+        status: String,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        message: Option<String>,
+    },
+
+    /// Agent reasoning update (why it chose specific tools).
+    #[serde(rename = "reasoning_update")]
+    ReasoningUpdate {
+        narrative: String,
+        decisions: Vec<ToolDecisionDto>,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        thread_id: Option<String>,
+    },
+
+    /// Reasoning update for a sandbox job.
+    #[serde(rename = "job_reasoning")]
+    JobReasoning {
+        job_id: String,
+        narrative: String,
+        decisions: Vec<ToolDecisionDto>,
+    },
+}
+
+impl AppEvent {
+    /// The wire-format event type string (matches the `#[serde(rename)]` value).
+    pub fn event_type(&self) -> &'static str {
+        match self {
+            Self::Response { .. } => "response",
+            Self::Thinking { .. } => "thinking",
+            Self::ToolStarted { .. } => "tool_started",
+            Self::ToolCompleted { .. } => "tool_completed",
+            Self::ToolResult { .. } => "tool_result",
+            Self::StreamChunk { .. } => "stream_chunk",
+            Self::Status { .. } => "status",
+            Self::JobStarted { .. } => "job_started",
+            Self::ApprovalNeeded { .. } => "approval_needed",
+            Self::AuthRequired { .. } => "auth_required",
+            Self::AuthCompleted { .. } => "auth_completed",
+            Self::Error { .. } => "error",
+            Self::Heartbeat => "heartbeat",
+            Self::JobMessage { .. } => "job_message",
+            Self::JobToolUse { .. } => "job_tool_use",
+            Self::JobToolResult { .. } => "job_tool_result",
+            Self::JobStatus { .. } => "job_status",
+            Self::JobResult { .. } => "job_result",
+            Self::ImageGenerated { .. } => "image_generated",
+            Self::Suggestions { .. } => "suggestions",
+            Self::TurnCost { .. } => "turn_cost",
+            Self::ExtensionStatus { .. } => "extension_status",
+            Self::ReasoningUpdate { .. } => "reasoning_update",
+            Self::JobReasoning { .. } => "job_reasoning",
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// Verify that `event_type()` returns the same string as the serde
+    /// `"type"` field for every variant.  This catches drift between the
+    /// `#[serde(rename)]` attributes and the manual match arms.
+    #[test]
+    fn event_type_matches_serde_type_field() {
+        let variants: Vec<AppEvent> = vec![
+            AppEvent::Response {
+                content: String::new(),
+                thread_id: String::new(),
+            },
+            AppEvent::Thinking {
+                message: String::new(),
+                thread_id: None,
+            },
+            AppEvent::ToolStarted {
+                name: String::new(),
+                thread_id: None,
+            },
+            AppEvent::ToolCompleted {
+                name: String::new(),
+                success: true,
+                error: None,
+                parameters: None,
+                thread_id: None,
+            },
+            AppEvent::ToolResult {
+                name: String::new(),
+                preview: String::new(),
+                thread_id: None,
+            },
+            AppEvent::StreamChunk {
+                content: String::new(),
+                thread_id: None,
+            },
+            AppEvent::Status {
+                message: String::new(),
+                thread_id: None,
+            },
+            AppEvent::JobStarted {
+                job_id: String::new(),
+                title: String::new(),
+                browse_url: String::new(),
+            },
+            AppEvent::ApprovalNeeded {
+                request_id: String::new(),
+                tool_name: String::new(),
+                description: String::new(),
+                parameters: String::new(),
+                thread_id: None,
+                allow_always: false,
+            },
+            AppEvent::AuthRequired {
+                extension_name: String::new(),
+                instructions: None,
+                auth_url: None,
+                setup_url: None,
+            },
+            AppEvent::AuthCompleted {
+                extension_name: String::new(),
+                success: true,
+                message: String::new(),
+            },
+            AppEvent::Error {
+                message: String::new(),
+                thread_id: None,
+            },
+            AppEvent::Heartbeat,
+            AppEvent::JobMessage {
+                job_id: String::new(),
+                role: String::new(),
+                content: String::new(),
+            },
+            AppEvent::JobToolUse {
+                job_id: String::new(),
+                tool_name: String::new(),
+                input: serde_json::Value::Null,
+            },
+            AppEvent::JobToolResult {
+                job_id: String::new(),
+                tool_name: String::new(),
+                output: String::new(),
+            },
+            AppEvent::JobStatus {
+                job_id: String::new(),
+                message: String::new(),
+            },
+            AppEvent::JobResult {
+                job_id: String::new(),
+                status: String::new(),
+                session_id: None,
+                fallback_deliverable: None,
+            },
+            AppEvent::ImageGenerated {
+                data_url: String::new(),
+                path: None,
+                thread_id: None,
+            },
+            AppEvent::Suggestions {
+                suggestions: vec![],
+                thread_id: None,
+            },
+            AppEvent::TurnCost {
+                input_tokens: 0,
+                output_tokens: 0,
+                cost_usd: String::new(),
+                thread_id: None,
+            },
+            AppEvent::ExtensionStatus {
+                extension_name: String::new(),
+                status: String::new(),
+                message: None,
+            },
+            AppEvent::ReasoningUpdate {
+                narrative: String::new(),
+                decisions: vec![],
+                thread_id: None,
+            },
+            AppEvent::JobReasoning {
+                job_id: String::new(),
+                narrative: String::new(),
+                decisions: vec![],
+            },
+        ];
+
+        for variant in &variants {
+            let json: serde_json::Value = serde_json::to_value(variant).unwrap();
+            let serde_type = json["type"].as_str().unwrap();
+            assert_eq!(
+                variant.event_type(),
+                serde_type,
+                "event_type() mismatch for variant: {:?}",
+                variant
+            );
+        }
+    }
+
+    #[test]
+    fn round_trip_deserialize() {
+        let original = AppEvent::Response {
+            content: "hello".to_string(),
+            thread_id: "t1".to_string(),
+        };
+        let json = serde_json::to_string(&original).unwrap();
+        let deserialized: AppEvent = serde_json::from_str(&json).unwrap();
+        assert_eq!(deserialized.event_type(), "response");
+    }
+}

crates/ironclaw_common/src/lib.rs

@@ -0,0 +1,7 @@
+//! Shared types and utilities for the IronClaw workspace.
+
+mod event;
+mod util;
+
+pub use event::{AppEvent, ToolDecisionDto};
+pub use util::truncate_preview;

crates/ironclaw_common/src/util.rs

@@ -0,0 +1,100 @@
+//! Shared utility functions.
+
+/// Truncate a string to at most `max_bytes` bytes at a char boundary, appending "...".
+///
+/// If the input is wrapped in `<tool_output ...>...</tool_output>` and truncation
+/// removes the closing tag, the tag is re-appended so downstream XML parsers
+/// never see an unclosed element.
+pub fn truncate_preview(s: &str, max_bytes: usize) -> String {
+    if s.len() <= max_bytes {
+        return s.to_string();
+    }
+    // Walk backwards from max_bytes to find a valid char boundary
+    let mut end = max_bytes;
+    while end > 0 && !s.is_char_boundary(end) {
+        end -= 1;
+    }
+    let mut result = format!("{}...", &s[..end]);
+
+    // Re-close <tool_output> if truncation cut through the closing tag.
+    if s.starts_with("<tool_output") && !result.ends_with("</tool_output>") {
+        result.push_str("\n</tool_output>");
+    }
+
+    result
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_truncate_preview_short_string() {
+        assert_eq!(truncate_preview("hello", 10), "hello");
+    }
+
+    #[test]
+    fn test_truncate_preview_exact_boundary() {
+        assert_eq!(truncate_preview("hello", 5), "hello");
+    }
+
+    #[test]
+    fn test_truncate_preview_truncates_ascii() {
+        assert_eq!(truncate_preview("hello world", 5), "hello...");
+    }
+
+    #[test]
+    fn test_truncate_preview_empty_string() {
+        assert_eq!(truncate_preview("", 10), "");
+    }
+
+    #[test]
+    fn test_truncate_preview_multibyte_char_boundary() {
+        let s = "a\u{20AC}b";
+        let result = truncate_preview(s, 3);
+        assert_eq!(result, "a...");
+    }
+
+    #[test]
+    fn test_truncate_preview_emoji() {
+        let s = "hi\u{1F980}";
+        let result = truncate_preview(s, 4);
+        assert_eq!(result, "hi...");
+    }
+
+    #[test]
+    fn test_truncate_preview_cjk() {
+        let s = "\u{4F60}\u{597D}\u{4E16}\u{754C}";
+        let result = truncate_preview(s, 7);
+        assert_eq!(result, "\u{4F60}\u{597D}...");
+    }
+
+    #[test]
+    fn test_truncate_preview_zero_max_bytes() {
+        assert_eq!(truncate_preview("hello", 0), "...");
+    }
+
+    #[test]
+    fn test_truncate_preview_closes_tool_output_tag() {
+        let s = "<tool_output name=\"search\">\nSome very long content here\n</tool_output>";
+        let result = truncate_preview(s, 60);
+        assert!(result.ends_with("</tool_output>"));
+        assert!(result.contains("..."));
+    }
+
+    #[test]
+    fn test_truncate_preview_no_extra_close_when_intact() {
+        let s = "<tool_output name=\"echo\">\nshort\n</tool_output>";
+        let result = truncate_preview(s, 500);
+        assert_eq!(result, s);
+        assert_eq!(result.matches("</tool_output>").count(), 1);
+    }
+
+    #[test]
+    fn test_truncate_preview_non_xml_unaffected() {
+        let s = "Just a plain long string that gets truncated";
+        let result = truncate_preview(s, 10);
+        assert_eq!(result, "Just a pla...");
+        assert!(!result.contains("</tool_output>"));
+    }
+}

crates/ironclaw_safety/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "ironclaw_safety"
-version = "0.1.0"
+version = "0.2.0"
 edition = "2024"
 rust-version = "1.92"
 description = "Prompt injection defense, input validation, secret leak detection, and safety policy enforcement"
@@ -8,7 +8,6 @@ authors = ["NEAR AI <[email protected]>"]
 license = "MIT OR Apache-2.0"
 homepage = "https://github.com/nearai/ironclaw"
 repository = "https://github.com/nearai/ironclaw"
-publish = false
 
 [package.metadata.dist]
 dist = false

registry/channels/feishu.json

@@ -2,7 +2,7 @@
   "name": "feishu",
   "display_name": "Feishu / Lark Channel",
   "kind": "channel",
-  "version": "0.1.1",
+  "version": "0.1.3",
   "wit_version": "0.3.0",
   "description": "Talk to your agent through a Feishu or Lark bot",
   "keywords": [
@@ -19,8 +19,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "sha256": "5fca74022264d1c8e78a0853766276f7ffa3cf0d8065b2f51ca10985acad4714",
+      "sha256": "a66ff0dafb67d2216d8161bb7e96e724a94acb0ab993b85d2782d30412f8fe94",
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/channel-feishu-0.1.1-wasm32-wasip2.tar.gz"
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/channel-feishu-0.1.3-wasm32-wasip2.tar.gz"
     }
   },
   "auth_summary": {

registry/channels/telegram.json

@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/channel-telegram-0.2.4-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/v0.20.0/channel-telegram-0.2.5-wasm32-wasip2.tar.gz",
-      "sha256": "a7cb300ec1c946831cfceaa95c1dc8f30d0f42a3924f3cb5de8098821573f4b8"
+      "sha256": "1ef20a538f55b379e049356e4d6758006251846bc3365ceaa1c87eba8379a329"
     }
   },
   "auth_summary": {

registry/tools/github.json

@@ -2,7 +2,7 @@
   "name": "github",
   "display_name": "GitHub",
   "kind": "tool",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "wit_version": "0.3.0",
   "description": "GitHub integration for issues, PRs, repos, and code search",
   "keywords": [
@@ -19,8 +19,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/tool-github-0.2.1-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-github-0.2.2-wasm32-wasip2.tar.gz",
-      "sha256": "92c530b3ad172e2372d819744b5233f1d8f65768e26eb5a6c213eba3ce1de758"
+      "sha256": "70b55af593193d8fa495c0f702ea23284d83a624124f8a5f7564916ec5032c3f"
     }
   },
   "auth_summary": {

registry/tools/gmail.json

@@ -2,7 +2,7 @@
   "name": "gmail",
   "display_name": "Gmail",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Read, send, and manage Gmail messages and threads",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.18.0/gmail-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-gmail-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "ee9574e02e92bc1d481f1310eb88afd99ee52bf6971074ab33bd76bf99b34b1d"
+      "sha256": "79025b40ee70ce1120acc4320bae50da095d7afb0ef67bd56d99b064b72ea779"
     }
   },
   "auth_summary": {

registry/tools/google-calendar.json

@@ -2,7 +2,7 @@
   "name": "google-calendar",
   "display_name": "Google Calendar",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Create, read, update, and delete Google Calendar events",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.18.0/google-calendar-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-google-calendar-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "2fa47150ea222e787c122182ad6f4dfa2ffaf5fe490d05e8de887a76445f8d2d"
+      "sha256": "86bcc075010b08f5ab2f98f504cec1c6c9e0ca144857d185cbecf72a11f504bf"
     }
   },
   "auth_summary": {

registry/tools/google-docs.json

@@ -2,7 +2,7 @@
   "name": "google-docs",
   "display_name": "Google Docs",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Create and edit Google Docs documents",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.18.0/google-docs-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-google-docs-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "40e134a1c1564f832ca861c3396895d4e33ec67b99313fc1f97baf8d971423a9"
+      "sha256": "39d476029764949498a53a6a223f9952b5f4df151be7b8b19bf3fe4d401a57cd"
     }
   },
   "auth_summary": {

registry/tools/google-drive.json

@@ -2,7 +2,7 @@
   "name": "google-drive",
   "display_name": "Google Drive",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Upload, download, search, and manage Google Drive files and folders",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.18.0/google-drive-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-google-drive-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "002a341a1d58125563a7c69561b26fbc2629b04ea723cade744102bdc0fbb71f"
+      "sha256": "6e9a700fab93865c852af718666af64c5b534ad6a419fb4b736e07740188f494"
     }
   },
   "auth_summary": {

registry/tools/google-sheets.json

@@ -2,7 +2,7 @@
   "name": "google-sheets",
   "display_name": "Google Sheets",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Read and write Google Sheets spreadsheet data",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.18.0/google-sheets-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-google-sheets-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "8aa2c9d52f033edea3a6c2311b0ec694ccb6d0a54ef07e94d72bf8be1ce8009a"
+      "sha256": "1f8c381799a916be83263cac9d497d52946e21b1b588592a3a42ca94a73b7051"
     }
   },
   "auth_summary": {

registry/tools/google-slides.json

@@ -2,7 +2,7 @@
   "name": "google-slides",
   "display_name": "Google Slides",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Create and edit Google Slides presentations",
   "keywords": [
@@ -17,8 +17,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.18.0/google-slides-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-google-slides-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "e931a97d4fd0b0b938e464dc7c7f2be6ea6b4d1508f5ea3cd931d44db23f05f5"
+      "sha256": "e2528be5da02f1b8cfc8ee9b0cdd849516c53d412e2f75c6175b3bded7f512cb"
     }
   },
   "auth_summary": {

registry/tools/llm-context.json

@@ -2,7 +2,7 @@
   "name": "llm-context",
   "display_name": "LLM Context",
   "kind": "tool",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "wit_version": "0.3.0",
   "description": "Fetch pre-extracted web content from Brave Search for grounding LLM answers (RAG, fact-checking)",
   "keywords": [
@@ -21,8 +21,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/tool-llm-context-0.1.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-llm-context-0.1.1-wasm32-wasip2.tar.gz",
-      "sha256": "d9ced2b1226b879135891e0ee40e072c7c95412e1b2462925a23853e1f92497e"
+      "sha256": "9b19e2fd05dbbbe3c8bd55309a91db09124e8415eb0f767828b6e10b55771e63"
     }
   },
   "auth_summary": {

registry/tools/slack.json

@@ -2,7 +2,7 @@
   "name": "slack-tool",
   "display_name": "Slack Tool",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Your agent uses Slack to post and read messages in your workspace",
   "keywords": [
@@ -17,8 +17,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/tool-slack-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-slack-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "ccfb0415d7a04f9497726c712d15216de36e86f498b849101283c017f5ab4efb"
+      "sha256": "927519e5b7734beeb022d3b8bbd152e0e6b9f67c9452a8ad47809d3c4221a137"
     }
   },
   "auth_summary": {

registry/tools/telegram.json

@@ -2,7 +2,7 @@
   "name": "telegram-mtproto",
   "display_name": "Telegram Tool",
   "kind": "tool",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "wit_version": "0.3.0",
   "description": "Your agent uses your Telegram account to read and send messages",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/tool-telegram-0.2.0-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-telegram-0.2.1-wasm32-wasip2.tar.gz",
-      "sha256": "c17065ca41fae5f2a7c43b36144686718cd310a2f22442313bb1aa82bbad0ae4"
+      "sha256": "1e57d0755fc9c7b3ec013d079f30168898b484a6919f9edd105f0cd80131c1cd"
     }
   },
   "auth_summary": {

registry/tools/web-search.json

@@ -2,7 +2,7 @@
   "name": "web-search",
   "display_name": "Web Search",
   "kind": "tool",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "wit_version": "0.3.0",
   "description": "Search the web using Brave Search API",
   "keywords": [
@@ -18,8 +18,8 @@
   },
   "artifacts": {
     "wasm32-wasip2": {
-      "url": "https://github.com/nearai/ironclaw/releases/download/v0.19.0/tool-web-search-0.2.1-wasm32-wasip2.tar.gz",
+      "url": "https://github.com/nearai/ironclaw/releases/download/ironclaw-v0.22.0/tool-web-search-0.2.2-wasm32-wasip2.tar.gz",
-      "sha256": "bad275ca4ec314adea5241d6b92c44ccf9cebcbca8e30ba2493cc0bcb4b57218"
+      "sha256": "47382b50c1ea7525b20d59dc02fab04e336d018665826c2f24710bdf460779ae"
     }
   },
   "auth_summary": {

release-plz.toml

@@ -1,7 +1,2 @@
 [workspace]
 git_release_enable = false
-
-[[package]]
-name = "ironclaw_safety"
-publish = false
-release = false

src/agent/agent_loop.rs

@@ -13,9 +13,10 @@ use futures::StreamExt;
 use uuid::Uuid;
 
 use crate::agent::context_monitor::ContextMonitor;
-use crate::agent::heartbeat::spawn_heartbeat;
+use crate::agent::heartbeat::{spawn_heartbeat, spawn_multi_user_heartbeat};
 use crate::agent::routine_engine::{RoutineEngine, spawn_cron_ticker};
 use crate::agent::self_repair::{DefaultSelfRepair, RepairResult, SelfRepair};
+use crate::agent::session::ThreadState;
 use crate::agent::session_manager::SessionManager;
 use crate::agent::submission::{Submission, SubmissionParser, SubmissionResult};
 use crate::agent::{HeartbeatConfig as AgentHeartbeatConfig, Router, Scheduler, SchedulerDeps};
@@ -84,6 +85,15 @@ fn resolve_owner_scope_notification_user(
     trimmed_option(explicit_user).or_else(|| trimmed_option(owner_fallback))
 }
 
+fn is_single_message_repl(message: &IncomingMessage) -> bool {
+    message.channel == "repl"
+        && message
+            .metadata
+            .get("single_message_mode")
+            .and_then(|value| value.as_bool())
+            .unwrap_or(false)
+}
+
 async fn resolve_channel_notification_user(
     extension_manager: Option<&Arc<ExtensionManager>>,
     channel: Option<&str>,
@@ -157,8 +167,8 @@ pub struct AgentDeps {
     pub hooks: Arc<HookRegistry>,
     /// Cost enforcement guardrails (daily budget, hourly rate limits).
     pub cost_guard: Arc<crate::agent::cost_guard::CostGuard>,
-    /// SSE broadcast sender for live job event streaming to the web gateway.
+    /// SSE manager for live job event streaming to the web gateway.
-    pub sse_tx: Option<tokio::sync::broadcast::Sender<crate::channels::web::types::SseEvent>>,
+    pub sse_tx: Option<Arc<crate::channels::web::sse::SseManager>>,
     /// HTTP interceptor for trace recording/replay.
     pub http_interceptor: Option<Arc<dyn crate::llm::recording::HttpInterceptor>>,
     /// Audio transcription middleware for voice messages.
@@ -169,6 +179,11 @@ pub struct AgentDeps {
     pub sandbox_readiness: crate::agent::routine_engine::SandboxReadiness,
     /// Software builder for self-repair tool rebuilding.
     pub builder: Option<Arc<dyn crate::tools::SoftwareBuilder>>,
+    /// Resolved LLM backend identifier (e.g., "nearai", "openai", "groq").
+    /// Used by `/model` persistence to determine which env var to update.
+    pub llm_backend: String,
+    /// Per-tenant rate limiting registry (lazily creates rate state per user).
+    pub tenant_rates: Arc<crate::tenant::TenantRateRegistry>,
 }
 
 /// The main agent that coordinates all components.
@@ -231,12 +246,15 @@ impl Agent {
             SchedulerDeps {
                 tools: deps.tools.clone(),
                 extension_manager: deps.extension_manager.clone(),
-                store: deps.store.clone(),
+                store: deps
+                    .store
+                    .as_ref()
+                    .map(|db| crate::tenant::AdminScope::new(Arc::clone(db))),
                 hooks: deps.hooks.clone(),
             },
         );
-        if let Some(ref tx) = deps.sse_tx {
+        if let Some(ref sse) = deps.sse_tx {
-            scheduler.set_sse_sender(tx.clone());
+            scheduler.set_sse_sender(Arc::clone(sse));
         }
         if let Some(ref interceptor) = deps.http_interceptor {
             scheduler.set_http_interceptor(Arc::clone(interceptor));
@@ -312,6 +330,50 @@ impl Agent {
         &self.deps.cost_guard
     }
 
+    /// Build a tenant-scoped execution context for the given user.
+    ///
+    /// This is the standard entry point for per-user operations. The returned
+    /// [`TenantCtx`] provides a [`TenantScope`] that auto-binds `user_id` on
+    /// every database operation and a per-user rate limiter.
+    pub(super) async fn tenant_ctx(&self, user_id: &str) -> crate::tenant::TenantCtx {
+        let rate = self.deps.tenant_rates.get_or_create(user_id).await;
+
+        let store = self
+            .deps
+            .store
+            .as_ref()
+            .map(|db| crate::tenant::TenantScope::new(user_id, Arc::clone(db)));
+
+        // Reuse the owner workspace if user matches, otherwise create per-user.
+        let workspace = match &self.deps.workspace {
+            Some(ws) if ws.user_id() == user_id => Some(Arc::clone(ws)),
+            _ => self
+                .deps
+                .store
+                .as_ref()
+                .map(|db| Arc::new(Workspace::new_with_db(user_id, Arc::clone(db)))),
+        };
+
+        crate::tenant::TenantCtx::new(
+            user_id,
+            store,
+            workspace,
+            Arc::clone(&self.deps.cost_guard),
+            rate,
+        )
+    }
+
+    /// Get an admin-scoped database accessor for cross-tenant operations.
+    ///
+    /// Only for system-level components (heartbeat, routine engine, self-repair,
+    /// scheduler). Handler code should use [`tenant_ctx()`](Self::tenant_ctx) instead.
+    pub(super) fn admin_store(&self) -> Option<crate::tenant::AdminScope> {
+        self.deps
+            .store
+            .as_ref()
+            .map(|db| crate::tenant::AdminScope::new(Arc::clone(db)))
+    }
+
     pub(super) fn skill_registry(&self) -> Option<&Arc<std::sync::RwLock<SkillRegistry>>> {
         self.deps.skill_registry.as_ref()
     }
@@ -397,8 +459,8 @@ impl Agent {
             self.config.stuck_threshold,
             self.config.max_repair_attempts,
         );
-        if let Some(ref store) = self.deps.store {
+        if let Some(admin) = self.admin_store() {
-            self_repair = self_repair.with_store(Arc::clone(store));
+            self_repair = self_repair.with_store(admin);
         }
         if let Some(ref builder) = self.deps.builder {
             self_repair = self_repair.with_builder(Arc::clone(builder), Arc::clone(self.tools()));
@@ -505,6 +567,7 @@ impl Agent {
                         .with_interval(std::time::Duration::from_secs(hb_config.interval_secs));
                     config.quiet_hours_start = hb_config.quiet_hours_start;
                     config.quiet_hours_end = hb_config.quiet_hours_end;
+                    config.multi_tenant = hb_config.multi_tenant;
                     config.timezone = hb_config
                         .timezone
                         .clone()
@@ -534,30 +597,52 @@ impl Agent {
                     .await;
                     let notify_user = heartbeat_notify_user;
                     let channels = self.channels.clone();
+                    let is_multi_tenant = hb_config.multi_tenant;
                     tokio::spawn(async move {
                         while let Some(response) = notify_rx.recv().await {
+                            // In multi-tenant mode, extract the owning user_id from
+                            // the response metadata so notifications reach the
+                            // correct user rather than the agent's owner.
+                            // This intentionally overrides the configured notify_target
+                            // because each user's heartbeat should notify that user.
+                            let effective_user = if is_multi_tenant {
+                                response
+                                    .metadata
+                                    .get("owner_id")
+                                    .and_then(|v| v.as_str())
+                                    .map(String::from)
+                            } else {
+                                None
+                            };
+
                             // Try the configured channel first, fall back to
                             // broadcasting on all channels.
-                            let targeted_ok = if let Some(ref channel) = notify_channel
+                            let targeted_ok = if let Some(ref channel) = notify_channel {
-                                && let Some(ref user) = notify_target
+                                let target = effective_user.as_deref().or(notify_target.as_deref());
-                            {
+                                if let Some(user) = target {
-                                channels
+                                    channels
-                                    .broadcast(channel, user, response.clone())
+                                        .broadcast(channel, user, response.clone())
-                                    .await
+                                        .await
-                                    .is_ok()
+                                        .is_ok()
+                                } else {
+                                    false
+                                }
                             } else {
                                 false
                             };
 
-                            if !targeted_ok && let Some(ref user) = notify_user {
+                            if !targeted_ok {
-                                let results = channels.broadcast_all(user, response).await;
+                                let fallback = effective_user.as_deref().or(notify_user.as_deref());
-                                for (ch, result) in results {
+                                if let Some(user) = fallback {
-                                    if let Err(e) = result {
+                                    let results = channels.broadcast_all(user, response).await;
-                                        tracing::warn!(
+                                    for (ch, result) in results {
-                                            "Failed to broadcast heartbeat to {}: {}",
+                                        if let Err(e) = result {
-                                            ch,
+                                            tracing::warn!(
-                                            e
+                                                "Failed to broadcast heartbeat to {}: {}",
-                                        );
+                                                ch,
+                                                e
+                                            );
+                                        }
                                     }
                                 }
                             }
@@ -570,14 +655,29 @@ impl Agent {
                         .map(|h| h.to_workspace_config())
                         .unwrap_or_default();
 
-                    Some(spawn_heartbeat(
+                    if config.multi_tenant {
-                        config,
+                        if let Some(admin) = self.admin_store() {
-                        hygiene,
+                            Some(spawn_multi_user_heartbeat(
-                        workspace.clone(),
+                                config,
-                        self.cheap_llm().clone(),
+                                hygiene,
-                        Some(notify_tx),
+                                self.cheap_llm().clone(),
-                        self.store().map(Arc::clone),
+                                Some(notify_tx),
-                    ))
+                                admin,
+                            ))
+                        } else {
+                            tracing::warn!("Multi-tenant heartbeat requires a database store");
+                            None
+                        }
+                    } else {
+                        Some(spawn_heartbeat(
+                            config,
+                            hygiene,
+                            workspace.clone(),
+                            self.cheap_llm().clone(),
+                            Some(notify_tx),
+                            self.admin_store(),
+                        ))
+                    }
                 } else {
                     tracing::warn!("Heartbeat enabled but no workspace available");
                     None
@@ -599,7 +699,7 @@ impl Agent {
 
                     let engine = Arc::new(RoutineEngine::new(
                         rt_config.clone(),
-                        Arc::clone(store),
+                        crate::tenant::AdminScope::new(Arc::clone(store)),
                         self.llm().clone(),
                         Arc::clone(workspace),
                         notify_tx,
@@ -1052,10 +1152,11 @@ impl Agent {
             } else {
                 drop(sess);
                 self.session_manager
-                    .resolve_thread(
+                    .resolve_thread_with_parsed_uuid(
                         &message.user_id,
                         &message.channel,
                         message.conversation_scope(),
+                        approval_thread_uuid,
                     )
                     .await
             }
@@ -1136,9 +1237,14 @@ impl Agent {
             && let Submission::UserInput { ref content } = submission
             && let Some(engine) = self.routine_engine().await
         {
-            let fired = engine
+            let single_message_repl = is_single_message_repl(message);
-                .check_event_triggers(&message.user_id, &message.channel, content)
+            // Use post-hook content so that BeforeInbound hooks that rewrite
-                .await;
+            // input are respected by event trigger matching.
+            let fired = if single_message_repl {
+                engine.check_event_triggers_and_wait(message, content).await
+            } else {
+                engine.check_event_triggers(message, content).await
+            };
             if fired > 0 {
                 tracing::debug!(
                     channel = %message.channel,
@@ -1146,15 +1252,30 @@ impl Agent {
                     fired,
                     "Consumed inbound user message with matching event-triggered routine(s)"
                 );
-                return Ok(Some(String::new()));
+                return if single_message_repl {
+                    Ok(None)
+                } else {
+                    Ok(Some(String::new()))
+                };
             }
         }
 
+        // Build per-tenant execution context once; threaded through all handlers.
+        let tenant = self.tenant_ctx(&message.user_id).await;
+
+        let session_for_empty_exit = Arc::clone(&session);
+
         // Process based on submission type
         let result = match submission {
             Submission::UserInput { content } => {
                 let mut result = self
-                    .process_user_input(message, session.clone(), thread_id, &content)
+                    .process_user_input(
+                        message,
+                        tenant.clone(),
+                        session.clone(),
+                        thread_id,
+                        &content,
+                    )
                     .await;
 
                 // Drain any messages queued during processing.
@@ -1221,7 +1342,13 @@ impl Agent {
                     let mut queued_msg = message.clone();
                     queued_msg.attachments.clear();
                     result = self
-                        .process_user_input(&queued_msg, session.clone(), thread_id, &next_content)
+                        .process_user_input(
+                            &queued_msg,
+                            tenant.clone(),
+                            session.clone(),
+                            thread_id,
+                            &next_content,
+                        )
                         .await;
 
                     // If processing failed, re-queue the drained content so it
@@ -1246,8 +1373,30 @@ impl Agent {
                     command,
                     message.channel
                 );
+                // /reasoning is special-cased here (not in handle_system_command)
+                // because it needs the session + thread_id to read turn reasoning
+                // data, which handle_system_command's signature doesn't provide.
+                if command == "reasoning" {
+                    let result = self
+                        .handle_reasoning_command(&args, &session, thread_id)
+                        .await;
+                    return match result {
+                        SubmissionResult::Response { content } => Ok(Some(content)),
+                        SubmissionResult::Ok { message } => Ok(message),
+                        SubmissionResult::Error { message } => {
+                            Ok(Some(format!("Error: {}", message)))
+                        }
+                        _ => {
+                            if is_single_message_repl(message) {
+                                Ok(None)
+                            } else {
+                                Ok(Some(String::new()))
+                            }
+                        }
+                    };
+                }
                 // Authorization checks (including restart channel check) are enforced in handle_system_command
-                self.handle_system_command(&command, &args, &message.channel)
+                self.handle_system_command(&command, &args, &message.channel, &tenant)
                     .await
             }
             Submission::Undo => self.process_undo(session, thread_id).await,
@@ -1260,12 +1409,9 @@ impl Agent {
             Submission::Summarize => self.process_summarize(session, thread_id).await,
             Submission::Suggest => self.process_suggest(session, thread_id).await,
             Submission::JobStatus { job_id } => {
-                self.process_job_status(&message.user_id, job_id.as_deref())
+                self.process_job_status(&tenant, job_id.as_deref()).await
-                    .await
-            }
-            Submission::JobCancel { job_id } => {
-                self.process_job_cancel(&message.user_id, &job_id).await
             }
+            Submission::JobCancel { job_id } => self.process_job_cancel(&tenant, &job_id).await,
             Submission::Quit => return Ok(None),
             Submission::SwitchThread { thread_id: target } => {
                 self.process_switch_thread(message, target).await
@@ -1305,7 +1451,26 @@ impl Agent {
                     Ok(Some(content))
                 }
             }
-            SubmissionResult::Ok { message } => Ok(message),
+            SubmissionResult::Ok {
+                message: output_message,
+            } => {
+                let should_exit =
+                    if output_message.as_deref() == Some("") && is_single_message_repl(message) {
+                        let sess = session_for_empty_exit.lock().await;
+                        sess.threads
+                            .get(&thread_id)
+                            .map(|thread| thread.state != ThreadState::AwaitingApproval)
+                            .unwrap_or(true)
+                    } else {
+                        false
+                    };
+
+                if should_exit {
+                    Ok(None)
+                } else {
+                    Ok(output_message)
+                }
+            }
             SubmissionResult::Error { message } => Ok(Some(format!("Error: {}", message))),
             SubmissionResult::Interrupted => Ok(Some("Interrupted.".into())),
             SubmissionResult::NeedApproval { .. } => {
@@ -1321,7 +1486,7 @@ impl Agent {
 #[cfg(test)]
 mod tests {
     use super::{
-        chat_tool_execution_metadata, resolve_routine_notification_user,
+        chat_tool_execution_metadata, is_single_message_repl, resolve_routine_notification_user,
         should_fallback_routine_notification, truncate_for_preview,
     };
     use crate::channels::IncomingMessage;
@@ -1483,4 +1648,17 @@ mod tests {
 
         assert!(should_fallback_routine_notification(&error)); // safety: test-only assertion
     }
+
+    #[test]
+    fn single_message_repl_detection_requires_repl_channel_and_metadata_flag() {
+        let repl = IncomingMessage::new("repl", "owner-scope", "hello")
+            .with_metadata(serde_json::json!({ "single_message_mode": true }));
+        let gateway = IncomingMessage::new("gateway", "owner-scope", "hello")
+            .with_metadata(serde_json::json!({ "single_message_mode": true }));
+        let plain_repl = IncomingMessage::new("repl", "owner-scope", "hello");
+
+        assert!(is_single_message_repl(&repl)); // safety: test-only assertion
+        assert!(!is_single_message_repl(&gateway)); // safety: test-only assertion
+        assert!(!is_single_message_repl(&plain_repl)); // safety: test-only assertion
+    }
 }

src/agent/agentic_loop.rs

@@ -10,7 +10,7 @@ use std::borrow::Cow;
 
 use crate::agent::session::PendingApproval;
 use crate::error::Error;
-use crate::llm::{ChatMessage, Reasoning, ReasoningContext, RespondResult};
+use crate::llm::{ChatMessage, FinishReason, Reasoning, ReasoningContext, RespondResult};
 
 /// Signal from the delegate indicating how the loop should proceed.
 pub enum LoopSignal {
@@ -134,6 +134,9 @@ pub async fn run_agentic_loop(
     config: &AgenticLoopConfig,
 ) -> Result<LoopOutcome, Error> {
     let mut consecutive_tool_intent_nudges: u32 = 0;
+    // Accumulates across all iterations (not reset by text responses) so
+    // non-consecutive truncations still escalate to force_text.
+    let mut truncation_count: u32 = 0;
 
     for iteration in 1..=config.max_iterations {
         // Check for external signals (stop, cancellation, user messages)
@@ -215,7 +218,35 @@ pub async fn run_agentic_loop(
                 tool_calls,
                 content,
             } => {
+                // If the response was truncated, tool call parameters are likely
+                // incomplete. Discard them and tell the LLM to try a different
+                // approach rather than executing malformed tool calls.
+                if output.finish_reason == FinishReason::Length {
+                    truncation_count += 1;
+                    let names: Vec<&str> = tool_calls.iter().map(|tc| tc.name.as_str()).collect();
+                    tracing::warn!(
+                        iteration,
+                        tools = ?names,
+                        truncation_count,
+                        "Discarding truncated tool calls (finish_reason=Length)"
+                    );
+                    if let Some(ref text) = content {
+                        reason_ctx.messages.push(ChatMessage::assistant(text));
+                    }
+                    reason_ctx
+                        .messages
+                        .push(ChatMessage::user(crate::llm::TRUNCATED_TOOL_CALL_NOTICE));
+                    // After repeated truncations, force text-only mode so the LLM
+                    // stops attempting tool calls it can't fit in the output budget.
+                    if truncation_count >= 3 {
+                        reason_ctx.force_text = true;
+                    }
+                    delegate.after_iteration(iteration).await;
+                    continue;
+                }
+
                 consecutive_tool_intent_nudges = 0;
+                truncation_count = 0;
 
                 if let Some(outcome) = delegate
                     .execute_tool_calls(tool_calls, content, reason_ctx)
@@ -271,6 +302,7 @@ mod tests {
         RespondOutput {
             result: RespondResult::Text(text.to_string()),
             usage: zero_usage(),
+            finish_reason: FinishReason::Stop,
         }
     }
 
@@ -281,6 +313,7 @@ mod tests {
                 content: None,
             },
             usage: zero_usage(),
+            finish_reason: FinishReason::ToolUse,
         }
     }
 
@@ -414,6 +447,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "echo".to_string(),
             arguments: serde_json::json!({}),
+            reasoning: None,
         };
         let delegate = MockDelegate::new(vec![
             tool_calls_output(vec![tool_call]),
@@ -621,4 +655,95 @@ mod tests {
         let result = truncate_for_preview("café", 4);
         assert_eq!(result, "caf...");
     }
+
+    #[tokio::test]
+    async fn test_truncated_tool_calls_discarded_on_length() {
+        let truncated_tool_call = ToolCall {
+            id: "call_1".to_string(),
+            name: "memory_write".to_string(),
+            arguments: serde_json::json!({}), // empty — truncated
+            reasoning: None,
+        };
+        let truncated_output = RespondOutput {
+            result: RespondResult::ToolCalls {
+                tool_calls: vec![truncated_tool_call],
+                content: Some("I'll write the report.".to_string()),
+            },
+            usage: zero_usage(),
+            finish_reason: FinishReason::Length, // response was truncated
+        };
+        let delegate = MockDelegate::new(vec![truncated_output, text_output("Summarized it.")]);
+        let reasoning = stub_reasoning();
+        let mut ctx = ReasoningContext::new();
+        let config = AgenticLoopConfig {
+            max_iterations: 5,
+            ..Default::default()
+        };
+
+        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
+            .await
+            .unwrap();
+
+        // Tool calls should NOT have been executed
+        assert_eq!(delegate.tool_exec_count.load(Ordering::SeqCst), 0);
+        // The loop should have continued and returned the text response
+        assert!(matches!(outcome, LoopOutcome::Response(ref t) if t == "Summarized it."));
+        // A truncation notice should have been injected into context
+        assert!(
+            ctx.messages
+                .iter()
+                .any(|m| m.role == crate::llm::Role::User && m.content.contains("truncated")),
+            "Should inject truncation notice into context"
+        );
+        // The partial assistant content should have been preserved
+        assert!(
+            ctx.messages
+                .iter()
+                .any(|m| m.role == crate::llm::Role::Assistant
+                    && m.content.contains("write the report")),
+            "Should preserve partial assistant content"
+        );
+    }
+
+    #[tokio::test]
+    async fn test_repeated_truncations_force_text_mode() {
+        let make_truncated = || RespondOutput {
+            result: RespondResult::ToolCalls {
+                tool_calls: vec![ToolCall {
+                    id: "call_1".to_string(),
+                    name: "memory_write".to_string(),
+                    arguments: serde_json::json!({}),
+                    reasoning: None,
+                }],
+                content: None,
+            },
+            usage: zero_usage(),
+            finish_reason: FinishReason::Length,
+        };
+        // Three truncated responses, then a text response
+        let delegate = MockDelegate::new(vec![
+            make_truncated(),
+            make_truncated(),
+            make_truncated(),
+            text_output("Gave up on tool calls."),
+        ]);
+        let reasoning = stub_reasoning();
+        let mut ctx = ReasoningContext::new();
+        let config = AgenticLoopConfig {
+            max_iterations: 5,
+            ..Default::default()
+        };
+
+        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
+            .await
+            .unwrap();
+
+        assert!(matches!(outcome, LoopOutcome::Response(_)));
+        assert_eq!(delegate.tool_exec_count.load(Ordering::SeqCst), 0);
+        // After 3 truncations, force_text should be set
+        assert!(
+            ctx.force_text,
+            "Should escalate to force_text after repeated truncations"
+        );
+    }
 }

src/agent/commands.rs

@@ -33,6 +33,7 @@ impl Agent {
         &self,
         intent: MessageIntent,
         message: &IncomingMessage,
+        tenant: &crate::tenant::TenantCtx,
     ) -> Result<SubmissionResult, Error> {
         // Send thinking status for non-trivial operations
         if let MessageIntent::CreateJob { .. } = &intent {
@@ -52,24 +53,18 @@ impl Agent {
                 description,
                 category,
             } => {
-                self.handle_create_job(&message.user_id, title, description, category)
+                self.handle_create_job(tenant, title, description, category)
                     .await?
             }
             MessageIntent::CheckJobStatus { job_id } => {
-                self.handle_check_status(&message.user_id, job_id).await?
+                self.handle_check_status(tenant, job_id).await?
-            }
-            MessageIntent::CancelJob { job_id } => {
-                self.handle_cancel_job(&message.user_id, &job_id).await?
-            }
-            MessageIntent::ListJobs { filter } => {
-                self.handle_list_jobs(&message.user_id, filter).await?
-            }
-            MessageIntent::HelpJob { job_id } => {
-                self.handle_help_job(&message.user_id, &job_id).await?
             }
+            MessageIntent::CancelJob { job_id } => self.handle_cancel_job(tenant, &job_id).await?,
+            MessageIntent::ListJobs { filter } => self.handle_list_jobs(tenant, filter).await?,
+            MessageIntent::HelpJob { job_id } => self.handle_help_job(tenant, &job_id).await?,
             MessageIntent::Command { command, args } => {
                 match self
-                    .handle_command(&command, &args, &message.channel)
+                    .handle_command(&command, &args, &message.channel, tenant)
                     .await?
                 {
                     Some(s) => s,
@@ -83,14 +78,14 @@ impl Agent {
 
     async fn handle_create_job(
         &self,
-        user_id: &str,
+        tenant: &crate::tenant::TenantCtx,
         title: String,
         description: String,
         category: Option<String>,
     ) -> Result<String, Error> {
         let job_id = self
             .scheduler
-            .dispatch_job(user_id, &title, &description, None)
+            .dispatch_job(tenant.user_id(), &title, &description, None)
             .await?;
 
         // Set the dedicated category field (not stored in metadata)
@@ -113,7 +108,7 @@ impl Agent {
 
     async fn handle_check_status(
         &self,
-        user_id: &str,
+        tenant: &crate::tenant::TenantCtx,
         job_id: Option<String>,
     ) -> Result<String, Error> {
         match job_id {
@@ -122,7 +117,8 @@ impl Agent {
                     .map_err(|_| crate::error::JobError::NotFound { id: Uuid::nil() })?;
 
                 // Try DB first for persistent state, fall back to ContextManager.
-                if let Some(store) = self.store()
+                // TenantScope.get_job() auto-filters by ownership — no manual check needed.
+                if let Some(store) = tenant.store()
                     && let Ok(Some(ctx)) = store.get_job(uuid).await
                 {
                     return Ok(format!(
@@ -138,7 +134,7 @@ impl Agent {
                 }
 
                 let ctx = self.context_manager.get_context(uuid).await?;
-                if ctx.user_id != user_id {
+                if ctx.user_id != tenant.user_id() {
                     return Err(crate::error::JobError::NotFound { id: uuid }.into());
                 }
 
@@ -155,7 +151,8 @@ impl Agent {
             }
             None => {
                 // Show summary from DB for consistency with Jobs tab.
-                if let Some(store) = self.store() {
+                // TenantScope methods auto-scope to user — no user_id parameter needed.
+                if let Some(store) = tenant.store() {
                     let mut total = 0;
                     let mut in_progress = 0;
                     let mut completed = 0;
@@ -183,7 +180,7 @@ impl Agent {
                 }
 
                 // Fallback to ContextManager if no DB.
-                let summary = self.context_manager.summary_for(user_id).await;
+                let summary = self.context_manager.summary_for(tenant.user_id()).await;
                 Ok(format!(
                     "Jobs summary: Total: {} In Progress: {} Completed: {} Failed: {} Stuck: {}",
                     summary.total,
@@ -196,19 +193,24 @@ impl Agent {
         }
     }
 
-    async fn handle_cancel_job(&self, user_id: &str, job_id: &str) -> Result<String, Error> {
+    async fn handle_cancel_job(
+        &self,
+        tenant: &crate::tenant::TenantCtx,
+        job_id: &str,
+    ) -> Result<String, Error> {
         let uuid = Uuid::parse_str(job_id)
             .map_err(|_| crate::error::JobError::NotFound { id: Uuid::nil() })?;
 
         let ctx = self.context_manager.get_context(uuid).await?;
-        if ctx.user_id != user_id {
+        if ctx.user_id != tenant.user_id() {
             return Err(crate::error::JobError::NotFound { id: uuid }.into());
         }
 
         self.scheduler.stop(uuid).await?;
 
         // Also update DB so the Jobs tab reflects cancellation immediately.
-        if let Some(store) = self.store()
+        // Use TenantScope — ownership already verified above.
+        if let Some(store) = tenant.store()
             && let Err(e) = store
                 .update_job_status(uuid, JobState::Cancelled, Some("Cancelled by user"))
                 .await
@@ -221,11 +223,12 @@ impl Agent {
 
     async fn handle_list_jobs(
         &self,
-        user_id: &str,
+        tenant: &crate::tenant::TenantCtx,
         _filter: Option<String>,
     ) -> Result<String, Error> {
         // List from DB for consistency with Jobs tab.
-        if let Some(store) = self.store() {
+        // TenantScope methods auto-scope to user.
+        if let Some(store) = tenant.store() {
             let agent_jobs = match store.list_agent_jobs().await {
                 Ok(jobs) => jobs,
                 Err(e) => {
@@ -256,7 +259,7 @@ impl Agent {
         }
 
         // Fallback to ContextManager if no DB.
-        let jobs = self.context_manager.all_jobs_for(user_id).await;
+        let jobs = self.context_manager.all_jobs_for(tenant.user_id()).await;
         if jobs.is_empty() {
             return Ok("No jobs found.".to_string());
         }
@@ -270,12 +273,16 @@ impl Agent {
         Ok(output)
     }
 
-    async fn handle_help_job(&self, user_id: &str, job_id: &str) -> Result<String, Error> {
+    async fn handle_help_job(
+        &self,
+        tenant: &crate::tenant::TenantCtx,
+        job_id: &str,
+    ) -> Result<String, Error> {
         let uuid = Uuid::parse_str(job_id)
             .map_err(|_| crate::error::JobError::NotFound { id: Uuid::nil() })?;
 
         let ctx = self.context_manager.get_context(uuid).await?;
-        if ctx.user_id != user_id {
+        if ctx.user_id != tenant.user_id() {
             return Err(crate::error::JobError::NotFound { id: uuid }.into());
         }
 
@@ -308,11 +315,11 @@ impl Agent {
     /// Show job status inline — either all jobs (no id) or a specific job.
     pub(super) async fn process_job_status(
         &self,
-        user_id: &str,
+        tenant: &crate::tenant::TenantCtx,
         job_id: Option<&str>,
     ) -> Result<SubmissionResult, Error> {
         match self
-            .handle_check_status(user_id, job_id.map(|s| s.to_string()))
+            .handle_check_status(tenant, job_id.map(|s| s.to_string()))
             .await
         {
             Ok(text) => Ok(SubmissionResult::response(text)),
@@ -323,10 +330,10 @@ impl Agent {
     /// Cancel a job by ID.
     pub(super) async fn process_job_cancel(
         &self,
-        user_id: &str,
+        tenant: &crate::tenant::TenantCtx,
         job_id: &str,
     ) -> Result<SubmissionResult, Error> {
-        match self.handle_cancel_job(user_id, job_id).await {
+        match self.handle_cancel_job(tenant, job_id).await {
             Ok(text) => Ok(SubmissionResult::response(text)),
             Err(e) => Ok(SubmissionResult::error(format!("Cancel error: {}", e))),
         }
@@ -465,12 +472,101 @@ impl Agent {
         }
     }
 
+    /// Handle `/reasoning [N|all]` — show reasoning history for the active thread.
+    pub(super) async fn handle_reasoning_command(
+        &self,
+        args: &[String],
+        session: &Arc<Mutex<Session>>,
+        thread_id: Uuid,
+    ) -> SubmissionResult {
+        // Clone the turn data we need, then drop the session lock.
+        let turns_snapshot: Vec<(
+            usize,
+            Option<String>,
+            Vec<crate::agent::session::TurnToolCall>,
+        )>;
+        {
+            let sess = session.lock().await;
+            let thread = match sess.threads.get(&thread_id) {
+                Some(t) => t,
+                None => return SubmissionResult::error("No active thread."),
+            };
+
+            if thread.turns.is_empty() {
+                return SubmissionResult::ok_with_message("No turns yet.");
+            }
+
+            // Parse argument: default=last turn, "all"=all turns, N=specific turn (1-based).
+            let selected: Vec<&crate::agent::session::Turn> = match args.first().map(|s| s.as_str())
+            {
+                Some("all") => thread.turns.iter().collect(),
+                Some(n) => match n.parse::<usize>() {
+                    Ok(0) => return SubmissionResult::error("Turn numbers start at 1."),
+                    Ok(num) if num > thread.turns.len() => {
+                        return SubmissionResult::error(format!(
+                            "Turn {} does not exist (max: {}).",
+                            num,
+                            thread.turns.len()
+                        ));
+                    }
+                    Ok(num) => vec![&thread.turns[num - 1]],
+                    Err(_) => return SubmissionResult::error("Usage: /reasoning [N|all]"),
+                },
+                None => {
+                    // Default: last turn that has tool calls
+                    match thread.turns.iter().rev().find(|t| !t.tool_calls.is_empty()) {
+                        Some(t) => vec![t],
+                        None => {
+                            return SubmissionResult::ok_with_message("No turns with tool calls.");
+                        }
+                    }
+                }
+            };
+
+            turns_snapshot = selected
+                .into_iter()
+                .map(|t| (t.turn_number, t.narrative.clone(), t.tool_calls.clone()))
+                .collect();
+        }
+        // Session lock is now dropped — format output without holding it.
+
+        let mut output = String::new();
+        for (turn_number, narrative, tool_calls) in &turns_snapshot {
+            output.push_str(&format!("--- Turn {} ---\n", turn_number + 1));
+            if let Some(narrative) = narrative {
+                output.push_str(&format!("Reasoning: {}\n", narrative));
+            }
+            if tool_calls.is_empty() {
+                output.push_str("  (no tool calls)\n");
+            } else {
+                for tc in tool_calls {
+                    let status = if tc.error.is_some() {
+                        "error"
+                    } else if tc.result.is_some() {
+                        "ok"
+                    } else {
+                        "pending"
+                    };
+                    output.push_str(&format!("  {} [{}]", tc.name, status));
+                    if let Some(ref rationale) = tc.rationale {
+                        output.push_str(&format!(" — {}", rationale));
+                    }
+                    output.push('\n');
+                }
+            }
+            output.push('\n');
+        }
+
+        SubmissionResult::response(output.trim_end())
+    }
+
     /// Handle system commands that bypass thread-state checks entirely.
     pub(super) async fn handle_system_command(
         &self,
         command: &str,
         args: &[String],
         channel: &str,
+        tenant: &crate::tenant::TenantCtx,
     ) -> Result<SubmissionResult, Error> {
         match command {
             "help" => Ok(SubmissionResult::response(concat!(
@@ -480,6 +576,7 @@ impl Agent {
                 "  /version          Show version info\n",
                 "  /tools            List available tools\n",
                 "  /debug            Toggle debug mode\n",
+                "  /reasoning [N|all] Show agent reasoning for turns\n",
                 "  /ping             Connectivity check\n",
                 "\n",
                 "Jobs:\n",
@@ -663,19 +760,32 @@ impl Agent {
                         }
                     }
 
-                    match self.llm().set_model(requested) {
+                    if self.config.multi_tenant {
-                        Ok(()) => {
+                        // Multi-tenant: only persist to per-user DB settings.
-                            // Persist the model choice so it survives restarts.
+                        // Do NOT call set_model() on the shared provider — that
-                            self.persist_selected_model(requested).await;
+                        // would change the default for all users. The per-request
-                            Ok(SubmissionResult::response(format!(
+                        // model_override in the dispatcher reads from the same
-                                "Switched model to: {}",
+                        // "selected_model" setting and applies it per-user.
-                                requested
+                        self.persist_selected_model(tenant, requested).await;
-                            )))
+                        Ok(SubmissionResult::response(format!(
+                            "Model preference set to: {} (per-user)",
+                            requested
+                        )))
+                    } else {
+                        match self.llm().set_model(requested) {
+                            Ok(()) => {
+                                // Persist the model choice so it survives restarts.
+                                self.persist_selected_model(tenant, requested).await;
+                                Ok(SubmissionResult::response(format!(
+                                    "Switched model to: {}",
+                                    requested
+                                )))
+                            }
+                            Err(e) => Ok(SubmissionResult::error(format!(
+                                "Failed to switch model: {}",
+                                e
+                            ))),
                         }
-                        Err(e) => Ok(SubmissionResult::error(format!(
-                            "Failed to switch model: {}",
-                            e
-                        ))),
                     }
                 }
             }
@@ -817,10 +927,14 @@ impl Agent {
         command: &str,
         args: &[String],
         channel: &str,
+        tenant: &crate::tenant::TenantCtx,
     ) -> 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, channel).await? {
+        match self
+            .handle_system_command(command, args, channel, tenant)
+            .await?
+        {
             SubmissionResult::Response { content } => Ok(Some(content)),
             SubmissionResult::Ok { message } => Ok(message),
             SubmissionResult::Error { message } => Ok(Some(format!("Error: {}", message))),
@@ -832,21 +946,69 @@ impl Agent {
     ///
     /// Best-effort: logs warnings on failure but does not propagate errors,
     /// since the in-memory model switch already succeeded.
-    async fn persist_selected_model(&self, model: &str) {
+    ///
-        // 1. Persist to DB if available.
+    /// In multi-tenant mode, only the per-user DB setting is written — global
-        if let Some(store) = self.store() {
+    /// .env and TOML files are shared across users and must not be mutated.
+    async fn persist_selected_model(&self, tenant: &crate::tenant::TenantCtx, model: &str) {
+        // 1. Persist to DB if available (per-user scoped via TenantScope).
+        if let Some(store) = tenant.store() {
             let value = serde_json::Value::String(model.to_string());
-            if let Err(e) = store
+            if let Err(e) = store.set_setting("selected_model", &value).await {
-                .set_setting(self.owner_id(), "selected_model", &value)
-                .await
-            {
                 tracing::warn!("Failed to persist model to DB: {}", e);
+            } else {
+                tracing::debug!(
+                    user_id = tenant.user_id(),
+                    "Persisted selected_model to DB: {}",
+                    model
+                );
             }
+        } else {
+            tracing::warn!("No database store available — model choice will not persist to DB");
         }
 
-        // 2. Update TOML config file if it exists (sync I/O in spawn_blocking).
+        // 2. In multi-tenant mode, skip .env/TOML writes — these are global
+        // files shared by all users. The per-user DB setting is sufficient.
+        if self.config.multi_tenant {
+            return;
+        }
+
+        // 3. Update .env and TOML config file (sync I/O in spawn_blocking).
         let model_owned = model.to_string();
+        let backend = self.deps.llm_backend.clone();
         if let Err(e) = tokio::task::spawn_blocking(move || {
+            // 2a. Update the backend-specific model env var in ~/.ironclaw/.env.
+            //
+            // Env vars have the HIGHEST priority in LlmConfig::resolve_model()
+            // (env var > TOML > DB > default). If the .env file has e.g.
+            // NEARAI_MODEL=old-model, it shadows everything else. We must
+            // update this var or the /model change is invisible on restart.
+            let registry = crate::llm::ProviderRegistry::load();
+            let model_env = registry.model_env_var(&backend);
+            let env_var_prefix = format!("{}=", model_env);
+
+            // Only update the .env file if the var is actually set there
+            // (avoid injecting new vars the user never configured).
+            let env_path = crate::bootstrap::ironclaw_env_path();
+            let env_has_var = std::fs::read_to_string(&env_path)
+                .ok()
+                .is_some_and(|content| {
+                    content.lines().any(|line| {
+                        let trimmed = line.trim_start();
+                        !trimmed.starts_with('#') && trimmed.starts_with(&env_var_prefix)
+                    })
+                });
+            if env_has_var {
+                if let Err(e) = crate::bootstrap::upsert_bootstrap_var(model_env, &model_owned) {
+                    tracing::warn!("Failed to update {} in .env: {}", model_env, e);
+                } else {
+                    tracing::debug!("Updated {} in .env to {}", model_env, model_owned);
+                }
+            }
+
+            // 2b. Update (or create) the TOML config file.
+            //
+            // The TOML overlay has higher priority than DB settings on
+            // startup, so it MUST stay in sync with the DB.
             let toml_path = crate::settings::Settings::default_toml_path();
             match crate::settings::Settings::load_toml(&toml_path) {
                 Ok(Some(mut settings)) => {
@@ -856,7 +1018,15 @@ impl Agent {
                     }
                 }
                 Ok(None) => {
-                    // No config file on disk; nothing to update.
+                    // No config file yet — create one so the model choice
+                    // survives restarts even when the DB is unavailable.
+                    let settings = crate::settings::Settings {
+                        selected_model: Some(model_owned),
+                        ..Default::default()
+                    };
+                    if let Err(e) = settings.save_toml(&toml_path) {
+                        tracing::warn!("Failed to create config.toml for model persistence: {}", e);
+                    }
                 }
                 Err(e) => {
                     tracing::warn!("Failed to load config.toml for model persistence: {}", e);
@@ -865,7 +1035,7 @@ impl Agent {
         })
         .await
         {
-            tracing::warn!("Model TOML persistence task failed: {}", e);
+            tracing::warn!("Model persistence task failed: {}", e);
         }
     }
 }

src/agent/cost_guard.rs

@@ -21,6 +21,9 @@ pub struct CostGuardConfig {
     pub max_cost_per_day_cents: Option<u64>,
     /// Maximum LLM calls per hour. None = unlimited.
     pub max_actions_per_hour: Option<u64>,
+    /// Maximum spend per user per day in cents. None = unlimited.
+    /// Applied independently per user alongside the global budget.
+    pub max_cost_per_user_per_day_cents: Option<u64>,
 }
 
 /// Error returned when a cost limit is exceeded.
@@ -30,6 +33,12 @@ pub enum CostLimitExceeded {
     DailyBudget { spent_cents: u64, limit_cents: u64 },
     /// Hourly action rate limit reached.
     HourlyRate { actions: u64, limit: u64 },
+    /// Per-user daily spending cap reached.
+    UserDailyBudget {
+        user_id: String,
+        spent_cents: u64,
+        limit_cents: u64,
+    },
 }
 
 impl std::fmt::Display for CostLimitExceeded {
@@ -49,6 +58,17 @@ impl std::fmt::Display for CostLimitExceeded {
                 "Hourly action limit exceeded: {} actions of {} allowed per hour",
                 actions, limit
             ),
+            Self::UserDailyBudget {
+                user_id,
+                spent_cents,
+                limit_cents,
+            } => write!(
+                f,
+                "User '{}' daily cost limit exceeded: spent ${:.2} of ${:.2} allowed",
+                user_id,
+                *spent_cents as f64 / 100.0,
+                *limit_cents as f64 / 100.0
+            ),
         }
     }
 }
@@ -78,6 +98,9 @@ pub struct CostGuard {
 
     /// Per-model token usage since startup.
     model_tokens: Mutex<HashMap<String, ModelTokens>>,
+
+    /// Per-user daily cost tracking. Each entry resets independently at midnight UTC.
+    per_user_daily_cost: Mutex<HashMap<String, DailyCost>>,
 }
 
 struct DailyCost {
@@ -97,6 +120,7 @@ impl CostGuard {
             action_window: Mutex::new(VecDeque::new()),
             budget_exceeded: AtomicBool::new(false),
             model_tokens: Mutex::new(HashMap::new()),
+            per_user_daily_cost: Mutex::new(HashMap::new()),
         }
     }
 
@@ -203,6 +227,11 @@ impl CostGuard {
                 daily.reset_date = today;
                 self.budget_exceeded.store(false, Ordering::Relaxed);
                 tracing::info!("Cost guard: daily counter reset for {}", today);
+
+                // Prune per-user entries from previous days to prevent
+                // unbounded HashMap growth in long-lived deployments.
+                let mut per_user = self.per_user_daily_cost.lock().await;
+                per_user.retain(|_, entry| entry.reset_date == today);
             }
             daily.total += cost;
 
@@ -248,6 +277,85 @@ impl CostGuard {
         cost
     }
 
+    /// Record an LLM call with per-user attribution.
+    ///
+    /// Delegates to `record_llm_call` for global tracking, then additionally
+    /// records the cost against the user's daily budget.
+    #[allow(clippy::too_many_arguments)]
+    pub async fn record_llm_call_for_user(
+        &self,
+        user_id: &str,
+        model: &str,
+        input_tokens: u32,
+        output_tokens: u32,
+        cache_read_input_tokens: u32,
+        cache_creation_input_tokens: u32,
+        cache_read_discount: Decimal,
+        cache_write_multiplier: Decimal,
+        cost_per_token: Option<(Decimal, Decimal)>,
+    ) -> Decimal {
+        let cost = self
+            .record_llm_call(
+                model,
+                input_tokens,
+                output_tokens,
+                cache_read_input_tokens,
+                cache_creation_input_tokens,
+                cache_read_discount,
+                cache_write_multiplier,
+                cost_per_token,
+            )
+            .await;
+
+        // Track per-user daily cost
+        {
+            let today = chrono::Utc::now().date_naive();
+            let mut per_user = self.per_user_daily_cost.lock().await;
+            let entry = per_user
+                .entry(user_id.to_string())
+                .or_insert_with(|| DailyCost {
+                    total: Decimal::ZERO,
+                    reset_date: today,
+                });
+            if today != entry.reset_date {
+                entry.total = Decimal::ZERO;
+                entry.reset_date = today;
+            }
+            entry.total += cost;
+        }
+
+        cost
+    }
+
+    /// Check whether the next action is allowed for a specific user.
+    ///
+    /// Checks the global limits first (via `check_allowed`), then additionally
+    /// checks the per-user daily budget if configured.
+    pub async fn check_allowed_for_user(&self, user_id: &str) -> Result<(), CostLimitExceeded> {
+        // Check global limits first
+        self.check_allowed().await?;
+
+        // Check per-user daily budget
+        if let Some(limit_cents) = self.config.max_cost_per_user_per_day_cents {
+            let today = chrono::Utc::now().date_naive();
+            let per_user = self.per_user_daily_cost.lock().await;
+            if let Some(entry) = per_user.get(user_id)
+                && entry.reset_date == today
+            {
+                let spent_cents = to_cents(entry.total);
+                if spent_cents >= limit_cents {
+                    return Err(CostLimitExceeded::UserDailyBudget {
+                        user_id: user_id.to_string(),
+                        spent_cents,
+                        limit_cents,
+                    });
+                }
+            }
+        }
+
+        Ok(())
+    }
+
     /// Current daily spend in USD (as Decimal).
     pub async fn daily_spend(&self) -> Decimal {
         let daily = self.daily_cost.lock().await;
@@ -259,6 +367,16 @@ impl CostGuard {
         }
     }
 
+    /// Current daily spend for a specific user in USD (as Decimal).
+    pub async fn daily_spend_for_user(&self, user_id: &str) -> Decimal {
+        let today = chrono::Utc::now().date_naive();
+        let per_user = self.per_user_daily_cost.lock().await;
+        match per_user.get(user_id) {
+            Some(entry) if entry.reset_date == today => entry.total,
+            _ => Decimal::ZERO,
+        }
+    }
+
     /// Number of actions in the current hourly window.
     pub async fn actions_this_hour(&self) -> u64 {
         let mut window = self.action_window.lock().await;
@@ -314,7 +432,7 @@ mod tests {
     async fn test_daily_budget_enforcement() {
         let guard = CostGuard::new(CostGuardConfig {
             max_cost_per_day_cents: Some(1), // $0.01 limit
-            max_actions_per_hour: None,
+            ..CostGuardConfig::default()
         });
 
         // First call allowed
@@ -350,8 +468,8 @@ mod tests {
     #[tokio::test]
     async fn test_hourly_rate_enforcement() {
         let guard = CostGuard::new(CostGuardConfig {
-            max_cost_per_day_cents: None,
             max_actions_per_hour: Some(3),
+            ..CostGuardConfig::default()
         });
 
         // First 3 actions allowed
@@ -633,8 +751,8 @@ mod tests {
         // A fresh CostGuard with rate limits should not panic even if
         // checked_sub returns None (simulating short uptime).
         let guard = CostGuard::new(CostGuardConfig {
-            max_cost_per_day_cents: None,
             max_actions_per_hour: Some(100),
+            ..CostGuardConfig::default()
         });
 
         // These must not panic regardless of system uptime
@@ -656,4 +774,119 @@ mod tests {
         let result = Instant::now().checked_sub(std::time::Duration::MAX);
         assert!(result.is_none());
     }
+
+    #[tokio::test]
+    async fn test_per_user_daily_budget_enforcement() {
+        let guard = CostGuard::new(CostGuardConfig {
+            max_cost_per_day_cents: None,
+            max_actions_per_hour: None,
+            max_cost_per_user_per_day_cents: Some(1), // $0.01 per user
+        });
+
+        // Both users initially allowed
+        assert!(guard.check_allowed_for_user("alice").await.is_ok());
+        assert!(guard.check_allowed_for_user("bob").await.is_ok());
+
+        // Alice makes an expensive call
+        guard
+            .record_llm_call_for_user(
+                "alice",
+                "gpt-4o",
+                10_000,
+                10_000,
+                0,
+                0,
+                Decimal::ONE,
+                Decimal::ONE,
+                None,
+            )
+            .await;
+
+        // Alice should be blocked, Bob should still be allowed
+        let result = guard.check_allowed_for_user("alice").await;
+        assert!(result.is_err());
+        match result.unwrap_err() {
+            CostLimitExceeded::UserDailyBudget {
+                user_id,
+                limit_cents,
+                ..
+            } => {
+                assert_eq!(user_id, "alice");
+                assert_eq!(limit_cents, 1);
+            }
+            other => panic!("Expected UserDailyBudget, got {:?}", other),
+        }
+        assert!(guard.check_allowed_for_user("bob").await.is_ok());
+    }
+
+    #[tokio::test]
+    async fn test_per_user_daily_spend_tracking() {
+        let guard = CostGuard::new(CostGuardConfig::default());
+
+        assert_eq!(guard.daily_spend_for_user("alice").await, Decimal::ZERO);
+        assert_eq!(guard.daily_spend_for_user("bob").await, Decimal::ZERO);
+
+        let cost = guard
+            .record_llm_call_for_user(
+                "alice",
+                "gpt-4o",
+                1000,
+                500,
+                0,
+                0,
+                Decimal::ONE,
+                Decimal::ONE,
+                None,
+            )
+            .await;
+
+        assert_eq!(guard.daily_spend_for_user("alice").await, cost);
+        assert_eq!(guard.daily_spend_for_user("bob").await, Decimal::ZERO);
+        // Global spend should also be tracked
+        assert_eq!(guard.daily_spend().await, cost);
+    }
+
+    #[tokio::test]
+    async fn test_per_user_budget_independent_of_global() {
+        let guard = CostGuard::new(CostGuardConfig {
+            max_cost_per_day_cents: Some(100_000), // $1000 global limit
+            max_actions_per_hour: None,
+            max_cost_per_user_per_day_cents: Some(1), // $0.01 per user
+        });
+
+        // User hits their personal limit
+        guard
+            .record_llm_call_for_user(
+                "alice",
+                "gpt-4o",
+                10_000,
+                10_000,
+                0,
+                0,
+                Decimal::ONE,
+                Decimal::ONE,
+                None,
+            )
+            .await;
+
+        // Alice blocked by per-user limit, not global
+        assert!(guard.check_allowed_for_user("alice").await.is_err());
+        // Global limit is far from reached
+        assert!(guard.check_allowed().await.is_ok());
+        // Bob is unaffected
+        assert!(guard.check_allowed_for_user("bob").await.is_ok());
+    }
+
+    #[test]
+    fn test_user_cost_limit_display() {
+        let limit = CostLimitExceeded::UserDailyBudget {
+            user_id: "alice".to_string(),
+            spent_cents: 150,
+            limit_cents: 100,
+        };
+        let msg = limit.to_string();
+        assert!(msg.contains("alice"));
+        assert!(msg.contains("$1.50"));
+        assert!(msg.contains("$1.00"));
+    }
 }

src/agent/dispatcher.rs

@@ -42,6 +42,7 @@ impl Agent {
     pub(super) async fn run_agentic_loop(
         &self,
         message: &IncomingMessage,
+        tenant: crate::tenant::TenantCtx,
         session: Arc<Mutex<Session>>,
         thread_id: Uuid,
         initial_messages: Vec<ChatMessage>,
@@ -63,7 +64,12 @@ impl Agent {
         );
 
         let system_prompt = if let Some(ws) = self.workspace() {
-            match ws
+            let scoped_workspace = if ws.user_id() == message.user_id {
+                Arc::clone(ws)
+            } else {
+                Arc::new(ws.scoped_to_user(&message.user_id))
+            };
+            match scoped_workspace
                 .system_prompt_for_context_tz(is_group_chat, user_tz)
                 .await
             {
@@ -163,6 +169,7 @@ impl Agent {
 
         let delegate = ChatDelegate {
             agent: self,
+            tenant,
             session: session.clone(),
             thread_id,
             message,
@@ -235,6 +242,7 @@ impl Agent {
 /// auth intercept, and cost tracking.
 struct ChatDelegate<'a> {
     agent: &'a Agent,
+    tenant: crate::tenant::TenantCtx,
     session: Arc<Mutex<Session>>,
     thread_id: Uuid,
     message: &'a IncomingMessage,
@@ -298,6 +306,8 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
 
         // Update context for this iteration
         reason_ctx.available_tools = tool_defs;
+        // Preserve force_text if already set (e.g. by truncation escalation).
+        let force_text = force_text || reason_ctx.force_text;
         reason_ctx.system_prompt = Some(if force_text {
             self.cached_prompt_no_tools.clone()
         } else {
@@ -331,8 +341,8 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
         reason_ctx: &mut ReasoningContext,
         iteration: usize,
     ) -> Result<crate::llm::RespondOutput, Error> {
-        // Enforce cost guardrails before the LLM call
+        // Enforce cost guardrails before the LLM call (global + per-user)
-        if let Err(limit) = self.agent.cost_guard().check_allowed().await {
+        if let Err(limit) = self.tenant.check_cost_allowed().await {
             return Err(crate::error::LlmError::InvalidResponse {
                 provider: "agent".to_string(),
                 reason: limit.to_string(),
@@ -340,6 +350,21 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
             .into());
         }
 
+        // Apply per-user model override from settings (first iteration only
+        // to avoid repeated DB lookups within the same agentic loop).
+        // Uses "selected_model" — the same key the /model command persists to
+        // via SettingsStore (per-user scoped via TenantScope).
+        if iteration == 0
+            && let Some(store) = self.tenant.store()
+            && let Ok(Some(value)) = store.get_setting("selected_model").await
+            && let Some(model) = value.as_str()
+        {
+            let model = model.trim();
+            if !model.is_empty() {
+                reason_ctx.model_override = Some(model.to_string());
+            }
+        }
+
         let output = match reasoning.respond_with_tools(reason_ctx).await {
             Ok(output) => output,
             Err(crate::error::LlmError::ContextLengthExceeded { used, limit }) => {
@@ -374,13 +399,22 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
             Err(e) => return Err(e.into()),
         };
 
-        // Record cost and track token usage
+        // Record cost and track token usage (global + per-user).
-        let model_name = self.agent.llm().active_model_name();
+        // When a model override is active, use the override name for attribution
+        // and let CostGuard look up pricing via costs::model_cost() instead of
+        // using the default provider's cost_per_token (which reflects the wrong model).
+        let (model_name, cost_per_token) = if let Some(ref ovr) = reason_ctx.model_override {
+            (ovr.clone(), None)
+        } else {
+            (
+                self.agent.llm().active_model_name(),
+                Some(self.agent.llm().cost_per_token()),
+            )
+        };
         let read_discount = self.agent.llm().cache_read_discount();
         let write_multiplier = self.agent.llm().cache_write_multiplier();
         let call_cost = self
-            .agent
+            .tenant
-            .cost_guard()
             .record_llm_call(
                 &model_name,
                 output.usage.input_tokens,
@@ -389,7 +423,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                 output.usage.cache_creation_input_tokens,
                 read_discount,
                 write_multiplier,
-                Some(self.agent.llm().cost_per_token()),
+                cost_per_token,
             )
             .await;
         tracing::debug!(
@@ -420,6 +454,19 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
         content: Option<String>,
         reason_ctx: &mut ReasoningContext,
     ) -> Result<Option<LoopOutcome>, Error> {
+        // Extract and sanitize the narrative before consuming `content`.
+        let narrative = content
+            .as_deref()
+            .filter(|c| !c.trim().is_empty())
+            .map(|c| {
+                let sanitized = self
+                    .agent
+                    .safety()
+                    .sanitize_tool_output("agent_narrative", c);
+                sanitized.content
+            })
+            .filter(|c| !c.trim().is_empty());
+
         // Add the assistant message with tool_calls to context.
         // OpenAI protocol requires this before tool-result messages.
         reason_ctx
@@ -440,6 +487,41 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
             )
             .await;
 
+        // Build per-tool decisions for the reasoning update.
+        // Sanitize each rationale through SafetyLayer (parity with JobDelegate).
+        let decisions: Vec<crate::channels::ToolDecision> = tool_calls
+            .iter()
+            .filter_map(|tc| {
+                tc.reasoning.as_ref().map(|r| {
+                    let sanitized = self
+                        .agent
+                        .safety()
+                        .sanitize_tool_output("tool_rationale", r)
+                        .content;
+                    crate::channels::ToolDecision {
+                        tool_name: tc.name.clone(),
+                        rationale: sanitized,
+                    }
+                })
+            })
+            .collect();
+
+        // Emit reasoning update to channels.
+        if narrative.is_some() || !decisions.is_empty() {
+            let _ = self
+                .agent
+                .channels
+                .send_status(
+                    &self.message.channel,
+                    StatusUpdate::ReasoningUpdate {
+                        narrative: narrative.clone().unwrap_or_default(),
+                        decisions: decisions.clone(),
+                    },
+                    &self.message.metadata,
+                )
+                .await;
+        }
+
         // Record tool calls in the thread with sensitive params redacted.
         {
             let mut redacted_args: Vec<serde_json::Value> = Vec::with_capacity(tool_calls.len());
@@ -455,8 +537,23 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
             if let Some(thread) = sess.threads.get_mut(&self.thread_id)
                 && let Some(turn) = thread.last_turn_mut()
             {
+                // Set turn-level narrative.
+                if turn.narrative.is_none() {
+                    turn.narrative = narrative;
+                }
                 for (tc, safe_args) in tool_calls.iter().zip(redacted_args) {
-                    turn.record_tool_call(&tc.name, safe_args);
+                    let sanitized_rationale = tc.reasoning.as_ref().map(|r| {
+                        self.agent
+                            .safety()
+                            .sanitize_tool_output("tool_rationale", r)
+                            .content
+                    });
+                    turn.record_tool_call_with_reasoning(
+                        &tc.name,
+                        safe_args,
+                        sanitized_rationale,
+                        Some(tc.id.clone()),
+                    );
                 }
             }
         }
@@ -726,7 +823,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                         if let Some(thread) = sess.threads.get_mut(&self.thread_id)
                             && let Some(turn) = thread.last_turn_mut()
                         {
-                            turn.record_tool_error(error_msg.clone());
+                            turn.record_tool_error_for(&tc.id, error_msg.clone());
                         }
                     }
                     reason_ctx
@@ -852,16 +949,19 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                         Err(e) => format!("Tool '{}' failed: {}", tc.name, e),
                     };
 
-                    // Record sanitized result in thread
+                    // Record sanitized result in thread (identity-based matching).
                     {
                         let mut sess = self.session.lock().await;
                         if let Some(thread) = sess.threads.get_mut(&self.thread_id)
                             && let Some(turn) = thread.last_turn_mut()
                         {
                             if is_tool_error {
-                                turn.record_tool_error(result_content.clone());
+                                turn.record_tool_error_for(&tc.id, result_content.clone());
                             } else {
-                                turn.record_tool_result(serde_json::json!(result_content));
+                                turn.record_tool_result_for(
+                                    &tc.id,
+                                    serde_json::json!(result_content),
+                                );
                             }
                         }
                     }
@@ -1098,15 +1198,23 @@ pub(crate) fn extract_suggestions(text: &str) -> (String, Vec<String>) {
         Regex::new(r"(?s)<suggestions>\s*(.*?)\s*</suggestions>").expect("valid regex") // safety: constant pattern
     });
 
-    // Find the position of the last closing code fence to avoid matching inside code blocks
+    // Build a sorted list of code fence positions to determine open/close pairing.
-    let last_code_fence = text.rfind("```").unwrap_or(0);
+    // A position is "inside" a fenced block when it falls between an odd-numbered
+    // fence (opening) and the next even-numbered fence (closing).
+    let fence_positions: Vec<usize> = text.match_indices("```").map(|(pos, _)| pos).collect();
 
-    // Find all matches, take the last one that's after the last code fence
+    let is_inside_fence = |pos: usize| -> bool {
+        // Count how many fences appear before `pos`. If odd, we're inside a fence.
+        let count = fence_positions.iter().take_while(|&&fp| fp <= pos).count();
+        count % 2 == 1
+    };
+
+    // Find all matches, take the last one that's outside any code fence
     let mut best_match: Option<regex::Match<'_>> = None;
     let mut best_capture: Option<String> = None;
     for caps in RE.captures_iter(text) {
         if let (Some(full), Some(inner)) = (caps.get(0), caps.get(1))
-            && full.start() >= last_code_fence
+            && !is_inside_fence(full.start())
         {
             best_match = Some(full);
             best_capture = Some(inner.as_str().to_string());
@@ -1225,6 +1333,8 @@ mod tests {
             document_extraction: None,
             sandbox_readiness: crate::agent::routine_engine::SandboxReadiness::DisabledByConfig,
             builder: None,
+            llm_backend: "nearai".to_string(),
+            tenant_rates: Arc::new(crate::tenant::TenantRateRegistry::new(4, 3)),
         };
 
         Agent::new(
@@ -1240,10 +1350,14 @@ mod tests {
                 allow_local_tools: false,
                 max_cost_per_day_cents: None,
                 max_actions_per_hour: None,
+                max_cost_per_user_per_day_cents: None,
                 max_tool_iterations: 50,
                 auto_approve_tools: false,
                 default_timezone: "UTC".to_string(),
                 max_tokens_per_job: 0,
+                multi_tenant: false,
+                max_llm_concurrent_per_user: None,
+                max_jobs_concurrent_per_user: None,
             },
             deps,
             Arc::new(ChannelManager::new()),
@@ -1453,11 +1567,13 @@ mod tests {
                     id: "call_2".to_string(),
                     name: "http".to_string(),
                     arguments: serde_json::json!({"url": "https://example.com"}),
+                    reasoning: None,
                 },
                 ToolCall {
                     id: "call_3".to_string(),
                     name: "echo".to_string(),
                     arguments: serde_json::json!({"message": "done"}),
+                    reasoning: None,
                 },
             ],
             user_timezone: None,
@@ -1643,6 +1759,7 @@ mod tests {
                     id: "call_1".to_string(),
                     name: "echo".to_string(),
                     arguments: serde_json::json!({"message": "hi"}),
+                    reasoning: None,
                 }],
             ),
             ChatMessage::tool_result("call_1", "echo", "hi"),
@@ -1735,11 +1852,13 @@ mod tests {
                         id: "c1".to_string(),
                         name: "http".to_string(),
                         arguments: serde_json::json!({}),
+                        reasoning: None,
                     },
                     ToolCall {
                         id: "c2".to_string(),
                         name: "echo".to_string(),
                         arguments: serde_json::json!({}),
+                        reasoning: None,
                     },
                 ],
             ),
@@ -1773,6 +1892,7 @@ mod tests {
                     id: "c1".to_string(),
                     name: "echo".to_string(),
                     arguments: serde_json::json!({}),
+                    reasoning: None,
                 }],
             ),
             ChatMessage::tool_result("c1", "echo", "done"),
@@ -1903,6 +2023,7 @@ mod tests {
                     id: crate::llm::generate_tool_call_id(0, 0),
                     name: "echo".to_string(),
                     arguments: serde_json::json!({"message": "looping"}),
+                    reasoning: None,
                 }],
                 input_tokens: 0,
                 output_tokens: 5,
@@ -2056,6 +2177,7 @@ mod tests {
                     id: crate::llm::generate_tool_call_id(0, 0),
                     name: "nonexistent_tool".to_string(),
                     arguments: serde_json::json!({}),
+                    reasoning: None,
                 }],
                 input_tokens: 0,
                 output_tokens: 5,
@@ -2092,6 +2214,8 @@ mod tests {
             document_extraction: None,
             sandbox_readiness: crate::agent::routine_engine::SandboxReadiness::DisabledByConfig,
             builder: None,
+            llm_backend: "nearai".to_string(),
+            tenant_rates: Arc::new(crate::tenant::TenantRateRegistry::new(4, 3)),
         };
 
         Agent::new(
@@ -2107,10 +2231,14 @@ mod tests {
                 allow_local_tools: false,
                 max_cost_per_day_cents: None,
                 max_actions_per_hour: None,
+                max_cost_per_user_per_day_cents: None,
                 max_tool_iterations,
                 auto_approve_tools: true,
                 default_timezone: "UTC".to_string(),
                 max_tokens_per_job: 0,
+                multi_tenant: false,
+                max_llm_concurrent_per_user: None,
+                max_jobs_concurrent_per_user: None,
             },
             deps,
             Arc::new(ChannelManager::new()),
@@ -2145,13 +2273,14 @@ mod tests {
 
         let message = IncomingMessage::new("test", "test-user", "do something");
         let initial_messages = vec![ChatMessage::user("do something")];
+        let tenant = agent.tenant_ctx("test-user").await;
 
         // The dispatcher must terminate within 5 seconds. If there is an
         // infinite loop bug (e.g., index not advancing on tool failure), the
         // timeout will fire and the test will fail.
         let result = tokio::time::timeout(
             Duration::from_secs(5),
-            agent.run_agentic_loop(&message, session, thread_id, initial_messages),
+            agent.run_agentic_loop(&message, tenant, session, thread_id, initial_messages),
         )
         .await;
 
@@ -2212,6 +2341,8 @@ mod tests {
                 document_extraction: None,
                 sandbox_readiness: crate::agent::routine_engine::SandboxReadiness::DisabledByConfig,
                 builder: None,
+                llm_backend: "nearai".to_string(),
+                tenant_rates: Arc::new(crate::tenant::TenantRateRegistry::new(4, 3)),
             };
 
             Agent::new(
@@ -2227,10 +2358,14 @@ mod tests {
                     allow_local_tools: false,
                     max_cost_per_day_cents: None,
                     max_actions_per_hour: None,
+                    max_cost_per_user_per_day_cents: None,
                     max_tool_iterations: max_iter,
                     auto_approve_tools: true,
                     default_timezone: "UTC".to_string(),
                     max_tokens_per_job: 0,
+                    multi_tenant: false,
+                    max_llm_concurrent_per_user: None,
+                    max_jobs_concurrent_per_user: None,
                 },
                 deps,
                 Arc::new(ChannelManager::new()),
@@ -2250,13 +2385,14 @@ mod tests {
 
         let message = IncomingMessage::new("test", "test-user", "keep calling tools");
         let initial_messages = vec![ChatMessage::user("keep calling tools")];
+        let tenant = agent.tenant_ctx("test-user").await;
 
         // Even with an LLM that always wants to call tools, the dispatcher
         // must terminate within the timeout thanks to force_text at
         // max_tool_iterations.
         let result = tokio::time::timeout(
             Duration::from_secs(5),
-            agent.run_agentic_loop(&message, session, thread_id, initial_messages),
+            agent.run_agentic_loop(&message, tenant, session, thread_id, initial_messages),
         )
         .await;
 
@@ -2345,6 +2481,16 @@ mod tests {
         assert!(suggestions.is_empty()); // safety: test
     }
 
+    #[test]
+    fn test_extract_suggestions_inside_unclosed_code_fence() {
+        // Regression: odd number of fences (unclosed fence) must still be
+        // treated as "inside a code block".
+        let input = "```\ncode\n<suggestions>[\"bar\"]</suggestions>";
+        let (text, suggestions) = super::extract_suggestions(input);
+        assert_eq!(text, input); // safety: test
+        assert!(suggestions.is_empty()); // safety: test
+    }
+
     #[test]
     fn test_extract_suggestions_after_code_fence() {
         let input = "```\ncode\n```\nAnswer.\n<suggestions>[\"foo\"]</suggestions>";

src/agent/heartbeat.rs

@@ -31,8 +31,8 @@ use chrono_tz::Tz;
 use tokio::sync::mpsc;
 
 use crate::channels::OutgoingResponse;
-use crate::db::Database;
 use crate::llm::{ChatMessage, CompletionRequest, LlmProvider, Reasoning};
+use crate::tenant::AdminScope;
 use crate::workspace::Workspace;
 use crate::workspace::hygiene::HygieneConfig;
 
@@ -57,6 +57,9 @@ pub struct HeartbeatConfig {
     pub quiet_hours_end: Option<u32>,
     /// Timezone for fire_at and quiet hours evaluation (IANA name).
     pub timezone: Option<String>,
+    /// When true, cycle through all users with routines instead of
+    /// running heartbeat for a single user. Requires a database store.
+    pub multi_tenant: bool,
 }
 
 impl Default for HeartbeatConfig {
@@ -71,6 +74,7 @@ impl Default for HeartbeatConfig {
             quiet_hours_start: None,
             quiet_hours_end: None,
             timezone: None,
+            multi_tenant: false,
         }
     }
 }
@@ -178,7 +182,7 @@ pub struct HeartbeatRunner {
     workspace: Arc<Workspace>,
     llm: Arc<dyn LlmProvider>,
     response_tx: Option<mpsc::Sender<OutgoingResponse>>,
-    store: Option<Arc<dyn Database>>,
+    store: Option<AdminScope>,
     consecutive_failures: u32,
 }
 
@@ -207,8 +211,8 @@ impl HeartbeatRunner {
         self
     }
 
-    /// Set the database store for persistent heartbeat conversations.
+    /// Set the admin-scoped database store for persistent heartbeat conversations.
-    pub fn with_store(mut self, store: Arc<dyn Database>) -> Self {
+    pub fn with_store(mut self, store: AdminScope) -> Self {
         self.store = Some(store);
         self
     }
@@ -396,7 +400,7 @@ impl HeartbeatRunner {
     }
 
     /// Send a notification about heartbeat findings.
-    async fn send_notification(&self, message: &str) {
+    pub(crate) async fn send_notification(&self, message: &str) {
         let Some(ref tx) = self.response_tx else {
             tracing::debug!("No response channel configured for heartbeat notifications");
             return;
@@ -493,7 +497,7 @@ pub fn spawn_heartbeat(
     workspace: Arc<Workspace>,
     llm: Arc<dyn LlmProvider>,
     response_tx: Option<mpsc::Sender<OutgoingResponse>>,
-    store: Option<Arc<dyn Database>>,
+    store: Option<AdminScope>,
 ) -> tokio::task::JoinHandle<()> {
     let mut runner = HeartbeatRunner::new(config, hygiene_config, workspace, llm);
     if let Some(tx) = response_tx {
@@ -508,6 +512,179 @@ pub fn spawn_heartbeat(
     })
 }
 
+/// Spawn a multi-user heartbeat runner that cycles through all users that
+/// own routines (enabled or not). Each tick, it queries the DB for distinct
+/// user_ids, creates a per-user workspace, and runs a heartbeat check for
+/// each user concurrently. Per-user failure counts are tracked independently.
+pub fn spawn_multi_user_heartbeat(
+    config: HeartbeatConfig,
+    hygiene_config: HygieneConfig,
+    llm: Arc<dyn LlmProvider>,
+    response_tx: Option<mpsc::Sender<OutgoingResponse>>,
+    store: AdminScope,
+) -> tokio::task::JoinHandle<()> {
+    tokio::spawn(async move {
+        if !config.enabled {
+            tracing::info!("Multi-user heartbeat is disabled");
+            return;
+        }
+
+        let mut tick_interval = if config.fire_at.is_none() {
+            let mut iv = tokio::time::interval(config.interval);
+            iv.tick().await; // skip immediate tick
+            Some(iv)
+        } else {
+            None
+        };
+
+        // Track consecutive failures per user so we can disable heartbeat
+        // for persistently-failing users (same semantics as single-user mode).
+        let mut user_failures: std::collections::HashMap<String, u32> =
+            std::collections::HashMap::new();
+
+        tracing::info!("Starting multi-user heartbeat loop");
+
+        loop {
+            if let Some(fire_at) = config.fire_at {
+                let sleep_dur = duration_until_next_fire(fire_at, config.resolved_tz());
+                tokio::time::sleep(sleep_dur).await;
+            } else if let Some(ref mut iv) = tick_interval {
+                iv.tick().await;
+            }
+
+            if config.is_quiet_hours() {
+                continue;
+            }
+
+            // Get distinct user_ids from routines
+            let user_ids = match store.list_all_routines().await {
+                Ok(routines) => {
+                    let mut ids: Vec<String> = routines
+                        .iter()
+                        .map(|r| r.user_id.clone())
+                        .collect::<std::collections::HashSet<_>>()
+                        .into_iter()
+                        .collect();
+                    ids.sort();
+                    ids
+                }
+                Err(e) => {
+                    tracing::error!("Multi-user heartbeat: failed to list routines: {}", e);
+                    continue;
+                }
+            };
+
+            // Run user heartbeats concurrently so one slow LLM call doesn't
+            // block others. Cap concurrency to avoid flooding the LLM provider.
+            const MAX_CONCURRENT_HEARTBEATS: usize = 8;
+            let mut join_set = tokio::task::JoinSet::new();
+
+            for user_id in &user_ids {
+                // Skip users that have exceeded max_failures
+                let failures = user_failures.get(user_id).copied().unwrap_or(0);
+                if failures >= config.max_failures {
+                    continue;
+                }
+
+                let workspace = Arc::new(Workspace::new_with_db(user_id, Arc::clone(store.db())));
+
+                // Run memory hygiene per user (same as single-user heartbeat).
+                let hygiene_ws = Arc::clone(&workspace);
+                let hygiene_cfg = hygiene_config.clone();
+                let hygiene_user = user_id.clone();
+                tokio::spawn(async move {
+                    let report =
+                        crate::workspace::hygiene::run_if_due(&hygiene_ws, &hygiene_cfg).await;
+                    if report.had_work() {
+                        tracing::info!(
+                            user_id = hygiene_user,
+                            daily_logs_deleted = report.daily_logs_deleted,
+                            conversation_docs_deleted = report.conversation_docs_deleted,
+                            "multi-user heartbeat: memory hygiene deleted stale documents"
+                        );
+                    }
+                });
+
+                // Drain completed tasks to stay within the concurrency cap.
+                while join_set.len() >= MAX_CONCURRENT_HEARTBEATS {
+                    if let Some(join_result) = join_set.join_next().await {
+                        collect_heartbeat_result(join_result, &mut user_failures, &config);
+                    }
+                }
+
+                let uid = user_id.clone();
+                let cfg = config.clone();
+                let hyg = hygiene_config.clone();
+                let llm_clone = llm.clone();
+                let tx = response_tx.clone();
+                let admin = store.clone();
+
+                join_set.spawn(async move {
+                    let mut runner = HeartbeatRunner::new(cfg, hyg, workspace, llm_clone);
+                    if let Some(tx) = tx {
+                        runner = runner.with_response_channel(tx);
+                    }
+                    runner = runner.with_store(admin);
+
+                    let result = runner.check_heartbeat().await;
+                    if let HeartbeatResult::NeedsAttention(msg) = &result {
+                        runner.send_notification(msg).await;
+                    }
+                    (uid, result)
+                });
+            }
+
+            // Collect remaining results and update failure counts
+            while let Some(join_result) = join_set.join_next().await {
+                collect_heartbeat_result(join_result, &mut user_failures, &config);
+            }
+        }
+    })
+}
+
+/// Process a single JoinSet result from the multi-user heartbeat loop.
+fn collect_heartbeat_result(
+    join_result: Result<(String, HeartbeatResult), tokio::task::JoinError>,
+    user_failures: &mut std::collections::HashMap<String, u32>,
+    config: &HeartbeatConfig,
+) {
+    let (uid, result) = match join_result {
+        Ok(pair) => pair,
+        Err(e) => {
+            tracing::error!("Multi-user heartbeat task panicked: {}", e);
+            return;
+        }
+    };
+    match result {
+        HeartbeatResult::Ok => {
+            tracing::trace!(user_id = uid, "Multi-user heartbeat OK");
+            user_failures.remove(&uid);
+        }
+        HeartbeatResult::NeedsAttention(_) => {
+            tracing::info!(user_id = uid, "Multi-user heartbeat needs attention");
+            user_failures.remove(&uid);
+        }
+        HeartbeatResult::Skipped => {}
+        HeartbeatResult::Failed(err) => {
+            let count = user_failures.entry(uid.clone()).or_insert(0);
+            *count += 1;
+            tracing::error!(
+                user_id = uid,
+                consecutive_failures = *count,
+                "Multi-user heartbeat failed: {}",
+                err
+            );
+            if *count >= config.max_failures {
+                tracing::error!(
+                    user_id = uid,
+                    "Multi-user heartbeat disabled for user after {} consecutive failures",
+                    count
+                );
+            }
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -726,7 +903,7 @@ mod tests {
             Arc<crate::workspace::Workspace>,
             Arc<dyn crate::llm::LlmProvider>,
             Option<tokio::sync::mpsc::Sender<crate::channels::OutgoingResponse>>,
-            Option<Arc<dyn crate::db::Database>>,
+            Option<AdminScope>,
         ) -> tokio::task::JoinHandle<()> = spawn_heartbeat;
         let _ = _fn_ptr;
     }

src/agent/job_monitor.rs

@@ -21,8 +21,8 @@ use tokio::task::JoinHandle;
 use uuid::Uuid;
 
 use crate::channels::IncomingMessage;
-use crate::channels::web::types::SseEvent;
 use crate::context::{ContextManager, JobState};
+use ironclaw_common::AppEvent;
 
 /// Route context for forwarding job monitor events back to the user's channel.
 #[derive(Debug, Clone)]
@@ -36,15 +36,15 @@ pub struct JobMonitorRoute {
 /// injects assistant messages into the agent loop.
 ///
 /// The monitor forwards:
-/// - `SseEvent::JobMessage` (assistant role): injected as incoming messages so
+/// - `AppEvent::JobMessage` (assistant role): injected as incoming messages so
 ///   the main agent can read and relay to the user.
-/// - `SseEvent::JobResult`: injected as a completion notice, then the task exits.
+/// - `AppEvent::JobResult`: injected as a completion notice, then the task exits.
 ///
 /// Tool use/result and status events are intentionally skipped (too noisy for
 /// the main agent's context window).
 pub fn spawn_job_monitor(
     job_id: Uuid,
-    event_rx: broadcast::Receiver<(Uuid, String, SseEvent)>,
+    event_rx: broadcast::Receiver<(Uuid, String, AppEvent)>,
     inject_tx: mpsc::Sender<IncomingMessage>,
     route: JobMonitorRoute,
 ) -> JoinHandle<()> {
@@ -56,7 +56,7 @@ pub fn spawn_job_monitor(
 /// jobs don't stay `InProgress` forever in the `ContextManager`.
 pub fn spawn_job_monitor_with_context(
     job_id: Uuid,
-    mut event_rx: broadcast::Receiver<(Uuid, String, SseEvent)>,
+    mut event_rx: broadcast::Receiver<(Uuid, String, AppEvent)>,
     inject_tx: mpsc::Sender<IncomingMessage>,
     route: JobMonitorRoute,
     context_manager: Option<Arc<ContextManager>>,
@@ -74,7 +74,7 @@ pub fn spawn_job_monitor_with_context(
                     }
 
                     match event {
-                        SseEvent::JobMessage { role, content, .. } if role == "assistant" => {
+                        AppEvent::JobMessage { role, content, .. } if role == "assistant" => {
                             let mut msg = IncomingMessage::new(
                                 route.channel.clone(),
                                 route.user_id.clone(),
@@ -92,7 +92,7 @@ pub fn spawn_job_monitor_with_context(
                                 break;
                             }
                         }
-                        SseEvent::JobResult { status, .. } => {
+                        AppEvent::JobResult { status, .. } => {
                             // Transition in-memory state so the job frees its
                             // max_jobs slot and query tools show the final state.
                             if let Some(ref cm) = context_manager {
@@ -162,7 +162,7 @@ pub fn spawn_job_monitor_with_context(
 /// inject messages into) but we still need to free the `max_jobs` slot.
 pub fn spawn_completion_watcher(
     job_id: Uuid,
-    mut event_rx: broadcast::Receiver<(Uuid, String, SseEvent)>,
+    mut event_rx: broadcast::Receiver<(Uuid, String, AppEvent)>,
     context_manager: Arc<ContextManager>,
 ) -> JoinHandle<()> {
     let short_id = job_id.to_string()[..8].to_string();
@@ -170,7 +170,7 @@ pub fn spawn_completion_watcher(
     tokio::spawn(async move {
         loop {
             match event_rx.recv().await {
-                Ok((ev_job_id, _user_id, SseEvent::JobResult { status, .. }))
+                Ok((ev_job_id, _user_id, AppEvent::JobResult { status, .. }))
                     if ev_job_id == job_id =>
                 {
                     let target = if status == "completed" {
@@ -229,7 +229,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_monitor_forwards_assistant_messages() {
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let (inject_tx, mut inject_rx) = mpsc::channel::<IncomingMessage>(16);
 
         let job_id = Uuid::new_v4();
@@ -240,7 +240,7 @@ mod tests {
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobMessage {
+                AppEvent::JobMessage {
                     job_id: job_id.to_string(),
                     role: "assistant".to_string(),
                     content: "I found a bug".to_string(),
@@ -262,7 +262,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_monitor_ignores_other_jobs() {
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let (inject_tx, mut inject_rx) = mpsc::channel::<IncomingMessage>(16);
 
         let job_id = Uuid::new_v4();
@@ -274,7 +274,7 @@ mod tests {
             .send((
                 other_job_id,
                 "test-user".to_string(),
-                SseEvent::JobMessage {
+                AppEvent::JobMessage {
                     job_id: other_job_id.to_string(),
                     role: "assistant".to_string(),
                     content: "wrong job".to_string(),
@@ -293,7 +293,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_monitor_exits_on_job_result() {
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let (inject_tx, mut inject_rx) = mpsc::channel::<IncomingMessage>(16);
 
         let job_id = Uuid::new_v4();
@@ -304,7 +304,7 @@ mod tests {
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobResult {
+                AppEvent::JobResult {
                     job_id: job_id.to_string(),
                     status: "completed".to_string(),
                     session_id: None,
@@ -329,7 +329,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_monitor_skips_tool_events() {
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let (inject_tx, mut inject_rx) = mpsc::channel::<IncomingMessage>(16);
 
         let job_id = Uuid::new_v4();
@@ -340,7 +340,7 @@ mod tests {
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobToolUse {
+                AppEvent::JobToolUse {
                     job_id: job_id.to_string(),
                     tool_name: "shell".to_string(),
                     input: serde_json::json!({"command": "ls"}),
@@ -353,7 +353,7 @@ mod tests {
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobMessage {
+                AppEvent::JobMessage {
                     job_id: job_id.to_string(),
                     role: "user".to_string(),
                     content: "user prompt".to_string(),
@@ -409,7 +409,7 @@ mod tests {
             .await
             .unwrap();
 
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let (inject_tx, mut inject_rx) = mpsc::channel::<IncomingMessage>(16);
 
         let handle = spawn_job_monitor_with_context(
@@ -425,7 +425,7 @@ mod tests {
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobResult {
+                AppEvent::JobResult {
                     job_id: job_id.to_string(),
                     status: "completed".to_string(),
                     session_id: None,
@@ -458,7 +458,7 @@ mod tests {
             .await
             .unwrap();
 
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let (inject_tx, mut inject_rx) = mpsc::channel::<IncomingMessage>(16);
 
         let handle = spawn_job_monitor_with_context(
@@ -474,7 +474,7 @@ mod tests {
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobResult {
+                AppEvent::JobResult {
                     job_id: job_id.to_string(),
                     status: "failed".to_string(),
                     session_id: None,
@@ -507,14 +507,14 @@ mod tests {
             .await
             .unwrap();
 
-        let (event_tx, _) = broadcast::channel::<(Uuid, String, SseEvent)>(16);
+        let (event_tx, _) = broadcast::channel::<(Uuid, String, AppEvent)>(16);
         let handle = spawn_completion_watcher(job_id, event_tx.subscribe(), Arc::clone(&cm));
 
         event_tx
             .send((
                 job_id,
                 "test-user".to_string(),
-                SseEvent::JobResult {
+                AppEvent::JobResult {
                     job_id: job_id.to_string(),
                     status: "completed".to_string(),
                     session_id: None,

src/agent/mod.rs

@@ -36,7 +36,9 @@ pub(crate) use agent_loop::truncate_for_preview;
 pub use agent_loop::{Agent, AgentDeps};
 pub use compaction::{CompactionResult, ContextCompactor};
 pub use context_monitor::{CompactionStrategy, ContextBreakdown, ContextMonitor};
-pub use heartbeat::{HeartbeatConfig, HeartbeatResult, HeartbeatRunner, spawn_heartbeat};
+pub use heartbeat::{
+    HeartbeatConfig, HeartbeatResult, HeartbeatRunner, spawn_heartbeat, spawn_multi_user_heartbeat,
+};
 pub use router::{MessageIntent, Router};
 pub use routine::{Routine, RoutineAction, RoutineRun, Trigger};
 pub use routine_engine::{RoutineEngine, SandboxReadiness};

src/agent/routine_engine.rs

@@ -18,21 +18,22 @@ use std::time::Duration;
 use chrono::Utc;
 use regex::Regex;
 use tokio::sync::{RwLock, mpsc};
+use tokio::task::JoinHandle;
 use uuid::Uuid;
 
 use crate::agent::Scheduler;
 use crate::agent::routine::{
     NotifyConfig, Routine, RoutineAction, RoutineRun, RunStatus, Trigger, next_cron_fire,
 };
-use crate::channels::OutgoingResponse;
+use crate::channels::{IncomingMessage, OutgoingResponse};
 use crate::config::RoutineConfig;
 use crate::context::{JobContext, JobState};
-use crate::db::Database;
 use crate::error::RoutineError;
 use crate::extensions::ExtensionManager;
 use crate::llm::{
     ChatMessage, CompletionRequest, FinishReason, LlmProvider, ToolCall, ToolCompletionRequest,
 };
+use crate::tenant::AdminScope;
 use crate::tools::{
     ToolError, ToolRegistry, autonomous_allowed_tool_names, autonomous_unavailable_message,
     prepare_tool_params,
@@ -45,6 +46,11 @@ enum EventMatcher {
     System { routine: Routine },
 }
 
+struct TriggeredRoutine {
+    routine: Routine,
+    detail: String,
+}
+
 /// Distinguishes why sandbox is unavailable so error messages are accurate.
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum SandboxReadiness {
@@ -56,10 +62,44 @@ pub enum SandboxReadiness {
     DockerUnavailable,
 }
 
+/// Check whether an event-triggered routine's user/channel filters match an
+/// incoming message.
+///
+/// Returns `true` if:
+/// - The routine has an `Event` trigger (non-Event routines always return `false`)
+/// - The routine's `user_id` matches the message's user scope
+/// - The routine's channel filter (if any) matches the message channel
+///   case-insensitively
+///
+/// This is a pure function extracted from `check_event_triggers` so the
+/// filter logic can be unit-tested without async infrastructure.
+pub(crate) fn routine_matches_message(routine: &Routine, message: &IncomingMessage) -> bool {
+    // Only Event-triggered routines can match incoming messages.
+    if !matches!(routine.trigger, Trigger::Event { .. }) {
+        return false;
+    }
+
+    // User ownership filter — only fire routines scoped to this user.
+    if routine.user_id != message.user_id {
+        return false;
+    }
+
+    // Channel filter (case-insensitive, matching emit_system_event behavior)
+    if let Trigger::Event {
+        channel: Some(ch), ..
+    } = &routine.trigger
+        && !ch.eq_ignore_ascii_case(&message.channel)
+    {
+        return false;
+    }
+
+    true
+}
+
 /// The routine execution engine.
 pub struct RoutineEngine {
     config: RoutineConfig,
-    store: Arc<dyn Database>,
+    store: AdminScope,
     llm: Arc<dyn LlmProvider>,
     workspace: Arc<Workspace>,
     /// Sender for notifications (routed to channel manager).
@@ -88,7 +128,7 @@ impl RoutineEngine {
     #[allow(clippy::too_many_arguments)]
     pub fn new(
         config: RoutineConfig,
-        store: Arc<dyn Database>,
+        store: AdminScope,
         llm: Arc<dyn LlmProvider>,
         workspace: Arc<Workspace>,
         notify_tx: mpsc::Sender<OutgoingResponse>,
@@ -167,10 +207,45 @@ impl RoutineEngine {
     }
 
     /// Check incoming message against event triggers. Returns number of routines fired.
+    pub async fn check_event_triggers(&self, message: &IncomingMessage, content: &str) -> usize {
+        let triggered = self.matching_event_triggers(message, content).await;
+        let fired = triggered.len();
+        for triggered in triggered {
+            std::mem::drop(self.spawn_fire(triggered.routine, "event", Some(triggered.detail)));
+        }
+        fired
+    }
+
+    /// Fire matching event-triggered routines and wait for them to complete.
     ///
-    /// Accepts only the three fields needed for matching (user scope, channel,
+    /// Used by single-message REPL mode so the process does not exit before
-    /// message content) so callers never need to clone a full `IncomingMessage`.
+    /// background event-triggered routines finish.
-    pub async fn check_event_triggers(&self, user_id: &str, channel: &str, content: &str) -> usize {
+    pub async fn check_event_triggers_and_wait(
+        &self,
+        message: &IncomingMessage,
+        content: &str,
+    ) -> usize {
+        let triggered = self.matching_event_triggers(message, content).await;
+        let fired = triggered.len();
+        let handles: Vec<JoinHandle<()>> = triggered
+            .into_iter()
+            .map(|triggered| self.spawn_fire(triggered.routine, "event", Some(triggered.detail)))
+            .collect();
+
+        for handle in handles {
+            if let Err(e) = handle.await {
+                tracing::warn!(error = %e, "Event-triggered routine task failed");
+            }
+        }
+
+        fired
+    }
+
+    async fn matching_event_triggers(
+        &self,
+        message: &IncomingMessage,
+        content: &str,
+    ) -> Vec<TriggeredRoutine> {
         let cache = self.event_cache.read().await;
 
         // Early return if there are no message matchers at all.
@@ -178,10 +253,9 @@ impl RoutineEngine {
             .iter()
             .any(|m| matches!(m, EventMatcher::Message { .. }))
         {
-            return 0;
+            return Vec::new();
         }
-
+        let mut triggered = Vec::new();
-        let mut fired = 0;
 
         // Collect routine IDs for batch query
         let routine_ids: Vec<Uuid> = cache
@@ -193,13 +267,13 @@ impl RoutineEngine {
             .collect();
 
         if routine_ids.is_empty() {
-            return 0;
+            return Vec::new();
         }
 
         // Single batch query instead of N queries
         let concurrent_counts = match self.batch_concurrent_counts(&routine_ids).await {
             Some(counts) => counts,
-            None => return 0,
+            None => return Vec::new(),
         };
 
         for matcher in cache.iter() {
@@ -208,16 +282,24 @@ impl RoutineEngine {
                 EventMatcher::System { .. } => continue,
             };
 
-            if routine.user_id != user_id {
+            // User ownership + channel filter (extracted for testability).
-                continue;
+            if !routine_matches_message(routine, message) {
-            }
+                // User mismatch is expected for multi-user setups — keep at
-
+                // trace to avoid one log per routine per inbound message.
-            // Channel filter
+                if routine.user_id != message.user_id {
-            if let Trigger::Event {
+                    tracing::trace!(
-                channel: Some(ch), ..
+                        routine = %routine.name,
-            } = &routine.trigger
+                        routine_user = %routine.user_id,
-                && ch != channel
+                        message_user = %message.user_id,
-            {
+                        "Skipped: user scope mismatch"
+                    );
+                } else {
+                    tracing::debug!(
+                        routine = %routine.name,
+                        channel = %message.channel,
+                        "Skipped: channel mismatch"
+                    );
+                }
                 continue;
             }
 
@@ -228,14 +310,14 @@ impl RoutineEngine {
 
             // Cooldown check
             if !self.check_cooldown(routine) {
-                tracing::trace!(routine = %routine.name, "Skipped: cooldown active");
+                tracing::debug!(routine = %routine.name, "Skipped: cooldown active");
                 continue;
             }
 
             // Concurrent run check (using batch-loaded counts)
             let running_count = concurrent_counts.get(&routine.id).copied().unwrap_or(0);
             if running_count >= routine.guardrails.max_concurrent as i64 {
-                tracing::trace!(routine = %routine.name, "Skipped: max concurrent reached");
+                tracing::debug!(routine = %routine.name, "Skipped: max concurrent reached");
                 continue;
             }
 
@@ -246,11 +328,13 @@ impl RoutineEngine {
             }
 
             let detail = truncate(content, 200);
-            self.spawn_fire(routine.clone(), "event", Some(detail));
+            triggered.push(TriggeredRoutine {
-            fired += 1;
+                routine: routine.clone(),
+                detail,
+            });
         }
 
-        fired
+        triggered
     }
 
     /// Emit a structured event to system-event routines.
@@ -698,12 +782,22 @@ impl RoutineEngine {
             });
         }
 
+        // Per-user workspace (same pattern as spawn_fire).
+        let routine_workspace = if routine.user_id == self.workspace.user_id() {
+            self.workspace.clone()
+        } else {
+            Arc::new(Workspace::new_with_db(
+                &routine.user_id,
+                Arc::clone(self.store.db()),
+            ))
+        };
+
         // Execute inline for manual triggers (caller wants to wait)
         let engine = EngineContext {
             config: self.config.clone(),
             store: self.store.clone(),
             llm: self.llm.clone(),
-            workspace: self.workspace.clone(),
+            workspace: routine_workspace,
             notify_tx: self.notify_tx.clone(),
             running_count: self.running_count.clone(),
             scheduler: self.scheduler.clone(),
@@ -806,7 +900,12 @@ impl RoutineEngine {
     }
 
     /// Spawn a fire in a background task.
-    fn spawn_fire(&self, routine: Routine, trigger_type: &str, trigger_detail: Option<String>) {
+    fn spawn_fire(
+        &self,
+        routine: Routine,
+        trigger_type: &str,
+        trigger_detail: Option<String>,
+    ) -> JoinHandle<()> {
         let run = RoutineRun {
             id: Uuid::new_v4(),
             routine_id: routine.id,
@@ -821,11 +920,23 @@ impl RoutineEngine {
             created_at: Utc::now(),
         };
 
+        // Use per-user workspace so each routine executes in the correct
+        // user's context. Fall back to the engine-wide workspace when the
+        // routine belongs to the same user (avoids unnecessary allocation).
+        let routine_workspace = if routine.user_id == self.workspace.user_id() {
+            self.workspace.clone()
+        } else {
+            Arc::new(Workspace::new_with_db(
+                &routine.user_id,
+                Arc::clone(self.store.db()),
+            ))
+        };
+
         let engine = EngineContext {
             config: self.config.clone(),
             store: self.store.clone(),
             llm: self.llm.clone(),
-            workspace: self.workspace.clone(),
+            workspace: routine_workspace,
             notify_tx: self.notify_tx.clone(),
             running_count: self.running_count.clone(),
             scheduler: self.scheduler.clone(),
@@ -843,7 +954,7 @@ impl RoutineEngine {
                 return;
             }
             execute_routine(engine, routine, run).await;
-        });
+        })
     }
 
     fn check_cooldown(&self, routine: &Routine) -> bool {
@@ -878,7 +989,7 @@ impl RoutineEngine {
 /// an active state (Pending/InProgress/Stuck). Maps the final `JobState` to
 /// a `RunStatus` for the routine run.
 struct FullJobWatcher {
-    store: Arc<dyn Database>,
+    store: AdminScope,
     job_id: Uuid,
     routine_name: String,
 }
@@ -889,7 +1000,7 @@ impl FullJobWatcher {
     /// Safety ceiling: 24 hours, derived from POLL_INTERVAL.
     const MAX_POLLS: u32 = (24 * 60 * 60) / Self::POLL_INTERVAL.as_secs() as u32;
 
-    fn new(store: Arc<dyn Database>, job_id: Uuid, routine_name: String) -> Self {
+    fn new(store: AdminScope, job_id: Uuid, routine_name: String) -> Self {
         Self {
             store,
             job_id,
@@ -961,7 +1072,7 @@ impl FullJobWatcher {
 /// Shared context passed to the execution function.
 struct EngineContext {
     config: RoutineConfig,
-    store: Arc<dyn Database>,
+    store: AdminScope,
     llm: Arc<dyn LlmProvider>,
     workspace: Arc<Workspace>,
     notify_tx: mpsc::Sender<OutgoingResponse>,
@@ -1305,6 +1416,19 @@ async fn execute_lightweight(
     }
 }
 
+/// Sanitize a user-controlled string before interpolation into an LLM prompt.
+/// Strips newlines (which could break prompt structure) and truncates to a
+/// reasonable length to limit abuse surface.
+fn sanitize_prompt_field(value: &str) -> String {
+    const MAX_LEN: usize = 128;
+    value
+        .chars()
+        .filter(|&c| c != '\n' && c != '\r')
+        .take(MAX_LEN)
+        .map(|c| if c == '`' { '\'' } else { c })
+        .collect()
+}
+
 fn build_lightweight_prompt(
     prompt: &str,
     context_parts: &[String],
@@ -1323,14 +1447,16 @@ fn build_lightweight_prompt(
         );
 
         if let Some(channel) = notify.channel.as_deref() {
+            let sanitized = sanitize_prompt_field(channel);
             full_prompt.push_str(&format!(
-                "The configured delivery channel for this routine is `{channel}`.\n"
+                "The configured delivery channel for this routine is `{sanitized}`.\n"
             ));
         }
 
         if let Some(user) = notify.user.as_deref() {
+            let sanitized = sanitize_prompt_field(user);
             full_prompt.push_str(&format!(
-                "The configured delivery target for this routine is `{user}`.\n"
+                "The configured delivery target for this routine is `{sanitized}`.\n"
             ));
         }
 
@@ -1440,6 +1566,7 @@ fn handle_text_response(
 /// This is a simplified version of the full dispatcher loop:
 /// - Max 3-5 iterations (configurable)
 /// - Sequential tool execution (not parallel)
+/// - Uses the owner's live autonomous tool scope when lightweight tools are enabled
 /// - Auto-approval of non-Always tools
 /// - No hooks or approval dialogs
 async fn execute_lightweight_with_tools(
@@ -1486,7 +1613,10 @@ async fn execute_lightweight_with_tools(
         let force_text = iteration >= max_iterations;
 
         if force_text {
-            // Final iteration: no tools, just get text response
+            // Final iteration: no tools, just get text response.
+            // Claude 4.6 rejects assistant prefill; NEAR AI rejects any non-user-ending
+            // conversation. Ensure the last message is user-role.
+            crate::util::ensure_ends_with_user_message(&mut messages);
             let request = CompletionRequest::new(messages)
                 .with_max_tokens(effective_max_tokens)
                 .with_temperature(0.3);
@@ -1765,6 +1895,13 @@ pub fn spawn_cron_ticker(
         engine.check_cron_triggers().await;
 
         let mut ticker = tokio::time::interval(interval);
+        ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
+        // Periodic event cache refresh so web/CLI mutations are picked up
+        // without requiring tool-path code to call refresh_event_cache().
+        // Uses wall-clock elapsed time so the refresh cadence is stable
+        // regardless of the cron tick interval configuration.
+        let refresh_interval = Duration::from_secs(60);
+        let mut last_refresh = tokio::time::Instant::now();
 
         loop {
             ticker.tick().await;
@@ -1772,7 +1909,11 @@ pub fn spawn_cron_ticker(
             // never races with FullJobWatcher instances from this process.
             engine.sync_dispatched_runs().await;
             engine.check_cron_triggers().await;
-            engine.sync_dispatched_runs().await;
+
+            if last_refresh.elapsed() >= refresh_interval {
+                engine.refresh_event_cache().await;
+                last_refresh = tokio::time::Instant::now();
+            }
         }
     })
 }
@@ -1838,7 +1979,13 @@ fn strip_html_tags(s: &str) -> String {
 
 #[cfg(test)]
 mod tests {
-    use crate::agent::routine::{NotifyConfig, RunStatus};
+    use chrono::Utc;
+    use uuid::Uuid;
+
+    use crate::agent::routine::{
+        NotifyConfig, Routine, RoutineAction, RoutineGuardrails, RunStatus, Trigger,
+    };
+    use crate::channels::IncomingMessage;
     use crate::config::RoutineConfig;
 
     #[test]
@@ -2036,6 +2183,117 @@ mod tests {
         }
     }
 
+    /// Helper to build a test routine with the given user_id and trigger.
+    fn make_routine(user_id: &str, trigger: Trigger) -> Routine {
+        Routine {
+            id: Uuid::new_v4(),
+            name: "test".to_string(),
+            description: String::new(),
+            user_id: user_id.to_string(),
+            enabled: true,
+            trigger,
+            action: RoutineAction::Lightweight {
+                prompt: String::new(),
+                context_paths: vec![],
+                max_tokens: 1000,
+                use_tools: false,
+                max_tool_rounds: 0,
+            },
+            guardrails: RoutineGuardrails::default(),
+            notify: Default::default(),
+            last_run_at: None,
+            next_fire_at: None,
+            run_count: 0,
+            consecutive_failures: 0,
+            state: serde_json::Value::Null,
+            created_at: Utc::now(),
+            updated_at: Utc::now(),
+        }
+    }
+
+    /// Helper to build a test IncomingMessage.
+    fn make_message(user_id: &str, channel: &str, content: &str) -> IncomingMessage {
+        IncomingMessage {
+            id: Uuid::new_v4(),
+            channel: channel.to_string(),
+            user_id: user_id.to_string(),
+            owner_id: user_id.to_string(),
+            sender_id: user_id.to_string(),
+            user_name: None,
+            content: content.to_string(),
+            thread_id: None,
+            conversation_scope_id: None,
+            received_at: Utc::now(),
+            metadata: serde_json::Value::Null,
+            timezone: None,
+            attachments: vec![],
+            is_internal: false,
+        }
+    }
+
+    /// Regression test for issue #1051: event triggers used case-sensitive
+    /// channel comparison, so "Telegram" != "telegram" caused silent mismatch.
+    /// Tests the actual `routine_matches_message` function used in `check_event_triggers`.
+    #[test]
+    fn test_channel_filter_is_case_insensitive() {
+        let routine = make_routine(
+            "user1",
+            Trigger::Event {
+                pattern: ".*".to_string(),
+                channel: Some("Telegram".to_string()),
+            },
+        );
+        let msg = make_message("user1", "telegram", "hello");
+
+        // Case-insensitive channel match must succeed
+        assert!(super::routine_matches_message(&routine, &msg));
+
+        // Exact case must also work
+        let msg_exact = make_message("user1", "Telegram", "hello");
+        assert!(super::routine_matches_message(&routine, &msg_exact));
+
+        // Different channel must not match
+        let msg_wrong = make_message("user1", "discord", "hello");
+        assert!(!super::routine_matches_message(&routine, &msg_wrong));
+    }
+
+    /// Regression test for issue #1051: event triggers did not filter by
+    /// user_id, so routines from user A could fire on messages from user B.
+    /// Tests the actual `routine_matches_message` function used in `check_event_triggers`.
+    #[test]
+    fn test_event_trigger_requires_user_match() {
+        let routine = make_routine(
+            "alice",
+            Trigger::Event {
+                pattern: ".*".to_string(),
+                channel: None,
+            },
+        );
+
+        // Different user must not match
+        let msg_bob = make_message("bob", "telegram", "hello");
+        assert!(!super::routine_matches_message(&routine, &msg_bob));
+
+        // Same user must match
+        let msg_alice = make_message("alice", "telegram", "hello");
+        assert!(super::routine_matches_message(&routine, &msg_alice));
+    }
+
+    /// When no channel filter is set, any channel should match (given user matches).
+    #[test]
+    fn test_no_channel_filter_matches_any_channel() {
+        let routine = make_routine(
+            "user1",
+            Trigger::Event {
+                pattern: ".*".to_string(),
+                channel: None,
+            },
+        );
+
+        let msg = make_message("user1", "whatever_channel", "hello");
+        assert!(super::routine_matches_message(&routine, &msg));
+    }
+
     #[test]
     fn test_routine_tool_denylist_blocks_self_management_tools() {
         let denylisted = vec![

src/agent/scheduler.rs

@@ -9,15 +9,14 @@ use tokio::task::JoinHandle;
 use uuid::Uuid;
 
 use crate::agent::task::{Task, TaskContext, TaskOutput};
-use crate::channels::web::types::SseEvent;
 use crate::config::AgentConfig;
 use crate::context::{ContextManager, JobContext, JobState};
-use crate::db::Database;
 use crate::error::{Error, JobError};
 use crate::extensions::ExtensionManager;
 use crate::hooks::HookRegistry;
 use crate::llm::LlmProvider;
 use crate::safety::SafetyLayer;
+use crate::tenant::AdminScope;
 use crate::tools::{
     ApprovalContext, ToolRegistry, autonomous_allowed_tool_names, autonomous_unavailable_error,
     prepare_tool_params,
@@ -53,7 +52,7 @@ struct ScheduledSubtask {
 pub struct SchedulerDeps {
     pub tools: Arc<ToolRegistry>,
     pub extension_manager: Option<Arc<ExtensionManager>>,
-    pub store: Option<Arc<dyn Database>>,
+    pub store: Option<AdminScope>,
     pub hooks: Arc<HookRegistry>,
 }
 
@@ -65,10 +64,10 @@ pub struct Scheduler {
     safety: Arc<SafetyLayer>,
     tools: Arc<ToolRegistry>,
     extension_manager: Option<Arc<ExtensionManager>>,
-    store: Option<Arc<dyn Database>>,
+    store: Option<AdminScope>,
     hooks: Arc<HookRegistry>,
-    /// SSE broadcast sender for live job event streaming.
+    /// SSE manager for live job event streaming.
-    sse_tx: Option<tokio::sync::broadcast::Sender<SseEvent>>,
+    sse_tx: Option<Arc<crate::channels::web::sse::SseManager>>,
     /// HTTP interceptor for trace recording/replay (propagated to workers).
     http_interceptor: Option<Arc<dyn crate::llm::recording::HttpInterceptor>>,
     /// Running jobs (main LLM-driven jobs).
@@ -102,9 +101,9 @@ impl Scheduler {
         }
     }
 
-    /// Set the SSE broadcast sender for live job event streaming.
+    /// Set the SSE manager for live job event streaming.
-    pub fn set_sse_sender(&mut self, tx: tokio::sync::broadcast::Sender<SseEvent>) {
+    pub fn set_sse_sender(&mut self, sse: Arc<crate::channels::web::sse::SseManager>) {
-        self.sse_tx = Some(tx);
+        self.sse_tx = Some(sse);
     }
 
     /// Set the HTTP interceptor for trace recording/replay.
@@ -781,10 +780,14 @@ mod tests {
             allow_local_tools: true,
             max_cost_per_day_cents: None,
             max_actions_per_hour: None,
+            max_cost_per_user_per_day_cents: None,
             max_tool_iterations: 10,
             auto_approve_tools: true,
             default_timezone: "UTC".to_string(),
             max_tokens_per_job,
+            multi_tenant: false,
+            max_llm_concurrent_per_user: None,
+            max_jobs_concurrent_per_user: None,
         };
         let cm = Arc::new(ContextManager::new(5));
         let llm: Arc<dyn LlmProvider> = Arc::new(StubLlm);

src/agent/self_repair.rs

@@ -8,8 +8,8 @@ use chrono::{DateTime, Utc};
 use uuid::Uuid;
 
 use crate::context::{ContextManager, JobState};
-use crate::db::Database;
 use crate::error::RepairError;
+use crate::tenant::AdminScope;
 use crate::tools::{BuildRequirement, Language, SoftwareBuilder, SoftwareType, ToolRegistry};
 
 /// A job that has been detected as stuck.
@@ -69,7 +69,7 @@ pub struct DefaultSelfRepair {
     /// Jobs in `InProgress` longer than this are treated as stuck.
     stuck_threshold: Duration,
     max_repair_attempts: u32,
-    store: Option<Arc<dyn Database>>,
+    store: Option<AdminScope>,
     builder: Option<Arc<dyn SoftwareBuilder>>,
     tools: Option<Arc<ToolRegistry>>,
 }
@@ -91,8 +91,8 @@ impl DefaultSelfRepair {
         }
     }
 
-    /// Add a Store for tool failure tracking.
+    /// Add an admin-scoped store for tool failure tracking.
-    pub fn with_store(mut self, store: Arc<dyn Database>) -> Self {
+    pub fn with_store(mut self, store: AdminScope) -> Self {
         self.store = Some(store);
         self
     }
@@ -806,7 +806,7 @@ mod tests {
         // Create self-repair with zero threshold (detect immediately),
         // wired with store, builder, and tools.
         let repair = DefaultSelfRepair::new(Arc::clone(&cm), Duration::from_secs(0), 3)
-            .with_store(Arc::clone(&db))
+            .with_store(crate::tenant::AdminScope::new(Arc::clone(&db)))
             .with_builder(
                 Arc::clone(&builder) as Arc<dyn crate::tools::SoftwareBuilder>,
                 tools,

src/agent/session.rs

@@ -16,8 +16,8 @@ use chrono::{DateTime, TimeDelta, Utc};
 use serde::{Deserialize, Serialize};
 use uuid::Uuid;
 
-use crate::channels::web::util::truncate_preview;
 use crate::llm::{ChatMessage, ToolCall, generate_tool_call_id};
+use ironclaw_common::truncate_preview;
 
 /// A session containing one or more threads.
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -449,6 +449,7 @@ impl Thread {
                         id: call_id.clone(),
                         name: tc.name.clone(),
                         arguments: tc.parameters.clone(),
+                        reasoning: None,
                     })
                     .collect();
 
@@ -522,7 +523,12 @@ impl Thread {
                             && let Some(ref tcs) = assistant_msg.tool_calls
                         {
                             for tc in tcs {
-                                turn.record_tool_call(&tc.name, tc.arguments.clone());
+                                turn.record_tool_call_with_reasoning(
+                                    &tc.name,
+                                    tc.arguments.clone(),
+                                    tc.reasoning.clone(),
+                                    Some(tc.id.clone()),
+                                );
                             }
                         }
 
@@ -602,6 +608,10 @@ pub struct Turn {
     pub completed_at: Option<DateTime<Utc>>,
     /// Error message (if failed).
     pub error: Option<String>,
+    /// Agent's reasoning narrative for this turn.
+    /// Cleaned via `clean_response` and sanitized through `SafetyLayer` before storage.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub narrative: Option<String>,
     /// Transient image content parts for multimodal LLM input.
     /// Not serialized — images are only needed for the current LLM call.
     /// The text description in `user_input` persists for compaction/context.
@@ -621,6 +631,7 @@ impl Turn {
             started_at: Utc::now(),
             completed_at: None,
             error: None,
+            narrative: None,
             image_content_parts: Vec::new(),
         }
     }
@@ -656,6 +667,26 @@ impl Turn {
             parameters: params,
             result: None,
             error: None,
+            rationale: None,
+            tool_call_id: None,
+        });
+    }
+
+    /// Record a tool call with reasoning context.
+    pub fn record_tool_call_with_reasoning(
+        &mut self,
+        name: impl Into<String>,
+        params: serde_json::Value,
+        rationale: Option<String>,
+        tool_call_id: Option<String>,
+    ) {
+        self.tool_calls.push(TurnToolCall {
+            name: name.into(),
+            parameters: params,
+            result: None,
+            error: None,
+            rationale,
+            tool_call_id,
         });
     }
 
@@ -672,6 +703,60 @@ impl Turn {
             call.error = Some(error.into());
         }
     }
+
+    /// Record a tool result by tool_call_id, with fallback to first pending call.
+    pub fn record_tool_result_for(&mut self, tool_call_id: &str, result: serde_json::Value) {
+        if let Some(call) = self
+            .tool_calls
+            .iter_mut()
+            .find(|c| c.tool_call_id.as_deref() == Some(tool_call_id))
+        {
+            call.result = Some(result);
+        } else if let Some(call) = self
+            .tool_calls
+            .iter_mut()
+            .find(|c| c.result.is_none() && c.error.is_none())
+        {
+            tracing::debug!(
+                tool_call_id = %tool_call_id,
+                fallback_tool = %call.name,
+                "tool_call_id not found, falling back to first pending call"
+            );
+            call.result = Some(result);
+        } else {
+            tracing::warn!(
+                tool_call_id = %tool_call_id,
+                "Tool result dropped: no matching or pending tool call"
+            );
+        }
+    }
+
+    /// Record a tool error by tool_call_id, with fallback to first pending call.
+    pub fn record_tool_error_for(&mut self, tool_call_id: &str, error: impl Into<String>) {
+        if let Some(call) = self
+            .tool_calls
+            .iter_mut()
+            .find(|c| c.tool_call_id.as_deref() == Some(tool_call_id))
+        {
+            call.error = Some(error.into());
+        } else if let Some(call) = self
+            .tool_calls
+            .iter_mut()
+            .find(|c| c.result.is_none() && c.error.is_none())
+        {
+            tracing::debug!(
+                tool_call_id = %tool_call_id,
+                fallback_tool = %call.name,
+                "tool_call_id not found, falling back to first pending call"
+            );
+            call.error = Some(error.into());
+        } else {
+            tracing::warn!(
+                tool_call_id = %tool_call_id,
+                "Tool error dropped: no matching or pending tool call"
+            );
+        }
+    }
 }
 
 /// Record of a tool call made during a turn.
@@ -685,6 +770,12 @@ pub struct TurnToolCall {
     pub result: Option<serde_json::Value>,
     /// Error from the tool (if failed).
     pub error: Option<String>,
+    /// Agent's reasoning for choosing this tool.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub rationale: Option<String>,
+    /// The tool_call_id from the LLM, for identity-based result matching.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub tool_call_id: Option<String>,
 }
 
 #[cfg(test)]
@@ -1309,6 +1400,7 @@ mod tests {
             id: "call_0".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"q": "test"}),
+            reasoning: None,
         };
         let messages = vec![
             ChatMessage::user("Find test"),
@@ -1339,6 +1431,7 @@ mod tests {
             id: "call_0".to_string(),
             name: "http".to_string(),
             arguments: serde_json::json!({}),
+            reasoning: None,
         };
         let messages = vec![
             ChatMessage::user("Fetch URL"),
@@ -1404,11 +1497,13 @@ mod tests {
             id: "call_a".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"q": "data"}),
+            reasoning: None,
         };
         let tc2 = ToolCall {
             id: "call_b".to_string(),
             name: "write".to_string(),
             arguments: serde_json::json!({"path": "out.txt"}),
+            reasoning: None,
         };
         let messages = vec![
             ChatMessage::user("Find and save"),
@@ -1620,4 +1715,100 @@ mod tests {
         let merged = thread.drain_pending_messages().unwrap();
         assert_eq!(merged, "failed batch\nnew msg");
     }
+
+    #[test]
+    fn test_record_tool_result_for_by_id() {
+        let mut turn = Turn::new(0, "test");
+        turn.record_tool_call_with_reasoning(
+            "tool_a",
+            serde_json::json!({}),
+            None,
+            Some("id_a".into()),
+        );
+        turn.record_tool_call_with_reasoning(
+            "tool_b",
+            serde_json::json!({}),
+            None,
+            Some("id_b".into()),
+        );
+
+        // Record result for second tool by ID
+        turn.record_tool_result_for("id_b", serde_json::json!("result_b"));
+        assert!(turn.tool_calls[0].result.is_none());
+        assert_eq!(
+            turn.tool_calls[1].result.as_ref().unwrap(),
+            &serde_json::json!("result_b")
+        );
+    }
+
+    #[test]
+    fn test_record_tool_error_for_by_id() {
+        let mut turn = Turn::new(0, "test");
+        turn.record_tool_call_with_reasoning(
+            "tool_a",
+            serde_json::json!({}),
+            None,
+            Some("id_a".into()),
+        );
+        turn.record_tool_call_with_reasoning(
+            "tool_b",
+            serde_json::json!({}),
+            None,
+            Some("id_b".into()),
+        );
+
+        turn.record_tool_error_for("id_a", "failed");
+        assert_eq!(turn.tool_calls[0].error.as_deref(), Some("failed"));
+        assert!(turn.tool_calls[1].error.is_none());
+    }
+
+    #[test]
+    fn test_record_tool_result_for_fallback_to_pending() {
+        let mut turn = Turn::new(0, "test");
+        turn.record_tool_call_with_reasoning(
+            "tool_a",
+            serde_json::json!({}),
+            None,
+            Some("id_a".into()),
+        );
+        turn.record_tool_call_with_reasoning(
+            "tool_b",
+            serde_json::json!({}),
+            None,
+            Some("id_b".into()),
+        );
+
+        // First tool already has a result
+        turn.tool_calls[0].result = Some(serde_json::json!("done"));
+
+        // Unknown ID should fall back to first pending (tool_b)
+        turn.record_tool_result_for("unknown_id", serde_json::json!("fallback"));
+        assert_eq!(
+            turn.tool_calls[0].result.as_ref().unwrap(),
+            &serde_json::json!("done")
+        );
+        assert_eq!(
+            turn.tool_calls[1].result.as_ref().unwrap(),
+            &serde_json::json!("fallback")
+        );
+    }
+
+    #[test]
+    fn test_record_tool_result_for_no_pending_is_noop() {
+        let mut turn = Turn::new(0, "test");
+        turn.record_tool_call_with_reasoning(
+            "tool_a",
+            serde_json::json!({}),
+            None,
+            Some("id_a".into()),
+        );
+        turn.tool_calls[0].result = Some(serde_json::json!("done"));
+
+        // No pending calls, unknown ID — should be a no-op
+        turn.record_tool_result_for("unknown_id", serde_json::json!("lost"));
+        assert_eq!(
+            turn.tool_calls[0].result.as_ref().unwrap(),
+            &serde_json::json!("done")
+        );
+    }
 }

src/agent/session_manager.rs

@@ -102,11 +102,30 @@ impl SessionManager {
     /// Resolve an external thread ID to an internal thread.
     ///
     /// Returns the session and thread ID. Creates both if they don't exist.
+    /// Delegates to [`resolve_thread_with_parsed_uuid`](Self::resolve_thread_with_parsed_uuid)
+    /// with `parsed_uuid: None`.
     pub async fn resolve_thread(
         &self,
         user_id: &str,
         channel: &str,
         external_thread_id: Option<&str>,
+    ) -> (Arc<Mutex<Session>>, Uuid) {
+        self.resolve_thread_with_parsed_uuid(user_id, channel, external_thread_id, None)
+            .await
+    }
+
+    /// Like [`resolve_thread`](Self::resolve_thread), but accepts a pre-parsed
+    /// UUID to skip redundant parsing when the caller has already validated
+    /// the external thread ID as a UUID (e.g. the approval routing path).
+    ///
+    /// Uses a single read-lock acquisition for both the key lookup and the UUID
+    /// adoption check to reduce contention under concurrent approval load.
+    pub async fn resolve_thread_with_parsed_uuid(
+        &self,
+        user_id: &str,
+        channel: &str,
+        external_thread_id: Option<&str>,
+        parsed_uuid: Option<Uuid>,
     ) -> (Arc<Mutex<Session>>, Uuid) {
         let session = self.get_or_create_session(user_id).await;
 
@@ -116,51 +135,65 @@ impl SessionManager {
             external_thread_id: external_thread_id.map(String::from),
         };
 
-        // Check if we have a mapping
+        // Use pre-parsed UUID if available, otherwise parse from string.
-        {
+        let ext_uuid = parsed_uuid
+            .or_else(|| external_thread_id.and_then(|ext_tid| Uuid::parse_str(ext_tid).ok()));
+
+        // Validate that parsed_uuid (if provided) is consistent with external_thread_id.
+        #[cfg(debug_assertions)]
+        if let (Some(parsed), Some(ext_tid)) = (&parsed_uuid, external_thread_id) {
+            debug_assert_eq!(
+                Uuid::parse_str(ext_tid).ok().as_ref(),
+                Some(parsed),
+                "parsed_uuid must be the parsed form of external_thread_id"
+            );
+        }
+
+        // Single read lock for both the key lookup and UUID adoption check
+        let adoptable_uuid = {
             let thread_map = self.thread_map.read().await;
+
+            // Fast path: exact key match
             if let Some(&thread_id) = thread_map.get(&key) {
-                // Verify thread still exists in session
                 let sess = session.lock().await;
                 if sess.threads.contains_key(&thread_id) {
                     return (Arc::clone(&session), thread_id);
                 }
             }
-        }
 
-        // Check if external_thread_id is itself a known thread UUID that
+            // UUID adoption check (still under the same read lock).
-        // exists in the session but was never registered in the thread_map
+            // If external_thread_id is a valid UUID not mapped elsewhere,
-        // (e.g. created by chat_new_thread_handler or hydrated from DB).
+            // it may be a thread created by chat_new_thread_handler or
-        // We only adopt it if no thread_map entry maps to this UUID —
+            // hydrated from DB that we can adopt.
-        // otherwise it belongs to a different channel scope.
+            // Only attempt adoption when external_thread_id is Some, preserving
-        if let Some(ext_tid) = external_thread_id
+            // the invariant that None external_thread_id never triggers adoption.
-            && let Ok(ext_uuid) = Uuid::parse_str(ext_tid)
+            if external_thread_id.is_some() {
-        {
+                ext_uuid.filter(|&uuid| !thread_map.values().any(|&v| v == uuid))
-            let thread_map = self.thread_map.read().await;
+            } else {
-            let mapped_elsewhere = thread_map.values().any(|&v| v == ext_uuid);
+                None
-            drop(thread_map);
+            }
+        }; // Single read lock dropped here
 
-            if !mapped_elsewhere {
+        // If we found an adoptable UUID, verify it exists in session and acquire write lock
-                let sess = session.lock().await;
+        if let Some(ext_uuid) = adoptable_uuid {
-                if sess.threads.contains_key(&ext_uuid) {
+            let sess = session.lock().await;
-                    drop(sess);
+            if sess.threads.contains_key(&ext_uuid) {
+                drop(sess);
 
-                    let mut thread_map = self.thread_map.write().await;
+                let mut thread_map = self.thread_map.write().await;
-                    // Re-check after acquiring write lock to prevent race condition
+                // Re-check after acquiring write lock to prevent race condition
-                    // where another task mapped this UUID between our read and write.
+                // where another task mapped this UUID between our read and write.
-                    if !thread_map.values().any(|&v| v == ext_uuid) {
+                if !thread_map.values().any(|&v| v == ext_uuid) {
-                        thread_map.insert(key, ext_uuid);
+                    thread_map.insert(key, ext_uuid);
-                        drop(thread_map);
+                    drop(thread_map);
-                        // Ensure undo manager exists
+                    // Ensure undo manager exists
-                        let mut undo_managers = self.undo_managers.write().await;
+                    let mut undo_managers = self.undo_managers.write().await;
-                        undo_managers
+                    undo_managers
-                            .entry(ext_uuid)
+                        .entry(ext_uuid)
-                            .or_insert_with(|| Arc::new(Mutex::new(UndoManager::new())));
+                        .or_insert_with(|| Arc::new(Mutex::new(UndoManager::new())));
-                        return (session, ext_uuid);
+                    return (session, ext_uuid);
-                    }
-                    // If it was mapped elsewhere while we were unlocked, fall through
-                    // to create a new thread, preserving channel isolation.
                 }
+                // If mapped elsewhere while unlocked, fall through to create new thread
             }
         }
 
@@ -909,6 +942,44 @@ mod tests {
         }
     }
 
+    #[tokio::test]
+    async fn test_resolve_thread_consolidates_read_path() {
+        // Verify that resolve_thread still correctly handles:
+        // 1. Fast path: key exists in thread_map
+        // 2. UUID adoption: external_thread_id is a UUID in session but not in map
+        // 3. New thread: neither path matches
+        use crate::agent::session::Thread;
+
+        let manager = SessionManager::new();
+
+        // Case 1: Normal resolution creates thread and maps it
+        let (session1, tid1) = manager
+            .resolve_thread("user1", "chan1", Some("ext-1"))
+            .await;
+        // Resolving again with same key should return same thread (fast path)
+        let (_, tid1_again) = manager
+            .resolve_thread("user1", "chan1", Some("ext-1"))
+            .await;
+        assert_eq!(tid1, tid1_again);
+
+        // Case 2: UUID adoption - insert a thread directly into session
+        let adopted_id = Uuid::new_v4();
+        {
+            let mut sess = session1.lock().await;
+            let thread = Thread::with_id(adopted_id, sess.id);
+            sess.threads.insert(adopted_id, thread);
+        }
+        // Resolve with the UUID as external_thread_id -- should adopt it
+        let (_, resolved) = manager
+            .resolve_thread("user1", "chan1", Some(&adopted_id.to_string()))
+            .await;
+        assert_eq!(resolved, adopted_id);
+
+        // Case 3: Different channel gets different thread
+        let (_, tid2) = manager.resolve_thread("user1", "chan2", None).await;
+        assert_ne!(tid1, tid2);
+    }
+
     #[tokio::test]
     async fn test_resolve_thread_finds_existing_session_thread_by_uuid() {
         use crate::agent::session::{Session, Thread};
@@ -947,4 +1018,88 @@ mod tests {
             "should have exactly 1 thread, not a duplicate"
         );
     }
+
+    #[tokio::test]
+    async fn test_resolve_thread_with_pre_parsed_uuid_adopts_thread() {
+        use crate::agent::session::Thread;
+
+        let manager = SessionManager::new();
+        let (session, _) = manager.resolve_thread("user1", "chan1", None).await;
+
+        // Manually insert a thread with a known UUID
+        let known_id = Uuid::new_v4();
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(known_id, sess.id);
+            sess.threads.insert(known_id, thread);
+        }
+
+        // Resolve with pre-parsed UUID -- should adopt it without re-parsing
+        let (_, resolved) = manager
+            .resolve_thread_with_parsed_uuid(
+                "user1",
+                "chan1",
+                Some(&known_id.to_string()),
+                Some(known_id),
+            )
+            .await;
+        assert_eq!(resolved, known_id);
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_with_parsed_uuid_none_delegates_to_parse() {
+        use crate::agent::session::Thread;
+
+        let manager = SessionManager::new();
+        let (session, _) = manager.resolve_thread("user2", "chan2", None).await;
+
+        // Insert a thread with a known UUID
+        let known_id = Uuid::new_v4();
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(known_id, sess.id);
+            sess.threads.insert(known_id, thread);
+        }
+
+        // Resolve with parsed_uuid=None but a valid UUID string -- should
+        // fall back to parsing the string and still adopt the thread
+        let (_, resolved) = manager
+            .resolve_thread_with_parsed_uuid("user2", "chan2", Some(&known_id.to_string()), None)
+            .await;
+        assert_eq!(resolved, known_id);
+    }
+
+    #[tokio::test]
+    async fn test_resolve_thread_with_none_external_thread_id_does_not_adopt() {
+        use crate::agent::session::Thread;
+
+        let manager = SessionManager::new();
+        let (session, default_tid) = manager.resolve_thread("user3", "chan3", None).await;
+
+        // Manually insert a thread with a known UUID (simulating a thread
+        // created by chat_new_thread_handler)
+        let known_id = Uuid::new_v4();
+        {
+            let mut sess = session.lock().await;
+            let thread = Thread::with_id(known_id, sess.id);
+            sess.threads.insert(known_id, thread);
+        }
+
+        // Resolve with external_thread_id=None but parsed_uuid=Some.
+        // This should NOT adopt the UUID — the old code prevented adoption
+        // when external_thread_id was None, and we preserve that invariant.
+        let (_, resolved) = manager
+            .resolve_thread_with_parsed_uuid("user3", "chan3", None, Some(known_id))
+            .await;
+
+        // Should return the existing default thread, not the injected UUID
+        assert_eq!(
+            resolved, default_tid,
+            "should return existing default thread when external_thread_id is None"
+        );
+        assert_ne!(
+            resolved, known_id,
+            "should NOT adopt UUID when external_thread_id is None"
+        );
+    }
 }

src/agent/submission.rs

@@ -92,6 +92,17 @@ impl SubmissionParser {
                 args: vec![],
             };
         }
+        if lower == "/reasoning" || lower.starts_with("/reasoning ") {
+            let args: Vec<String> = trimmed
+                .split_whitespace()
+                .skip(1)
+                .map(|s| s.to_string())
+                .collect();
+            return Submission::SystemCommand {
+                command: "reasoning".to_string(),
+                args,
+            };
+        }
         if lower == "/restart" {
             tracing::debug!("[SubmissionParser::parse] Recognized /restart command");
             return Submission::SystemCommand {

src/agent/thread_ops.rs

@@ -16,12 +16,12 @@ use crate::agent::dispatcher::{
 };
 use crate::agent::session::{MAX_PENDING_MESSAGES, PendingApproval, Session, ThreadState};
 use crate::agent::submission::SubmissionResult;
-use crate::channels::web::util::truncate_preview;
 use crate::channels::{IncomingMessage, StatusUpdate};
 use crate::context::JobContext;
 use crate::error::Error;
 use crate::llm::{ChatMessage, ToolCall};
 use crate::tools::redact_params;
+use ironclaw_common::truncate_preview;
 
 const FORGED_THREAD_ID_ERROR: &str = "Invalid or unauthorized thread ID.";
 
@@ -175,6 +175,7 @@ impl Agent {
     pub(super) async fn process_user_input(
         &self,
         message: &IncomingMessage,
+        tenant: crate::tenant::TenantCtx,
         session: Arc<Mutex<Session>>,
         thread_id: Uuid,
         content: &str,
@@ -351,7 +352,7 @@ impl Agent {
 
         if let Some(intent) = self.router.route_command(&temp_message) {
             // Explicit command like /status, /job, /list - handle directly
-            return self.handle_job_or_command(intent, message).await;
+            return self.handle_job_or_command(intent, message, &tenant).await;
         }
 
         // Natural language goes through the agentic loop
@@ -462,7 +463,7 @@ impl Agent {
 
         // Run the agentic tool execution loop
         let result = self
-            .run_agentic_loop(message, session.clone(), thread_id, turn_messages)
+            .run_agentic_loop(message, tenant, session.clone(), thread_id, turn_messages)
             .await;
 
         // Re-acquire lock and check if interrupted
@@ -513,10 +514,10 @@ impl Agent {
                 };
 
                 thread.complete_turn(&response);
-                let (turn_number, tool_calls) = thread
+                let (turn_number, tool_calls, narrative) = thread
                     .turns
                     .last()
-                    .map(|t| (t.turn_number, t.tool_calls.clone()))
+                    .map(|t| (t.turn_number, t.tool_calls.clone(), t.narrative.clone()))
                     .unwrap_or_default();
                 let _ = self
                     .channels
@@ -534,6 +535,7 @@ impl Agent {
                     &message.user_id,
                     turn_number,
                     &tool_calls,
+                    narrative.as_deref(),
                 )
                 .await;
                 self.persist_assistant_response(
@@ -725,7 +727,9 @@ impl Agent {
     ///
     /// Stored between the user and assistant messages so that
     /// `build_turns_from_db_messages` can reconstruct the tool call history.
-    /// Content is a JSON array of tool call summaries.
+    /// Content is a JSON object: `{ "calls": [...], "narrative": "..." }`.
+    /// The `calls` array contains tool call summaries with optional `rationale`
+    /// and `tool_call_id` fields. Legacy rows may be plain JSON arrays.
     pub(super) async fn persist_tool_calls(
         &self,
         thread_id: Uuid,
@@ -733,6 +737,7 @@ impl Agent {
         user_id: &str,
         turn_number: usize,
         tool_calls: &[crate::agent::session::TurnToolCall],
+        narrative: Option<&str>,
     ) {
         if tool_calls.is_empty() {
             return;
@@ -767,11 +772,30 @@ impl Agent {
                 if let Some(ref error) = tc.error {
                     obj["error"] = serde_json::Value::String(truncate_preview(error, 200));
                 }
+                if let Some(ref rationale) = tc.rationale {
+                    obj["rationale"] = serde_json::Value::String(truncate_preview(rationale, 500));
+                }
+                if let Some(ref tool_call_id) = tc.tool_call_id {
+                    obj["tool_call_id"] =
+                        serde_json::Value::String(truncate_preview(tool_call_id, 128));
+                }
                 obj
             })
             .collect();
 
-        let content = match serde_json::to_string(&summaries) {
+        // Wrap in an object with optional narrative so it can be reconstructed.
+        // safety: no byte-index slicing here; comment describes JSON shape
+        let wrapper = if let Some(n) = narrative {
+            serde_json::json!({
+                "narrative": truncate_preview(n, 1000),
+                "calls": summaries,
+            })
+        } else {
+            serde_json::json!({
+                "calls": summaries,
+            })
+        };
+        let content = match serde_json::to_string(&wrapper) {
             Ok(c) => c,
             Err(e) => {
                 tracing::warn!("Failed to serialize tool calls: {}", e);
@@ -1104,9 +1128,12 @@ impl Agent {
                     && let Some(turn) = thread.last_turn_mut()
                 {
                     if is_tool_error {
-                        turn.record_tool_error(result_content.clone());
+                        turn.record_tool_error_for(&pending.tool_call_id, result_content.clone());
                     } else {
-                        turn.record_tool_result(serde_json::json!(result_content));
+                        turn.record_tool_result_for(
+                            &pending.tool_call_id,
+                            serde_json::json!(result_content),
+                        );
                     }
                 }
             }
@@ -1358,9 +1385,12 @@ impl Agent {
                         && let Some(turn) = thread.last_turn_mut()
                     {
                         if is_deferred_error {
-                            turn.record_tool_error(deferred_content.clone());
+                            turn.record_tool_error_for(&tc.id, deferred_content.clone());
                         } else {
-                            turn.record_tool_result(serde_json::json!(deferred_content));
+                            turn.record_tool_result_for(
+                                &tc.id,
+                                serde_json::json!(deferred_content),
+                            );
                         }
                     }
                 }
@@ -1444,7 +1474,13 @@ impl Agent {
 
             // Continue the agentic loop (a tool was already executed this turn)
             let result = self
-                .run_agentic_loop(message, session.clone(), thread_id, context_messages)
+                .run_agentic_loop(
+                    message,
+                    self.tenant_ctx(&message.user_id).await,
+                    session.clone(),
+                    thread_id,
+                    context_messages,
+                )
                 .await;
 
             // Handle the result
@@ -1459,10 +1495,10 @@ impl Agent {
                     let (response, suggestions) =
                         crate::agent::dispatcher::extract_suggestions(&response);
                     thread.complete_turn(&response);
-                    let (turn_number, tool_calls) = thread
+                    let (turn_number, tool_calls, narrative) = thread
                         .turns
                         .last()
-                        .map(|t| (t.turn_number, t.tool_calls.clone()))
+                        .map(|t| (t.turn_number, t.tool_calls.clone(), t.narrative.clone()))
                         .unwrap_or_default();
                     // User message already persisted at turn start; save tool calls then assistant response
                     self.persist_tool_calls(
@@ -1471,6 +1507,7 @@ impl Agent {
                         &message.user_id,
                         turn_number,
                         &tool_calls,
+                        narrative.as_deref(),
                     )
                     .await;
                     self.persist_assistant_response(
@@ -1816,7 +1853,20 @@ fn rebuild_chat_messages_from_db(
             "assistant" => result.push(ChatMessage::assistant(&msg.content)),
             "tool_calls" => {
                 // Try to parse the enriched JSON and rebuild tool messages.
-                if let Ok(calls) = serde_json::from_str::<Vec<serde_json::Value>>(&msg.content) {
+                // Supports two formats:
+                // - Old: plain JSON array of tool call summaries
+                // - New: wrapped object { "calls": [...], "narrative": "..." }
+                let calls: Vec<serde_json::Value> =
+                    match serde_json::from_str::<serde_json::Value>(&msg.content) {
+                        Ok(serde_json::Value::Array(arr)) => arr,
+                        Ok(serde_json::Value::Object(obj)) => obj
+                            .get("calls")
+                            .and_then(|v| v.as_array())
+                            .cloned()
+                            .unwrap_or_default(),
+                        _ => Vec::new(),
+                    };
+                {
                     if calls.is_empty() {
                         continue;
                     }
@@ -1839,6 +1889,10 @@ fn rebuild_chat_messages_from_db(
                                     .get("parameters")
                                     .cloned()
                                     .unwrap_or(serde_json::json!({})),
+                                reasoning: c
+                                    .get("rationale")
+                                    .and_then(|v| v.as_str())
+                                    .map(String::from),
                             })
                             .collect();
 

src/app.rs

@@ -312,13 +312,7 @@ impl AppBuilder {
             .create_provider(&self.config.llm.nearai.base_url, self.session.clone());
 
         // Register memory tools if database is available
-        let workspace_user_id = self
+        let workspace_user_id = self.config.owner_id.as_str();
-            .config
-            .channels
-            .gateway
-            .as_ref()
-            .map(|gw| gw.user_id.as_str())
-            .unwrap_or("default");
         let workspace = if let Some(ref db) = self.db {
             let emb_cache_config = EmbeddingCacheConfig {
                 max_entries: self.config.embeddings.cache_size,
@@ -344,8 +338,9 @@ impl AppBuilder {
 
             // Detect multi-tenant mode: when GATEWAY_USER_TOKENS is configured,
             // each authenticated user needs their own workspace scope. Use
-            // PerUserWorkspaceResolver to create per-user workspaces on demand
+            // WorkspacePool (which implements WorkspaceResolver) to create
-            // instead of sharing the startup workspace across all users.
+            // per-user workspaces on demand instead of sharing the startup
+            // workspace across all users.
             let is_multi_tenant = self
                 .config
                 .channels
@@ -354,16 +349,14 @@ impl AppBuilder {
                 .is_some_and(|gw| gw.user_tokens.is_some());
 
             if is_multi_tenant {
-                let resolver = Arc::new(
+                let pool = Arc::new(crate::channels::web::server::WorkspacePool::new(
-                    crate::tools::builtin::memory::PerUserWorkspaceResolver::new(
+                    Arc::clone(db),
-                        Arc::clone(db),
+                    embeddings.clone(),
-                        embeddings.clone(),
+                    emb_cache_config,
-                        emb_cache_config,
+                    self.config.search.clone(),
-                        self.config.search.clone(),
+                    self.config.workspace.clone(),
-                        self.config.workspace.clone(),
+                ));
-                    ),
+                tools.register_memory_tools_with_resolver(pool);
-                );
-                tools.register_memory_tools_with_resolver(resolver);
                 tracing::info!(
                     "Memory tools configured with per-user workspace resolver (multi-tenant mode)"
                 );
@@ -887,6 +880,7 @@ impl AppBuilder {
             crate::agent::cost_guard::CostGuardConfig {
                 max_cost_per_day_cents: self.config.agent.max_cost_per_day_cents,
                 max_actions_per_hour: self.config.agent.max_actions_per_hour,
+                max_cost_per_user_per_day_cents: self.config.agent.max_cost_per_user_per_day_cents,
             },
         ));
 

src/channels/channel.rs

@@ -265,6 +265,15 @@ impl OutgoingResponse {
     }
 }
 
+/// A single tool decision within a reasoning update.
+#[derive(Debug, Clone)]
+pub struct ToolDecision {
+    /// Tool name.
+    pub tool_name: String,
+    /// Agent's reasoning for choosing this tool.
+    pub rationale: String,
+}
+
 /// Status update types for showing agent activity.
 #[derive(Debug, Clone)]
 pub enum StatusUpdate {
@@ -333,6 +342,13 @@ pub enum StatusUpdate {
     },
     /// Suggested follow-up messages for the user.
     Suggestions { suggestions: Vec<String> },
+    /// Agent reasoning update (why it chose specific tools).
+    ReasoningUpdate {
+        /// Human-readable summary of the agent's decision.
+        narrative: String,
+        /// Per-tool decisions.
+        decisions: Vec<ToolDecision>,
+    },
     /// Per-turn token usage and cost summary (shown as subtle metadata).
     TurnCost {
         input_tokens: u64,

src/channels/mod.rs

@@ -39,7 +39,7 @@ mod webhook_server;
 
 pub use channel::{
     AttachmentKind, Channel, ChannelSecretUpdater, IncomingAttachment, IncomingMessage,
-    MessageStream, OutgoingResponse, StatusUpdate, routing_target_from_metadata,
+    MessageStream, OutgoingResponse, StatusUpdate, ToolDecision, routing_target_from_metadata,
 };
 pub use http::{HttpChannel, HttpChannelState};
 pub use manager::ChannelManager;

src/channels/relay/client.rs

@@ -122,18 +122,32 @@ impl RelayClient {
     /// instance_url in chat-api. IronClaw only passes an optional CSRF nonce
     /// for validating the callback — no URLs.
     pub async fn initiate_oauth(&self, state_nonce: Option<&str>) -> Result<String, RelayError> {
+        let url = format!("{}/oauth/slack/auth", self.base_url);
+        tracing::debug!(relay_url = %url, "RelayClient::initiate_oauth: sending request");
         let mut query: Vec<(&str, &str)> = vec![];
         if let Some(nonce) = state_nonce {
             query.push(("state_nonce", nonce));
         }
         let resp = self
             .http
-            .get(format!("{}/oauth/slack/auth", self.base_url))
+            .get(&url)
             .bearer_auth(self.api_key.expose_secret())
             .query(&query)
             .send()
             .await
-            .map_err(|e| RelayError::Network(e.to_string()))?;
+            .map_err(|e| {
+                tracing::warn!(
+                    relay_url = %url,
+                    error = %e,
+                    "RelayClient::initiate_oauth: network request failed"
+                );
+                RelayError::Network(e.to_string())
+            })?;
+        tracing::debug!(
+            relay_url = %url,
+            status = %resp.status(),
+            "RelayClient::initiate_oauth: received response"
+        );
 
         let status = resp.status();
         if status.is_redirection() {
@@ -224,20 +238,39 @@ impl RelayClient {
         method: &str,
         body: serde_json::Value,
     ) -> Result<serde_json::Value, RelayError> {
+        let url = format!("{}/proxy/{}/{}", self.base_url, provider, method);
+        tracing::debug!(
+            relay_url = %url,
+            provider = %provider,
+            method = %method,
+            "RelayClient::proxy_provider: sending request"
+        );
         let query: Vec<(&str, &str)> = vec![("team_id", team_id)];
         let resp = self
             .http
-            .post(format!("{}/proxy/{}/{}", self.base_url, provider, method))
+            .post(&url)
             .bearer_auth(self.api_key.expose_secret())
             .query(&query)
             .json(&body)
             .send()
             .await
-            .map_err(|e| RelayError::Network(e.to_string()))?;
+            .map_err(|e| {
+                tracing::warn!(
+                    relay_url = %url,
+                    error = %e,
+                    "RelayClient::proxy_provider: network request failed"
+                );
+                RelayError::Network(e.to_string())
+            })?;
 
         if !resp.status().is_success() {
             let status = resp.status().as_u16();
             let body = resp.text().await.unwrap_or_default();
+            tracing::warn!(
+                relay_url = %url,
+                status = status,
+                "RelayClient::proxy_provider: channel-relay returned error"
+            );
             return Err(RelayError::Api {
                 status,
                 message: body,
@@ -255,23 +288,45 @@ impl RelayClient {
     /// 32-byte secret. Called once at activation time; the result is cached in the
     /// extension manager so subsequent calls to `relay_signing_secret()` use it.
     pub async fn get_signing_secret(&self, team_id: &str) -> Result<Vec<u8>, RelayError> {
+        let url = format!("{}/relay/signing-secret", self.base_url);
+        tracing::debug!(
+            relay_url = %url,
+            "RelayClient::get_signing_secret: fetching signing secret"
+        );
         let resp = self
             .http
-            .get(format!("{}/relay/signing-secret", self.base_url))
+            .get(&url)
             .bearer_auth(self.api_key.expose_secret())
             .query(&[("team_id", team_id)])
             .send()
             .await
-            .map_err(|e| RelayError::Network(e.to_string()))?;
+            .map_err(|e| {
+                tracing::warn!(
+                    relay_url = %url,
+                    error = %e,
+                    "RelayClient::get_signing_secret: network request failed"
+                );
+                RelayError::Network(e.to_string())
+            })?;
 
         if !resp.status().is_success() {
             let status = resp.status().as_u16();
             let body = resp.text().await.unwrap_or_default();
+            tracing::warn!(
+                relay_url = %url,
+                status = status,
+                body = %body,
+                "RelayClient::get_signing_secret: channel-relay returned error"
+            );
             return Err(RelayError::Api {
                 status,
                 message: body,
             });
         }
+        tracing::debug!(
+            relay_url = %url,
+            "RelayClient::get_signing_secret: received successful response"
+        );
 
         let body: serde_json::Value = resp
             .json()

src/channels/repl.rs

@@ -75,6 +75,7 @@ const SLASH_COMMANDS: &[&str] = &[
     "/suggest",
     "/thread",
     "/resume",
+    "/reasoning",
 ];
 
 /// Rustyline helper for slash-command tab completion.
@@ -430,6 +431,18 @@ impl ReplChannel {
             let _ = execute!(stderr, terminal::Clear(terminal::ClearType::FromCursorDown));
         }
     }
+
+    async fn finish_single_message_turn(&self) {
+        if self.single_message.is_none() {
+            return;
+        }
+
+        let tx = self.msg_tx.lock().ok().and_then(|mut guard| guard.take());
+        if let Some(tx) = tx {
+            let msg = IncomingMessage::new("repl", &self.user_id, "/quit");
+            let _ = tx.send(msg).await;
+        }
+    }
 }
 
 impl Default for ReplChannel {
@@ -479,7 +492,9 @@ impl Channel for ReplChannel {
 
     async fn start(&self) -> Result<MessageStream, ChannelError> {
         let (tx, rx) = mpsc::channel(32);
-        // Store tx so send_status can inject approval responses directly
+        // Approval prompts inject responses back through this sender.
+        // In single-message mode we keep it until the turn finishes, then
+        // drop it after enqueuing /quit so the receiver stream can close.
         if let Ok(mut guard) = self.msg_tx.lock() {
             *guard = Some(tx.clone());
         }
@@ -495,11 +510,10 @@ impl Channel for ReplChannel {
 
             // Single message mode: send it and return
             if let Some(msg) = single_message {
-                let incoming = IncomingMessage::new("repl", &user_id, &msg).with_timezone(&sys_tz);
+                let incoming = IncomingMessage::new("repl", &user_id, &msg)
+                    .with_metadata(serde_json::json!({ "single_message_mode": true }))
+                    .with_timezone(&sys_tz);
                 let _ = tx.blocking_send(incoming);
-                // Ensure the agent exits after handling exactly one turn in -m mode,
-                // even when other channels (gateway/http) are enabled.
-                let _ = tx.blocking_send(IncomingMessage::new("repl", &user_id, "/quit"));
                 return;
             }
 
@@ -662,6 +676,7 @@ impl Channel for ReplChannel {
             println!();
             println!();
             self.stdin_locked.store(false, Ordering::Relaxed);
+            self.finish_single_message_turn().await;
             return Ok(());
         }
 
@@ -680,6 +695,7 @@ impl Channel for ReplChannel {
         println!();
         // Unlock stdin so readline can resume
         self.stdin_locked.store(false, Ordering::Relaxed);
+        self.finish_single_message_turn().await;
         Ok(())
     }
 
@@ -779,6 +795,7 @@ impl Channel for ReplChannel {
                 let msg_tx = Arc::clone(&self.msg_tx);
                 let user_id = self.user_id.clone();
                 let lock_flag = Arc::clone(&self.stdin_locked);
+                let single_message_mode = self.single_message.is_some();
                 tokio::task::spawn_blocking(move || {
                     let action = run_approval_selector(allow_always).unwrap_or("n");
                     // Unlock stdin so readline can resume after approval
@@ -787,7 +804,12 @@ impl Channel for ReplChannel {
                         return;
                     };
                     if let Some(tx) = guard.as_ref() {
-                        let msg = IncomingMessage::new("repl", &user_id, action);
+                        let msg = if single_message_mode {
+                            IncomingMessage::new("repl", &user_id, action)
+                                .with_metadata(serde_json::json!({ "single_message_mode": true }))
+                        } else {
+                            IncomingMessage::new("repl", &user_id, action)
+                        };
                         let _ = tx.blocking_send(msg);
                     }
                 });
@@ -841,6 +863,19 @@ impl Channel for ReplChannel {
             StatusUpdate::Suggestions { .. } => {
                 // Suggestions are only rendered by the web gateway
             }
+            StatusUpdate::ReasoningUpdate {
+                narrative,
+                decisions,
+            } => {
+                if !narrative.is_empty() {
+                    let display = truncate_for_preview(&narrative, CLI_STATUS_MAX);
+                    eprintln!("  \x1b[94m\u{25B6} {display}\x1b[0m");
+                }
+                for d in &decisions {
+                    let display = truncate_for_preview(&d.rationale, CLI_STATUS_MAX);
+                    eprintln!("    \x1b[90m\u{2192} {}: {display}\x1b[0m", d.tool_name);
+                }
+            }
             StatusUpdate::TurnCost { .. } => {
                 // Cost display is handled by the TUI channel
             }
@@ -875,6 +910,7 @@ impl Channel for ReplChannel {
 #[cfg(test)]
 mod tests {
     use futures::StreamExt;
+    use tokio::time::{Duration, timeout};
 
     use super::*;
 
@@ -883,16 +919,36 @@ mod tests {
         let repl = ReplChannel::with_message("hi".to_string());
         let mut stream = repl.start().await.expect("repl start should succeed");
 
-        let first = stream.next().await.expect("first message missing");
+        let first = timeout(Duration::from_secs(1), stream.next())
+            .await
+            .expect("timed out waiting for first message")
+            .expect("first message missing");
         assert_eq!(first.channel, "repl");
         assert_eq!(first.content, "hi");
 
-        let second = stream.next().await.expect("quit message missing");
+        assert!(
+            timeout(Duration::from_millis(100), stream.next())
+                .await
+                .is_err(),
+            "single-message mode should wait for the turn to finish before quitting"
+        );
+
+        repl.respond(&first, OutgoingResponse::text("done"))
+            .await
+            .expect("respond should succeed");
+
+        let second = timeout(Duration::from_secs(1), stream.next())
+            .await
+            .expect("timed out waiting for quit message")
+            .expect("quit message missing");
         assert_eq!(second.channel, "repl");
         assert_eq!(second.content, "/quit");
 
         assert!(
-            stream.next().await.is_none(),
+            timeout(Duration::from_secs(1), stream.next())
+                .await
+                .expect("timed out waiting for stream to close")
+                .is_none(),
             "stream should end after /quit"
         );
     }

src/channels/wasm/router.rs

@@ -333,6 +333,9 @@ async fn webhook_handler(
 
     let channel_name = channel.channel_name();
 
+    // Track whether any authentication was performed and passed.
+    let mut did_authenticate = false;
+
     // Check if secret is required
     if state.router.requires_secret(channel_name).await {
         // Get the secret header name for this channel (from capabilities or default)
@@ -382,6 +385,7 @@ async fn webhook_handler(
                     );
                 }
                 tracing::debug!(channel = %channel_name, "Webhook secret validated");
+                did_authenticate = true;
             }
             None => {
                 tracing::warn!(
@@ -433,6 +437,7 @@ async fn webhook_handler(
                     );
                 }
                 tracing::debug!(channel = %channel_name, "Ed25519 signature verified");
+                did_authenticate = true;
             }
             _ => {
                 tracing::warn!(
@@ -484,6 +489,7 @@ async fn webhook_handler(
                     );
                 }
                 tracing::debug!(channel = %channel_name, "HMAC-SHA256 signature verified");
+                did_authenticate = true;
             }
             _ => {
                 tracing::warn!(
@@ -510,8 +516,9 @@ async fn webhook_handler(
         })
         .collect();
 
-    // Call the WASM channel
+    // Call the WASM channel. `did_authenticate` was set above by whichever
-    let secret_validated = state.router.requires_secret(channel_name).await;
+    // auth guard (secret / Ed25519 / HMAC) successfully validated the request.
+    let secret_validated = did_authenticate;
 
     tracing::info!(
         channel = %channel_name,

src/channels/wasm/wrapper.rs

@@ -3061,6 +3061,20 @@ fn status_to_wit(
         },
         // Suggestions and turn cost are web-gateway-only; skip for WASM channels
         StatusUpdate::Suggestions { .. } | StatusUpdate::TurnCost { .. } => return None,
+        StatusUpdate::ReasoningUpdate {
+            narrative,
+            decisions,
+        } => {
+            let mut msg = narrative.clone();
+            for d in decisions {
+                msg.push_str(&format!("\n  → {}: {}", d.tool_name, d.rationale));
+            }
+            wit_channel::StatusUpdate {
+                status: wit_channel::StatusType::Status,
+                message: msg,
+                metadata_json,
+            }
+        }
     })
 }
 

src/channels/web/handlers/chat.rs

@@ -175,7 +175,7 @@ pub async fn chat_auth_token_handler(
             if result.verification.is_some() {
                 state.sse.broadcast_for_user(
                     &user.user_id,
-                    SseEvent::AuthRequired {
+                    AppEvent::AuthRequired {
                         extension_name: req.extension_name.clone(),
                         instructions: Some(result.message),
                         auth_url: None,
@@ -187,7 +187,7 @@ pub async fn chat_auth_token_handler(
 
                 state.sse.broadcast_for_user(
                     &user.user_id,
-                    SseEvent::AuthCompleted {
+                    AppEvent::AuthCompleted {
                         extension_name: req.extension_name.clone(),
                         success: true,
                         message: result.message,
@@ -202,7 +202,7 @@ pub async fn chat_auth_token_handler(
             if matches!(e, crate::extensions::ExtensionError::ValidationFailed(_)) {
                 state.sse.broadcast_for_user(
                     &user.user_id,
-                    SseEvent::AuthRequired {
+                    AppEvent::AuthRequired {
                         extension_name: req.extension_name.clone(),
                         instructions: Some(msg.clone()),
                         auth_url: None,
@@ -398,8 +398,10 @@ pub async fn chat_history_handler(
                                 truncate_preview(&s, 500)
                             }),
                             error: tc.error.clone(),
+                            rationale: tc.rationale.clone(),
                         })
                         .collect(),
+                    narrative: t.narrative.clone(),
                 })
                 .collect();
 

src/channels/web/handlers/jobs.rs

@@ -598,11 +598,23 @@ pub async fn jobs_prompt_handler(
     }
 
     // Try agent job path: verify ownership, then send via scheduler.
-    if let Some(ref store) = state.store
+    if let Some(ref store) = state.store {
-        && let Ok(Some(agent_job)) = store.get_job(job_id).await
+        match store.get_job(job_id).await {
-        && agent_job.user_id != user.user_id
+            Ok(Some(agent_job)) => {
-    {
+                if agent_job.user_id != user.user_id {
-        return Err((StatusCode::NOT_FOUND, "Job not found".to_string()));
+                    return Err((StatusCode::NOT_FOUND, "Job not found".to_string()));
+                }
+            }
+            Ok(None) => {
+                return Err((StatusCode::NOT_FOUND, "Job not found".to_string()));
+            }
+            Err(e) => {
+                return Err((
+                    StatusCode::INTERNAL_SERVER_ERROR,
+                    format!("Database error: {}", e),
+                ));
+            }
+        }
     }
 
     let slot = state.scheduler.as_ref().ok_or((

src/channels/web/handlers/memory.rs

@@ -0,0 +1,225 @@
+//! Memory/workspace API handlers.
+
+use std::sync::Arc;
+
+use axum::{
+    Json,
+    extract::{Query, State},
+    http::StatusCode,
+};
+use serde::Deserialize;
+
+use crate::channels::web::auth::{AuthenticatedUser, UserIdentity};
+use crate::channels::web::server::GatewayState;
+use crate::channels::web::types::*;
+use crate::workspace::Workspace;
+
+/// Resolve the workspace for the authenticated user.
+///
+/// Prefers `workspace_pool` (multi-user mode) when available, falling back
+/// to the single-user `state.workspace`.
+pub(crate) async fn resolve_workspace(
+    state: &GatewayState,
+    user: &UserIdentity,
+) -> Result<Arc<Workspace>, (StatusCode, String)> {
+    if let Some(ref pool) = state.workspace_pool {
+        return Ok(pool.get_or_create(user).await);
+    }
+    state.workspace.as_ref().cloned().ok_or((
+        StatusCode::SERVICE_UNAVAILABLE,
+        "Workspace not available".to_string(),
+    ))
+}
+
+#[derive(Deserialize)]
+pub struct TreeQuery {
+    #[allow(dead_code)]
+    pub depth: Option<usize>,
+}
+
+pub async fn memory_tree_handler(
+    State(state): State<Arc<GatewayState>>,
+    AuthenticatedUser(user): AuthenticatedUser,
+    Query(_query): Query<TreeQuery>,
+) -> Result<Json<MemoryTreeResponse>, (StatusCode, String)> {
+    let workspace = resolve_workspace(&state, &user).await?;
+
+    // Build tree from list_all (flat list of all paths)
+    let all_paths = workspace
+        .list_all()
+        .await
+        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+
+    // Collect unique directories and files
+    let mut entries: Vec<TreeEntry> = Vec::new();
+    let mut seen_dirs: std::collections::HashSet<String> = std::collections::HashSet::new();
+
+    for path in &all_paths {
+        // Add parent directories
+        let parts: Vec<&str> = path.split('/').collect();
+        for i in 0..parts.len().saturating_sub(1) {
+            let dir_path = parts[..=i].join("/");
+            if seen_dirs.insert(dir_path.clone()) {
+                entries.push(TreeEntry {
+                    path: dir_path,
+                    is_dir: true,
+                });
+            }
+        }
+        // Add the file itself
+        entries.push(TreeEntry {
+            path: path.clone(),
+            is_dir: false,
+        });
+    }
+
+    entries.sort_by(|a, b| a.path.cmp(&b.path));
+
+    Ok(Json(MemoryTreeResponse { entries }))
+}
+
+#[derive(Deserialize)]
+pub struct ListQuery {
+    pub path: Option<String>,
+}
+
+pub async fn memory_list_handler(
+    State(state): State<Arc<GatewayState>>,
+    AuthenticatedUser(user): AuthenticatedUser,
+    Query(query): Query<ListQuery>,
+) -> Result<Json<MemoryListResponse>, (StatusCode, String)> {
+    let workspace = resolve_workspace(&state, &user).await?;
+
+    let path = query.path.as_deref().unwrap_or("");
+    let entries = workspace
+        .list(path)
+        .await
+        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+
+    let list_entries: Vec<ListEntry> = entries
+        .iter()
+        .map(|e| ListEntry {
+            name: e.path.rsplit('/').next().unwrap_or(&e.path).to_string(),
+            path: e.path.clone(),
+            is_dir: e.is_directory,
+            updated_at: e.updated_at.map(|dt| dt.to_rfc3339()),
+        })
+        .collect();
+
+    Ok(Json(MemoryListResponse {
+        path: path.to_string(),
+        entries: list_entries,
+    }))
+}
+
+#[derive(Deserialize)]
+pub struct ReadQuery {
+    pub path: String,
+}
+
+pub async fn memory_read_handler(
+    State(state): State<Arc<GatewayState>>,
+    AuthenticatedUser(user): AuthenticatedUser,
+    Query(query): Query<ReadQuery>,
+) -> Result<Json<MemoryReadResponse>, (StatusCode, String)> {
+    let workspace = resolve_workspace(&state, &user).await?;
+
+    let doc = workspace
+        .read(&query.path)
+        .await
+        .map_err(|e| (StatusCode::NOT_FOUND, e.to_string()))?;
+
+    Ok(Json(MemoryReadResponse {
+        path: query.path,
+        content: doc.content,
+        updated_at: Some(doc.updated_at.to_rfc3339()),
+    }))
+}
+
+pub async fn memory_write_handler(
+    State(state): State<Arc<GatewayState>>,
+    AuthenticatedUser(user): AuthenticatedUser,
+    Json(req): Json<MemoryWriteRequest>,
+) -> Result<Json<MemoryWriteResponse>, (StatusCode, String)> {
+    let workspace = resolve_workspace(&state, &user).await?;
+
+    // Route through layer-aware methods when a layer is specified.
+    //
+    // Note: unlike MemoryWriteTool, this endpoint does NOT block writes to
+    // identity files (IDENTITY.md, SOUL.md, etc.). The HTTP API is an
+    // authenticated admin interface; the supervisor uses it to seed identity
+    // files at startup. Identity-file protection is enforced at the tool
+    // layer (LLM-facing) where the write originates from an untrusted agent.
+    if let Some(ref layer_name) = req.layer {
+        let result = if req.append {
+            workspace
+                .append_to_layer(layer_name, &req.path, &req.content, req.force)
+                .await
+        } else {
+            workspace
+                .write_to_layer(layer_name, &req.path, &req.content, req.force)
+                .await
+        }
+        .map_err(|e| {
+            use crate::error::WorkspaceError;
+            let status = match &e {
+                WorkspaceError::LayerNotFound { .. } => StatusCode::BAD_REQUEST,
+                WorkspaceError::LayerReadOnly { .. } => StatusCode::FORBIDDEN,
+                WorkspaceError::PrivacyRedirectFailed => StatusCode::UNPROCESSABLE_ENTITY,
+                _ => StatusCode::INTERNAL_SERVER_ERROR,
+            };
+            (status, e.to_string())
+        })?;
+        return Ok(Json(MemoryWriteResponse {
+            path: req.path,
+            status: "written",
+            redirected: Some(result.redirected),
+            actual_layer: Some(result.actual_layer),
+        }));
+    }
+
+    // Non-layer path: honor the append field
+    if req.append {
+        workspace
+            .append(&req.path, &req.content)
+            .await
+            .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+    } else {
+        workspace
+            .write(&req.path, &req.content)
+            .await
+            .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+    }
+
+    Ok(Json(MemoryWriteResponse {
+        path: req.path,
+        status: "written",
+        redirected: None,
+        actual_layer: None,
+    }))
+}
+
+pub async fn memory_search_handler(
+    State(state): State<Arc<GatewayState>>,
+    AuthenticatedUser(user): AuthenticatedUser,
+    Json(req): Json<MemorySearchRequest>,
+) -> Result<Json<MemorySearchResponse>, (StatusCode, String)> {
+    let workspace = resolve_workspace(&state, &user).await?;
+
+    let limit = req.limit.unwrap_or(10);
+    let results = workspace
+        .search(&req.query, limit)
+        .await
+        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+
+    let hits: Vec<SearchHit> = results
+        .iter()
+        .map(|r| SearchHit {
+            path: r.document_id.to_string(),
+            content: r.content.clone(),
+            score: r.score as f64,
+        })
+        .collect();
+
+    Ok(Json(MemorySearchResponse { results: hits }))
+}

src/channels/web/handlers/mod.rs

@@ -3,6 +3,7 @@
 //! Each module groups related endpoint handlers by domain.
 
 pub mod jobs;
+pub mod memory;
 pub mod routines;
 pub mod skills;
 

src/channels/web/handlers/routines.rs

@@ -114,7 +114,7 @@ pub async fn routines_detail_handler(
             trigger_type: run.trigger_type.clone(),
             started_at: run.started_at.to_rfc3339(),
             completed_at: run.completed_at.map(|dt| dt.to_rfc3339()),
-            status: format!("{:?}", run.status),
+            status: run.status.to_string(),
             result_summary: run.result_summary.clone(),
             tokens_used: run.tokens_used,
             job_id: run.job_id,
@@ -324,7 +324,7 @@ pub async fn routines_runs_handler(
             trigger_type: run.trigger_type.clone(),
             started_at: run.started_at.to_rfc3339(),
             completed_at: run.completed_at.map(|dt| dt.to_rfc3339()),
-            status: format!("{:?}", run.status),
+            status: run.status.to_string(),
             result_summary: run.result_summary.clone(),
             tokens_used: run.tokens_used,
             job_id: run.job_id,

src/channels/web/handlers/webhooks.rs

@@ -54,10 +54,37 @@ fn validate_webhook_secret(
 ///
 /// This endpoint is **public** (no gateway auth token required) but protected
 /// by the per-routine webhook secret sent via the `X-Webhook-Secret` header.
+///
+/// **Single-user/backward-compatible**: looks up routines by path across all
+/// users. For multi-tenant isolation, use the user-scoped endpoint at
+/// `/api/webhooks/u/{user_id}/{path}` instead.
 pub async fn webhook_trigger_handler(
     State(state): State<Arc<GatewayState>>,
     Path(path): Path<String>,
     headers: HeaderMap,
+) -> Result<Json<serde_json::Value>, (StatusCode, String)> {
+    fire_webhook_inner(state, &path, None, &headers).await
+}
+
+/// Handle incoming webhook POST to `/api/webhooks/u/{user_id}/{path}`.
+///
+/// User-scoped variant for multi-tenant deployments. The `user_id` in the URL
+/// restricts the routine lookup to that user only, preventing cross-user
+/// webhook triggering even when paths collide.
+pub async fn webhook_trigger_user_scoped_handler(
+    State(state): State<Arc<GatewayState>>,
+    Path((user_id, path)): Path<(String, String)>,
+    headers: HeaderMap,
+) -> Result<Json<serde_json::Value>, (StatusCode, String)> {
+    fire_webhook_inner(state, &path, Some(&user_id), &headers).await
+}
+
+/// Shared webhook logic for both scoped and unscoped endpoints.
+async fn fire_webhook_inner(
+    state: Arc<GatewayState>,
+    path: &str,
+    user_id: Option<&str>,
+    headers: &HeaderMap,
 ) -> Result<Json<serde_json::Value>, (StatusCode, String)> {
     // Rate limit check
     if !state.webhook_rate_limiter.check() {
@@ -72,9 +99,9 @@ pub async fn webhook_trigger_handler(
         "Database not available".to_string(),
     ))?;
 
-    // Targeted query instead of loading all routines
+    // Targeted query — when user_id is provided, restrict to that user's routines
     let routine = store
-        .get_webhook_routine_by_path(&path)
+        .get_webhook_routine_by_path(path, user_id)
         .await
         .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
         .ok_or((
@@ -99,7 +126,7 @@ pub async fn webhook_trigger_handler(
         ))?
     };
 
-    let run_id = engine.fire_webhook(routine.id, &path).await.map_err(|e| {
+    let run_id = engine.fire_webhook(routine.id, path).await.map_err(|e| {
         let status = match &e {
             crate::error::RoutineError::NotFound { .. } => StatusCode::NOT_FOUND,
             crate::error::RoutineError::Disabled { .. }

src/channels/web/mod.rs

@@ -58,7 +58,7 @@ use self::log_layer::{LogBroadcaster, LogLevelHandle};
 use self::auth::MultiAuthState;
 use self::server::GatewayState;
 use self::sse::SseManager;
-use self::types::SseEvent;
+use self::types::AppEvent;
 
 /// Web gateway channel implementing the Channel trait.
 pub struct GatewayChannel {
@@ -98,7 +98,8 @@ impl GatewayChannel {
             job_manager: None,
             prompt_queue: None,
             scheduler: None,
-            default_user_id: config.user_id.clone(),
+            owner_id: config.user_id.clone(),
+            default_sender_id: config.user_id.clone(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: Some(Arc::new(ws::WsConnectionTracker::new())),
             llm_provider: None,
@@ -121,6 +122,22 @@ impl GatewayChannel {
         }
     }
 
+    /// Rebind the single-user auth identity to the durable owner scope while
+    /// preserving the configured gateway sender/routing identity.
+    pub fn with_owner_scope(mut self, owner_id: impl Into<String>) -> Self {
+        let owner_id = owner_id.into();
+        let single_user_token = if self.config.user_tokens.is_none() {
+            self.auth.first_token().map(ToOwned::to_owned)
+        } else {
+            None
+        };
+        if let Some(token) = single_user_token {
+            self.auth = MultiAuthState::single(token, owner_id.clone());
+        }
+        self.rebuild_state(|s| s.owner_id = owner_id);
+        self
+    }
+
     /// Create a gateway channel with a pre-built multi-user auth state.
     pub fn new_multi_auth(config: GatewayConfig, auth: MultiAuthState) -> Self {
         let state = Arc::new(GatewayState {
@@ -137,7 +154,8 @@ impl GatewayChannel {
             job_manager: None,
             prompt_queue: None,
             scheduler: None,
-            default_user_id: config.user_id.clone(),
+            owner_id: config.user_id.clone(),
+            default_sender_id: config.user_id.clone(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: Some(Arc::new(ws::WsConnectionTracker::new())),
             llm_provider: None,
@@ -177,7 +195,8 @@ impl GatewayChannel {
             job_manager: self.state.job_manager.clone(),
             prompt_queue: self.state.prompt_queue.clone(),
             scheduler: self.state.scheduler.clone(),
-            default_user_id: self.state.default_user_id.clone(),
+            owner_id: self.state.owner_id.clone(),
+            default_sender_id: self.state.default_sender_id.clone(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: self.state.ws_tracker.clone(),
             llm_provider: self.state.llm_provider.clone(),
@@ -367,7 +386,7 @@ impl Channel for GatewayChannel {
 
         self.state.sse.broadcast_for_user(
             &msg.user_id,
-            SseEvent::Response {
+            AppEvent::Response {
                 content: response.content,
                 thread_id,
             },
@@ -386,11 +405,11 @@ impl Channel for GatewayChannel {
             .and_then(|v| v.as_str())
             .map(String::from);
         let event = match status {
-            StatusUpdate::Thinking(msg) => SseEvent::Thinking {
+            StatusUpdate::Thinking(msg) => AppEvent::Thinking {
                 message: msg,
                 thread_id: thread_id.clone(),
             },
-            StatusUpdate::ToolStarted { name } => SseEvent::ToolStarted {
+            StatusUpdate::ToolStarted { name } => AppEvent::ToolStarted {
                 name,
                 thread_id: thread_id.clone(),
             },
@@ -399,23 +418,23 @@ impl Channel for GatewayChannel {
                 success,
                 error,
                 parameters,
-            } => SseEvent::ToolCompleted {
+            } => AppEvent::ToolCompleted {
                 name,
                 success,
                 error,
                 parameters,
                 thread_id: thread_id.clone(),
             },
-            StatusUpdate::ToolResult { name, preview } => SseEvent::ToolResult {
+            StatusUpdate::ToolResult { name, preview } => AppEvent::ToolResult {
                 name,
                 preview,
                 thread_id: thread_id.clone(),
             },
-            StatusUpdate::StreamChunk(content) => SseEvent::StreamChunk {
+            StatusUpdate::StreamChunk(content) => AppEvent::StreamChunk {
                 content,
                 thread_id: thread_id.clone(),
             },
-            StatusUpdate::Status(msg) => SseEvent::Status {
+            StatusUpdate::Status(msg) => AppEvent::Status {
                 message: msg,
                 thread_id: thread_id.clone(),
             },
@@ -423,7 +442,7 @@ impl Channel for GatewayChannel {
                 job_id,
                 title,
                 browse_url,
-            } => SseEvent::JobStarted {
+            } => AppEvent::JobStarted {
                 job_id,
                 title,
                 browse_url,
@@ -434,7 +453,7 @@ impl Channel for GatewayChannel {
                 description,
                 parameters,
                 allow_always,
-            } => SseEvent::ApprovalNeeded {
+            } => AppEvent::ApprovalNeeded {
                 request_id,
                 tool_name,
                 description,
@@ -448,7 +467,7 @@ impl Channel for GatewayChannel {
                 instructions,
                 auth_url,
                 setup_url,
-            } => SseEvent::AuthRequired {
+            } => AppEvent::AuthRequired {
                 extension_name,
                 instructions,
                 auth_url,
@@ -458,25 +477,39 @@ impl Channel for GatewayChannel {
                 extension_name,
                 success,
                 message,
-            } => SseEvent::AuthCompleted {
+            } => AppEvent::AuthCompleted {
                 extension_name,
                 success,
                 message,
             },
-            StatusUpdate::ImageGenerated { data_url, path } => SseEvent::ImageGenerated {
+            StatusUpdate::ImageGenerated { data_url, path } => AppEvent::ImageGenerated {
                 data_url,
                 path,
                 thread_id: thread_id.clone(),
             },
-            StatusUpdate::Suggestions { suggestions } => SseEvent::Suggestions {
+            StatusUpdate::Suggestions { suggestions } => AppEvent::Suggestions {
                 suggestions,
+                thread_id: thread_id.clone(),
+            },
+            StatusUpdate::ReasoningUpdate {
+                narrative,
+                decisions,
+            } => AppEvent::ReasoningUpdate {
+                narrative,
+                decisions: decisions
+                    .into_iter()
+                    .map(|d| crate::channels::web::types::ToolDecisionDto {
+                        tool_name: d.tool_name,
+                        rationale: d.rationale,
+                    })
+                    .collect(),
                 thread_id,
             },
             StatusUpdate::TurnCost {
                 input_tokens,
                 output_tokens,
                 cost_usd,
-            } => SseEvent::TurnCost {
+            } => AppEvent::TurnCost {
                 input_tokens,
                 output_tokens,
                 cost_usd,
@@ -512,7 +545,7 @@ impl Channel for GatewayChannel {
         };
         self.state.sse.broadcast_for_user(
             user_id,
-            SseEvent::Response {
+            AppEvent::Response {
                 content: response.content,
                 thread_id,
             },

src/channels/web/openai_compat.rs

@@ -231,6 +231,7 @@ pub fn convert_messages(messages: &[OpenAiMessage]) -> Result<Vec<ChatMessage>,
                                 name: tc.function.name.clone(),
                                 arguments: serde_json::from_str(&tc.function.arguments)
                                     .unwrap_or(serde_json::Value::Object(Default::default())),
+                                reasoning: None,
                             })
                             .collect();
                         Ok(ChatMessage::assistant_with_tool_calls(
@@ -954,6 +955,7 @@ mod tests {
             id: "call_abc".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"query": "rust"}),
+            reasoning: None,
         }];
 
         let converted = convert_tool_calls_to_openai(&calls);

src/channels/web/sse.rs

@@ -11,7 +11,7 @@ use tokio::sync::broadcast;
 use tokio_stream::StreamExt;
 use tokio_stream::wrappers::BroadcastStream;
 
-use crate::channels::web::types::SseEvent;
+use crate::channels::web::types::AppEvent;
 
 /// Maximum number of concurrent SSE/WebSocket connections.
 /// Prevents resource exhaustion from connection flooding.
@@ -25,7 +25,7 @@ const MAX_CONNECTIONS: u64 = 100;
 #[derive(Debug, Clone)]
 pub(crate) struct ScopedEvent {
     pub(crate) user_id: Option<String>,
-    pub(crate) event: SseEvent,
+    pub(crate) event: AppEvent,
 }
 
 /// Manages SSE broadcast to all connected browser tabs.
@@ -75,7 +75,7 @@ impl SseManager {
     }
 
     /// Broadcast an event to all connected clients (global/unscoped).
-    pub fn broadcast(&self, event: SseEvent) {
+    pub fn broadcast(&self, event: AppEvent) {
         let _ = self.tx.send(ScopedEvent {
             user_id: None,
             event,
@@ -86,7 +86,7 @@ impl SseManager {
     ///
     /// Only subscribers for this user_id (or unscoped subscribers) will
     /// receive the event.
-    pub fn broadcast_for_user(&self, user_id: &str, event: SseEvent) {
+    pub fn broadcast_for_user(&self, user_id: &str, event: AppEvent) {
         let _ = self.tx.send(ScopedEvent {
             user_id: Some(user_id.to_string()),
             event,
@@ -108,7 +108,7 @@ impl SseManager {
     pub fn subscribe_raw(
         &self,
         user_id: Option<String>,
-    ) -> Option<impl Stream<Item = SseEvent> + Send + 'static + use<>> {
+    ) -> Option<impl Stream<Item = AppEvent> + Send + 'static + use<>> {
         // Atomically increment only if below the limit. This prevents
         // concurrent callers from overshooting max_connections.
         let counter = Arc::clone(&self.connection_count);
@@ -186,30 +186,7 @@ impl SseManager {
                         return None;
                     }
                 };
-                let event_type = match &event {
+                let event_type = event.event_type();
-                    SseEvent::Response { .. } => "response",
-                    SseEvent::Thinking { .. } => "thinking",
-                    SseEvent::ToolStarted { .. } => "tool_started",
-                    SseEvent::ToolCompleted { .. } => "tool_completed",
-                    SseEvent::ToolResult { .. } => "tool_result",
-                    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",
-                    SseEvent::ImageGenerated { .. } => "image_generated",
-                    SseEvent::Suggestions { .. } => "suggestions",
-                    SseEvent::TurnCost { .. } => "turn_cost",
-                    SseEvent::ExtensionStatus { .. } => "extension_status",
-                };
                 Some(Ok(Event::default().event(event_type).data(data)))
             });
 
@@ -272,7 +249,7 @@ mod tests {
     fn test_broadcast_without_receivers() {
         let manager = SseManager::new();
         // Should not panic even with no receivers
-        manager.broadcast(SseEvent::Heartbeat);
+        manager.broadcast(AppEvent::Heartbeat);
     }
 
     #[tokio::test]
@@ -280,14 +257,14 @@ mod tests {
         let manager = SseManager::new();
         let mut stream = Box::pin(manager.subscribe_raw(None).expect("should subscribe"));
 
-        manager.broadcast(SseEvent::Status {
+        manager.broadcast(AppEvent::Status {
             message: "test".to_string(),
             thread_id: None,
         });
 
         let event = stream.next().await.unwrap();
         match event {
-            SseEvent::Status { message, .. } => assert_eq!(message, "test"),
+            AppEvent::Status { message, .. } => assert_eq!(message, "test"),
             _ => panic!("unexpected event type"),
         }
     }
@@ -299,14 +276,14 @@ mod tests {
 
         assert_eq!(manager.connection_count(), 1);
 
-        manager.broadcast(SseEvent::Thinking {
+        manager.broadcast(AppEvent::Thinking {
             message: "working".to_string(),
             thread_id: None,
         });
 
         let event = stream.next().await.unwrap();
         match event {
-            SseEvent::Thinking { message, .. } => assert_eq!(message, "working"),
+            AppEvent::Thinking { message, .. } => assert_eq!(message, "working"),
             _ => panic!("Expected Thinking event"),
         }
     }
@@ -329,12 +306,12 @@ mod tests {
         let mut s2 = Box::pin(manager.subscribe_raw(None).expect("should subscribe"));
         assert_eq!(manager.connection_count(), 2);
 
-        manager.broadcast(SseEvent::Heartbeat);
+        manager.broadcast(AppEvent::Heartbeat);
 
         let e1 = s1.next().await.unwrap();
         let e2 = s2.next().await.unwrap();
-        assert!(matches!(e1, SseEvent::Heartbeat));
+        assert!(matches!(e1, AppEvent::Heartbeat));
-        assert!(matches!(e2, SseEvent::Heartbeat));
+        assert!(matches!(e2, AppEvent::Heartbeat));
 
         drop(s1);
         assert_eq!(manager.connection_count(), 1);
@@ -373,25 +350,25 @@ mod tests {
         // Send event scoped to alice
         manager.broadcast_for_user(
             "alice",
-            SseEvent::Status {
+            AppEvent::Status {
                 message: "alice only".to_string(),
                 thread_id: None,
             },
         );
 
         // Send global event
-        manager.broadcast(SseEvent::Heartbeat);
+        manager.broadcast(AppEvent::Heartbeat);
 
         // Alice gets her scoped event
         let e = alice.next().await.unwrap();
-        assert!(matches!(e, SseEvent::Status { .. }));
+        assert!(matches!(e, AppEvent::Status { .. }));
 
         // Alice also gets the global heartbeat
         let e = alice.next().await.unwrap();
-        assert!(matches!(e, SseEvent::Heartbeat));
+        assert!(matches!(e, AppEvent::Heartbeat));
 
         // Bob only gets the global heartbeat (alice's event was filtered)
-        let e = bob.next().await.unwrap();
+        let e = bob.next().await.unwrap(); // safety: test-only
-        assert!(matches!(e, SseEvent::Heartbeat));
+        assert!(matches!(e, AppEvent::Heartbeat)); // safety: test assertion
     }
 }

src/channels/web/static/app.js

@@ -4265,9 +4265,9 @@ function renderRoutineDetail(routine) {
       + '<th>Trigger</th><th>Started</th><th>Completed</th><th>Status</th><th>Summary</th><th>Tokens</th>'
       + '</tr></thead><tbody>';
     for (const run of routine.recent_runs) {
-      const runStatusClass = run.status === 'Ok' ? 'completed'
+      const runStatusClass = run.status === 'ok' ? 'completed'
-        : run.status === 'Failed' ? 'failed'
+        : run.status === 'failed' ? 'failed'
-        : run.status === 'Attention' ? 'stuck'
+        : run.status === 'attention' ? 'stuck'
         : 'in_progress';
       html += '<tr>'
         + '<td>' + escapeHtml(run.trigger_type) + '</td>'

src/channels/web/test_helpers.rs

@@ -76,7 +76,8 @@ impl TestGatewayBuilder {
             store: None,
             job_manager: None,
             prompt_queue: None,
-            default_user_id: self.user_id,
+            owner_id: self.user_id.clone(),
+            default_sender_id: self.user_id,
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: Some(Arc::new(WsConnectionTracker::new())),
             llm_provider: self.llm_provider,

src/channels/web/tests/multi_tenant.rs

@@ -16,6 +16,7 @@ use axum::routing::{delete, get, post};
 use tower::ServiceExt;
 use uuid::Uuid;
 
+use crate::channels::web::GatewayChannel;
 use crate::channels::web::auth::{
     AuthenticatedUser, MultiAuthState, UserIdentity, auth_middleware,
 };
@@ -23,6 +24,7 @@ use crate::channels::web::server::{
     ActiveConfigSnapshot, GatewayState, PerUserRateLimiter, PromptQueue, RateLimiter, WorkspacePool,
 };
 use crate::channels::web::sse::SseManager;
+use crate::config::GatewayConfig;
 
 // ── Helpers ────────────────────────────────────────────────────────────
 
@@ -64,7 +66,8 @@ fn build_state(
         store,
         job_manager: None,
         prompt_queue,
-        default_user_id: "test".to_string(),
+        owner_id: "test".to_string(),
+        default_sender_id: "test".to_string(),
         shutdown_tx: tokio::sync::RwLock::new(None),
         ws_tracker: None,
         llm_provider: None,
@@ -82,6 +85,40 @@ fn build_state(
     })
 }
 
+fn gateway_config() -> GatewayConfig {
+    GatewayConfig {
+        host: "127.0.0.1".to_string(),
+        port: 3000,
+        auth_token: Some("gateway-auth".to_string()),
+        user_id: "gateway-sender".to_string(),
+        workspace_read_scopes: Vec::new(),
+        memory_layers: Vec::new(),
+        user_tokens: None,
+    }
+}
+
+#[test]
+fn with_owner_scope_updates_gateway_owner_scope_in_multi_user_mode() {
+    let mut gateway = GatewayChannel::new(gateway_config());
+    gateway.auth = two_user_auth();
+    gateway.config.user_tokens = Some(HashMap::new());
+    let gateway = gateway.with_owner_scope("owner-scope");
+
+    assert_eq!(gateway.state.owner_id, "owner-scope");
+    assert_eq!(gateway.state.default_sender_id, "gateway-sender");
+
+    let alice = gateway
+        .auth
+        .authenticate("tok-alice")
+        .expect("alice token should remain valid");
+    let bob = gateway
+        .auth
+        .authenticate("tok-bob")
+        .expect("bob token should remain valid");
+    assert_eq!(alice.user_id, "alice");
+    assert_eq!(bob.user_id, "bob");
+}
+
 /// Create a libSQL-backed test database in a temporary directory.
 ///
 /// Returns the database and a `TempDir` guard — the database file is

src/channels/web/types.rs

@@ -63,6 +63,9 @@ pub struct TurnInfo {
     pub started_at: String,
     pub completed_at: Option<String>,
     pub tool_calls: Vec<ToolCallInfo>,
+    /// Agent's reasoning narrative for this turn.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub narrative: Option<String>,
 }
 
 #[derive(Debug, Serialize)]
@@ -74,6 +77,9 @@ pub struct ToolCallInfo {
     pub result_preview: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub error: Option<String>,
+    /// Agent's reasoning for choosing this tool.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub rationale: Option<String>,
 }
 
 #[derive(Debug, Serialize)]
@@ -114,165 +120,9 @@ pub struct ApprovalRequest {
     pub thread_id: Option<String>,
 }
 
-// --- SSE Event Types ---
+// --- App Event (re-exported from ironclaw_common) ---
 
-#[derive(Debug, Clone, Serialize)]
+pub use ironclaw_common::{AppEvent, ToolDecisionDto};
-#[serde(tag = "type")]
-pub enum SseEvent {
-    #[serde(rename = "response")]
-    Response { content: String, thread_id: String },
-    #[serde(rename = "thinking")]
-    Thinking {
-        message: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-    #[serde(rename = "tool_started")]
-    ToolStarted {
-        name: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-    #[serde(rename = "tool_completed")]
-    ToolCompleted {
-        name: String,
-        success: bool,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        error: Option<String>,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        parameters: Option<String>,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-    #[serde(rename = "tool_result")]
-    ToolResult {
-        name: String,
-        preview: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-    #[serde(rename = "stream_chunk")]
-    StreamChunk {
-        content: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-    #[serde(rename = "status")]
-    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,
-        tool_name: String,
-        description: String,
-        parameters: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-        /// Whether the "always" auto-approve option should be shown.
-        allow_always: bool,
-    },
-    #[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,
-        #[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>,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        fallback_deliverable: Option<serde_json::Value>,
-    },
-
-    /// An image was generated by a tool.
-    #[serde(rename = "image_generated")]
-    ImageGenerated {
-        data_url: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        path: Option<String>,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-
-    /// Suggested follow-up messages for the user.
-    #[serde(rename = "suggestions")]
-    Suggestions {
-        suggestions: Vec<String>,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-
-    /// Per-turn token usage and cost summary.
-    #[serde(rename = "turn_cost")]
-    TurnCost {
-        input_tokens: u64,
-        output_tokens: u64,
-        cost_usd: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        thread_id: Option<String>,
-    },
-
-    /// Extension activation status change (WASM channels).
-    #[serde(rename = "extension_status")]
-    ExtensionStatus {
-        extension_name: String,
-        status: String,
-        #[serde(skip_serializing_if = "Option::is_none")]
-        message: Option<String>,
-    },
-}
 
 // --- Memory ---
 
@@ -784,32 +634,9 @@ pub enum WsServerMessage {
 }
 
 impl WsServerMessage {
-    /// Create a WsServerMessage from an SseEvent.
+    /// Create a WsServerMessage from an AppEvent.
-    pub fn from_sse_event(event: &SseEvent) -> Self {
+    pub fn from_app_event(event: &AppEvent) -> Self {
-        let event_type = match event {
+        let event_type = event.event_type();
-            SseEvent::Response { .. } => "response",
-            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",
-            SseEvent::ImageGenerated { .. } => "image_generated",
-            SseEvent::Suggestions { .. } => "suggestions",
-            SseEvent::TurnCost { .. } => "turn_cost",
-            SseEvent::ExtensionStatus { .. } => "extension_status",
-        };
         let data = serde_json::to_value(event).unwrap_or(serde_json::Value::Null);
         WsServerMessage::Event {
             event_type: event_type.to_string(),
@@ -1101,12 +928,12 @@ mod tests {
     }
 
     #[test]
-    fn test_ws_server_from_sse_response() {
+    fn test_ws_server_from_app_event_response() {
-        let sse = SseEvent::Response {
+        let event = AppEvent::Response {
             content: "hello".to_string(),
             thread_id: "t1".to_string(),
         };
-        let ws = WsServerMessage::from_sse_event(&sse);
+        let ws = WsServerMessage::from_app_event(&event);
         match ws {
             WsServerMessage::Event { event_type, data } => {
                 assert_eq!(event_type, "response");
@@ -1118,12 +945,12 @@ mod tests {
     }
 
     #[test]
-    fn test_ws_server_from_sse_thinking() {
+    fn test_ws_server_from_app_event_thinking() {
-        let sse = SseEvent::Thinking {
+        let event = AppEvent::Thinking {
             message: "reasoning...".to_string(),
             thread_id: None,
         };
-        let ws = WsServerMessage::from_sse_event(&sse);
+        let ws = WsServerMessage::from_app_event(&event);
         match ws {
             WsServerMessage::Event { event_type, data } => {
                 assert_eq!(event_type, "thinking");
@@ -1134,8 +961,8 @@ mod tests {
     }
 
     #[test]
-    fn test_ws_server_from_sse_approval_needed() {
+    fn test_ws_server_from_app_event_approval_needed() {
-        let sse = SseEvent::ApprovalNeeded {
+        let event = AppEvent::ApprovalNeeded {
             request_id: "r1".to_string(),
             tool_name: "shell".to_string(),
             description: "Run ls".to_string(),
@@ -1143,7 +970,7 @@ mod tests {
             thread_id: Some("t1".to_string()),
             allow_always: true,
         };
-        let ws = WsServerMessage::from_sse_event(&sse);
+        let ws = WsServerMessage::from_app_event(&event);
         match ws {
             WsServerMessage::Event { event_type, data } => {
                 assert_eq!(event_type, "approval_needed");
@@ -1155,9 +982,9 @@ mod tests {
     }
 
     #[test]
-    fn test_ws_server_from_sse_heartbeat() {
+    fn test_ws_server_from_app_event_heartbeat() {
-        let sse = SseEvent::Heartbeat;
+        let event = AppEvent::Heartbeat;
-        let ws = WsServerMessage::from_sse_event(&sse);
+        let ws = WsServerMessage::from_app_event(&event);
         match ws {
             WsServerMessage::Event { event_type, .. } => {
                 assert_eq!(event_type, "heartbeat");
@@ -1197,8 +1024,8 @@ mod tests {
     }
 
     #[test]
-    fn test_sse_auth_required_serialize() {
+    fn test_app_event_auth_required_serialize() {
-        let event = SseEvent::AuthRequired {
+        let event = AppEvent::AuthRequired {
             extension_name: "notion".to_string(),
             instructions: Some("Get your token from...".to_string()),
             auth_url: None,
@@ -1214,8 +1041,8 @@ mod tests {
     }
 
     #[test]
-    fn test_sse_auth_completed_serialize() {
+    fn test_app_event_auth_completed_serialize() {
-        let event = SseEvent::AuthCompleted {
+        let event = AppEvent::AuthCompleted {
             extension_name: "notion".to_string(),
             success: true,
             message: "notion authenticated (3 tools loaded)".to_string(),
@@ -1228,14 +1055,14 @@ mod tests {
     }
 
     #[test]
-    fn test_ws_server_from_sse_auth_required() {
+    fn test_ws_server_from_app_event_auth_required() {
-        let sse = SseEvent::AuthRequired {
+        let event = AppEvent::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);
+        let ws = WsServerMessage::from_app_event(&event);
         match ws {
             WsServerMessage::Event { event_type, data } => {
                 assert_eq!(event_type, "auth_required");
@@ -1246,13 +1073,13 @@ mod tests {
     }
 
     #[test]
-    fn test_ws_server_from_sse_auth_completed() {
+    fn test_ws_server_from_app_event_auth_completed() {
-        let sse = SseEvent::AuthCompleted {
+        let event = AppEvent::AuthCompleted {
             extension_name: "slack".to_string(),
             success: false,
             message: "Invalid token".to_string(),
         };
-        let ws = WsServerMessage::from_sse_event(&sse);
+        let ws = WsServerMessage::from_app_event(&event);
         match ws {
             WsServerMessage::Event { event_type, data } => {
                 assert_eq!(event_type, "auth_completed");

src/channels/web/util.rs

@@ -2,28 +2,21 @@
 
 use crate::channels::web::types::{ToolCallInfo, TurnInfo};
 
-/// Truncate a string to at most `max_bytes` bytes at a char boundary, appending "...".
+pub use ironclaw_common::truncate_preview;
-///
-/// If the input is wrapped in `<tool_output …>…</tool_output>` and truncation
-/// removes the closing tag, the tag is re-appended so downstream XML parsers
-/// never see an unclosed element.
-pub fn truncate_preview(s: &str, max_bytes: usize) -> String {
-    if s.len() <= max_bytes {
-        return s.to_string();
-    }
-    // Walk backwards from max_bytes to find a valid char boundary
-    let mut end = max_bytes;
-    while end > 0 && !s.is_char_boundary(end) {
-        end -= 1;
-    }
-    let mut result = format!("{}...", &s[..end]);
 
-    // Re-close <tool_output> if truncation cut through the closing tag.
+/// Parse tool call summary JSON objects into `ToolCallInfo` structs.
-    if s.starts_with("<tool_output") && !result.ends_with("</tool_output>") {
+fn parse_tool_call_infos(calls: &[serde_json::Value]) -> Vec<ToolCallInfo> {
-        result.push_str("\n</tool_output>");
+    calls
-    }
+        .iter()
-
+        .map(|c| ToolCallInfo {
-    result
+            name: c["name"].as_str().unwrap_or("unknown").to_string(),
+            has_result: c.get("result_preview").is_some_and(|v| !v.is_null()),
+            has_error: c.get("error").is_some_and(|v| !v.is_null()),
+            result_preview: c["result_preview"].as_str().map(String::from),
+            error: c["error"].as_str().map(String::from),
+            rationale: c["rationale"].as_str().map(String::from),
+        })
+        .collect()
 }
 
 /// Build TurnInfo pairs from flat DB messages (user/tool_calls/assistant triples).
@@ -49,6 +42,7 @@ pub fn build_turns_from_db_messages(
                 started_at: msg.created_at.to_rfc3339(),
                 completed_at: None,
                 tool_calls: Vec::new(),
+                narrative: None,
             };
 
             // Check if next message is a tool_calls record
@@ -56,18 +50,28 @@ pub fn build_turns_from_db_messages(
                 && next.role == "tool_calls"
             {
                 let tc_msg = iter.next().expect("peeked");
-                match serde_json::from_str::<Vec<serde_json::Value>>(&tc_msg.content) {
+                // Parse tool_calls JSON — supports two formats:
-                    Ok(calls) => {
+                // safety: no byte-index slicing; comment describes JSON shape
-                        turn.tool_calls = calls
+                match serde_json::from_str::<serde_json::Value>(&tc_msg.content) {
-                            .iter()
+                    Ok(serde_json::Value::Array(calls)) => {
-                            .map(|c| ToolCallInfo {
+                        // Old format: plain array
-                                name: c["name"].as_str().unwrap_or("unknown").to_string(),
+                        turn.tool_calls = parse_tool_call_infos(&calls);
-                                has_result: c.get("result_preview").is_some(),
+                    }
-                                has_error: c.get("error").is_some(),
+                    Ok(serde_json::Value::Object(obj)) => {
-                                result_preview: c["result_preview"].as_str().map(String::from),
+                        // New wrapped format with narrative
-                                error: c["error"].as_str().map(String::from),
+                        turn.narrative = obj
-                            })
+                            .get("narrative")
-                            .collect();
+                            .and_then(|v| v.as_str())
+                            .map(String::from);
+                        if let Some(serde_json::Value::Array(calls)) = obj.get("calls") {
+                            turn.tool_calls = parse_tool_call_infos(calls);
+                        }
+                    }
+                    Ok(_) => {
+                        tracing::warn!(
+                            message_id = %tc_msg.id,
+                            "Unexpected tool_calls JSON shape in DB, skipping"
+                        );
                     }
                     Err(e) => {
                         tracing::warn!(
@@ -105,6 +109,7 @@ pub fn build_turns_from_db_messages(
                 started_at: msg.created_at.to_rfc3339(),
                 completed_at: Some(msg.created_at.to_rfc3339()),
                 tool_calls: Vec::new(),
+                narrative: None,
             });
             turn_number += 1;
         }
@@ -118,88 +123,6 @@ mod tests {
     use super::*;
     use uuid::Uuid;
 
-    // ---- truncate_preview tests ----
-
-    #[test]
-    fn test_truncate_preview_short_string() {
-        assert_eq!(truncate_preview("hello", 10), "hello");
-    }
-
-    #[test]
-    fn test_truncate_preview_exact_boundary() {
-        assert_eq!(truncate_preview("hello", 5), "hello");
-    }
-
-    #[test]
-    fn test_truncate_preview_truncates_ascii() {
-        assert_eq!(truncate_preview("hello world", 5), "hello...");
-    }
-
-    #[test]
-    fn test_truncate_preview_empty_string() {
-        assert_eq!(truncate_preview("", 10), "");
-    }
-
-    #[test]
-    fn test_truncate_preview_multibyte_char_boundary() {
-        // '€' is 3 bytes (E2 82 AC). "a€b" = [61, E2, 82, AC, 62] = 5 bytes
-        // Truncating at max_bytes=3 should not split the euro sign.
-        let s = "a€b";
-        let result = truncate_preview(s, 3);
-        // max_bytes=3 lands mid-€, so it walks back to byte 1 ("a")
-        assert_eq!(result, "a...");
-    }
-
-    #[test]
-    fn test_truncate_preview_emoji() {
-        // '🦀' is 4 bytes. "hi🦀" = 6 bytes
-        let s = "hi🦀";
-        let result = truncate_preview(s, 4);
-        // max_bytes=4 lands mid-🦀, walks back to byte 2 ("hi")
-        assert_eq!(result, "hi...");
-    }
-
-    #[test]
-    fn test_truncate_preview_cjk() {
-        // CJK characters are 3 bytes each. "你好世界" = 12 bytes
-        let s = "你好世界";
-        let result = truncate_preview(s, 7);
-        // max_bytes=7 lands mid-character (byte 7 is inside 世), walks back to 6 ("你好")
-        assert_eq!(result, "你好...");
-    }
-
-    #[test]
-    fn test_truncate_preview_zero_max_bytes() {
-        assert_eq!(truncate_preview("hello", 0), "...");
-    }
-
-    #[test]
-    fn test_truncate_preview_closes_tool_output_tag() {
-        let s = "<tool_output name=\"search\">\nSome very long content here\n</tool_output>";
-        // Truncate so it cuts before the closing tag
-        let result = truncate_preview(s, 60);
-        assert!(result.ends_with("</tool_output>"));
-        assert!(result.contains("..."));
-    }
-
-    #[test]
-    fn test_truncate_preview_no_extra_close_when_intact() {
-        let s = "<tool_output name=\"echo\">\nshort\n</tool_output>";
-        // The string is short enough not to be truncated
-        let result = truncate_preview(s, 500);
-        assert_eq!(result, s);
-        // Should not have a duplicate closing tag
-        assert_eq!(result.matches("</tool_output>").count(), 1);
-    }
-
-    #[test]
-    fn test_truncate_preview_non_xml_unaffected() {
-        let s = "Just a plain long string that gets truncated";
-        let result = truncate_preview(s, 10);
-        assert_eq!(result, "Just a pla...");
-        assert!(!result.contains("</tool_output>"));
-    }
-
     // ---- build_turns_from_db_messages tests ----
 
     fn make_msg(role: &str, content: &str, offset_ms: i64) -> crate::history::ConversationMessage {
@@ -305,4 +228,52 @@ mod tests {
         assert!(turns[0].tool_calls.is_empty());
         assert_eq!(turns[0].state, "Completed");
     }
+
+    #[test]
+    fn test_build_turns_with_wrapped_tool_calls_format() {
+        let tc_json = serde_json::json!({
+            "narrative": "Searching memory for context before proceeding.",
+            "calls": [
+                {"name": "memory_search", "result_preview": "found 3 items", "rationale": "consult prior context"},
+                {"name": "shell", "error": "permission denied"}
+            ]
+        });
+        let messages = vec![
+            make_msg("user", "Find info", 0),
+            make_msg("tool_calls", &tc_json.to_string(), 500),
+            make_msg("assistant", "Here's what I found", 1000),
+        ];
+        let turns = build_turns_from_db_messages(&messages);
+        assert_eq!(turns.len(), 1);
+        assert_eq!(
+            turns[0].narrative.as_deref(),
+            Some("Searching memory for context before proceeding.")
+        );
+        assert_eq!(turns[0].tool_calls.len(), 2);
+        assert_eq!(turns[0].tool_calls[0].name, "memory_search");
+        assert_eq!(
+            turns[0].tool_calls[0].rationale.as_deref(),
+            Some("consult prior context")
+        );
+        assert!(turns[0].tool_calls[0].has_result);
+        assert_eq!(turns[0].tool_calls[1].name, "shell");
+        assert!(turns[0].tool_calls[1].has_error);
+        assert_eq!(turns[0].response.as_deref(), Some("Here's what I found"));
+    }
+
+    #[test]
+    fn test_build_turns_wrapped_format_without_narrative() {
+        let tc_json = serde_json::json!({
+            "calls": [{"name": "echo", "result_preview": "hello"}]
+        });
+        let messages = vec![
+            make_msg("user", "Say hi", 0),
+            make_msg("tool_calls", &tc_json.to_string(), 500),
+            make_msg("assistant", "Done", 1000),
+        ];
+        let turns = build_turns_from_db_messages(&messages);
+        assert_eq!(turns.len(), 1);
+        assert!(turns[0].narrative.is_none());
+        assert_eq!(turns[0].tool_calls.len(), 1);
+    }
 }

src/channels/web/ws.rs

@@ -97,7 +97,7 @@ pub async fn handle_ws_connection(
             let msg = tokio::select! {
                 event = event_stream.next() => {
                     match event {
-                        Some(sse_event) => WsServerMessage::from_sse_event(&sse_event),
+                        Some(app_event) => WsServerMessage::from_app_event(&app_event),
                         None => break, // Broadcast channel closed
                     }
                 }
@@ -275,7 +275,7 @@ async fn handle_client_message(
                         if result.verification.is_some() {
                             state.sse.broadcast_for_user(
                                 user_id,
-                                crate::channels::web::types::SseEvent::AuthRequired {
+                                crate::channels::web::types::AppEvent::AuthRequired {
                                     extension_name: extension_name.clone(),
                                     instructions: Some(result.message),
                                     auth_url: None,
@@ -286,7 +286,7 @@ async fn handle_client_message(
                             crate::channels::web::server::clear_auth_mode(state, user_id).await;
                             state.sse.broadcast_for_user(
                                 user_id,
-                                crate::channels::web::types::SseEvent::AuthCompleted {
+                                crate::channels::web::types::AppEvent::AuthCompleted {
                                     extension_name,
                                     success: true,
                                     message: result.message,
@@ -299,7 +299,7 @@ async fn handle_client_message(
                         if matches!(e, crate::extensions::ExtensionError::ValidationFailed(_)) {
                             state.sse.broadcast_for_user(
                                 user_id,
-                                crate::channels::web::types::SseEvent::AuthRequired {
+                                crate::channels::web::types::AppEvent::AuthRequired {
                                     extension_name: extension_name.clone(),
                                     instructions: Some(msg.clone()),
                                     auth_url: None,
@@ -520,7 +520,8 @@ mod tests {
             job_manager: None,
             prompt_queue: None,
             scheduler: None,
-            default_user_id: "test".to_string(),
+            owner_id: "test".to_string(),
+            default_sender_id: "test".to_string(),
             shutdown_tx: tokio::sync::RwLock::new(None),
             ws_tracker: Some(Arc::new(WsConnectionTracker::new())),
             llm_provider: None,

src/cli/oauth_defaults.rs

@@ -62,6 +62,30 @@ pub fn builtin_client_id_override_env(secret_name: &str) -> Option<&'static str>
     }
 }
 
+/// Suppress the baked-in desktop OAuth client secret when a hosted proxy is configured.
+///
+/// In hosted deployments, IronClaw may resolve the platform Google client ID from
+/// environment variables while still falling back to the baked-in desktop secret.
+/// That client_id/client_secret mismatch breaks Google token exchange and refresh.
+///
+/// When the proxy is configured, the platform will inject the correct server-side
+/// secret for matching platform credentials, so the baked-in secret must be omitted.
+pub fn hosted_proxy_client_secret(
+    client_secret: &Option<String>,
+    builtin: Option<&OAuthCredentials>,
+    exchange_proxy_configured: bool,
+) -> Option<String> {
+    if !exchange_proxy_configured {
+        return client_secret.clone();
+    }
+
+    let builtin_secret = builtin.map(|credentials| credentials.client_secret);
+    match (client_secret, builtin_secret) {
+        (Some(resolved), Some(baked_in)) if resolved == baked_in => None,
+        _ => client_secret.clone(),
+    }
+}
+
 // ── Shared callback server ──────────────────────────────────────────────
 
 // Core OAuth callback infrastructure is defined in `crate::llm::oauth_helpers`
@@ -449,7 +473,8 @@ pub struct PendingOAuthFlow {
     pub secrets: Arc<dyn SecretsStore + Send + Sync>,
     /// SSE broadcast manager for notifying the web UI.
     pub sse_manager: Option<Arc<crate::channels::web::sse::SseManager>>,
-    /// Gateway auth token for authenticating with the platform token exchange proxy.
+    /// OAuth proxy auth token for authenticating with the hosted token exchange proxy.
+    /// Kept as `gateway_token` for public API compatibility.
     pub gateway_token: Option<String>,
     /// Additional form params for the token exchange request.
     /// Used for provider-specific requirements such as RFC 8707 `resource`.
@@ -472,6 +497,12 @@ impl std::fmt::Debug for PendingOAuthFlow {
     }
 }
 
+impl PendingOAuthFlow {
+    pub fn oauth_proxy_auth_token(&self) -> Option<&str> {
+        self.gateway_token.as_deref()
+    }
+}
+
 /// Thread-safe registry of pending OAuth flows, keyed by CSRF `state` parameter.
 pub type PendingOAuthRegistry = Arc<RwLock<HashMap<String, PendingOAuthFlow>>>;
 
@@ -505,6 +536,22 @@ pub fn exchange_proxy_url() -> Option<String> {
         .filter(|url| !url.is_empty())
 }
 
+/// Returns the configured OAuth proxy auth token, if any.
+///
+/// New hosted infra can inject a dedicated shared proxy secret via
+/// `IRONCLAW_OAUTH_PROXY_AUTH_TOKEN`. Existing hosted instances continue to
+/// work by falling back to `GATEWAY_AUTH_TOKEN`.
+pub fn oauth_proxy_auth_token() -> Option<String> {
+    fn normalized_env_value(key: &str) -> Option<String> {
+        crate::config::helpers::env_or_override(key)
+            .map(|value| value.trim().to_string())
+            .filter(|value| !value.is_empty())
+    }
+
+    normalized_env_value("IRONCLAW_OAUTH_PROXY_AUTH_TOKEN")
+        .or_else(|| normalized_env_value("GATEWAY_AUTH_TOKEN"))
+}
+
 /// Maximum age for pending OAuth flows (5 minutes, matching TCP listener timeout).
 pub const OAUTH_FLOW_EXPIRY: Duration = Duration::from_secs(300);
 
@@ -650,6 +697,8 @@ pub fn strip_instance_prefix(state: &str) -> &str {
 
 pub struct ProxyTokenExchangeRequest<'a> {
     pub proxy_url: &'a str,
+    /// OAuth proxy auth token.
+    /// Kept as `gateway_token` for public API compatibility.
     pub gateway_token: &'a str,
     pub token_url: &'a str,
     pub client_id: &'a str,
@@ -661,9 +710,53 @@ pub struct ProxyTokenExchangeRequest<'a> {
     pub extra_token_params: &'a HashMap<String, String>,
 }
 
+pub struct ProxyRefreshTokenRequest<'a> {
+    pub proxy_url: &'a str,
+    /// OAuth proxy auth token.
+    /// Kept as `gateway_token` for public API compatibility.
+    pub gateway_token: &'a str,
+    pub token_url: &'a str,
+    pub client_id: &'a str,
+    pub client_secret: Option<&'a str>,
+    pub refresh_token: &'a str,
+    pub provider: Option<&'a str>,
+}
+
+fn oauth_token_response_from_json(
+    token_data: serde_json::Value,
+    access_token_field: &str,
+) -> Result<OAuthTokenResponse, OAuthCallbackError> {
+    let access_token = token_data
+        .get(access_token_field)
+        .and_then(|v| v.as_str())
+        .ok_or_else(|| {
+            let fields: Vec<&str> = token_data
+                .as_object()
+                .map(|o| o.keys().map(|k| k.as_str()).collect())
+                .unwrap_or_default();
+            OAuthCallbackError::Io(format!(
+                "No '{}' field in proxy response (fields present: {:?})",
+                access_token_field, fields
+            ))
+        })?
+        .to_string();
+
+    let refresh_token = token_data
+        .get("refresh_token")
+        .and_then(|v| v.as_str())
+        .map(String::from);
+    let expires_in = token_data.get("expires_in").and_then(|v| v.as_u64());
+
+    Ok(OAuthTokenResponse {
+        access_token,
+        refresh_token,
+        expires_in,
+    })
+}
+
 /// Exchange an OAuth authorization code via the platform's token exchange proxy.
 ///
-/// Authenticated via the gateway auth token (Bearer header). The caller may
+/// Authenticated via an OAuth proxy auth token (Bearer header). The caller may
 /// either rely on proxy-side secret lookup or forward a `client_secret` when
 /// the provider requires it.
 ///
@@ -675,13 +768,14 @@ pub async fn exchange_via_proxy(
 ) -> Result<OAuthTokenResponse, OAuthCallbackError> {
     if request.gateway_token.is_empty() {
         return Err(OAuthCallbackError::Io(
-            "Gateway auth token is required for proxy token exchange".to_string(),
+            "OAuth proxy auth token is required for proxy token exchange".to_string(),
         ));
     }
     let exchange_url = format!("{}/oauth/exchange", request.proxy_url.trim_end_matches('/'));
 
     let client = reqwest::Client::builder()
         .timeout(Duration::from_secs(60))
+        .redirect(reqwest::redirect::Policy::none())
         .build()
         .map_err(|e| OAuthCallbackError::Io(format!("Failed to build HTTP client: {}", e)))?;
     let mut params = vec![
@@ -724,41 +818,454 @@ pub async fn exchange_via_proxy(
         .json()
         .await
         .map_err(|e| OAuthCallbackError::Io(format!("Failed to parse proxy response: {}", e)))?;
+    oauth_token_response_from_json(token_data, request.access_token_field)
+}
 
-    let access_token = token_data
+/// Refresh an OAuth access token via the platform's token refresh proxy.
-        .get(request.access_token_field)
+///
-        .and_then(|v| v.as_str())
+/// Authenticated via an OAuth proxy auth token (Bearer header). The caller may
-        .ok_or_else(|| {
+/// either rely on proxy-side secret lookup or forward a `client_secret` when
-            let fields: Vec<&str> = token_data
+/// the provider requires it.
-                .as_object()
+pub async fn refresh_token_via_proxy(
-                .map(|o| o.keys().map(|k| k.as_str()).collect())
+    request: ProxyRefreshTokenRequest<'_>,
-                .unwrap_or_default();
+) -> Result<OAuthTokenResponse, OAuthCallbackError> {
-            OAuthCallbackError::Io(format!(
+    if request.gateway_token.is_empty() {
-                "No '{}' field in proxy response (fields present: {:?})",
+        return Err(OAuthCallbackError::Io(
-                request.access_token_field, fields
+            "OAuth proxy auth token is required for proxy token refresh".to_string(),
-            ))
+        ));
-        })?
+    }
-        .to_string();
 
-    let refresh_token = token_data
+    let refresh_url = format!("{}/oauth/refresh", request.proxy_url.trim_end_matches('/'));
-        .get("refresh_token")
+    let client = reqwest::Client::builder()
-        .and_then(|v| v.as_str())
+        .timeout(Duration::from_secs(15))
-        .map(String::from);
+        .redirect(reqwest::redirect::Policy::none())
-    let expires_in = token_data.get("expires_in").and_then(|v| v.as_u64());
+        .build()
+        .map_err(|e| OAuthCallbackError::Io(format!("Failed to build HTTP client: {}", e)))?;
 
-    Ok(OAuthTokenResponse {
+    let mut params = vec![
-        access_token,
+        ("refresh_token", request.refresh_token.to_string()),
-        refresh_token,
+        ("token_url", request.token_url.to_string()),
-        expires_in,
+        ("client_id", request.client_id.to_string()),
-    })
+    ];
+    if let Some(secret) = request.client_secret {
+        params.push(("client_secret", secret.to_string()));
+    }
+    if let Some(provider) = request.provider {
+        params.push(("provider", provider.to_string()));
+    }
+
+    let response = client
+        .post(&refresh_url)
+        .bearer_auth(request.gateway_token)
+        .form(&params)
+        .send()
+        .await
+        .map_err(|e| {
+            OAuthCallbackError::Io(format!("Token refresh proxy request failed: {}", e))
+        })?;
+
+    if !response.status().is_success() {
+        let status = response.status();
+        let body = response.text().await.unwrap_or_default();
+        return Err(OAuthCallbackError::Io(format!(
+            "Token refresh proxy failed: {} - {}",
+            status, body
+        )));
+    }
+
+    let token_data: serde_json::Value = response
+        .json()
+        .await
+        .map_err(|e| OAuthCallbackError::Io(format!("Failed to parse proxy response: {}", e)))?;
+
+    oauth_token_response_from_json(token_data, "access_token")
 }
 
 #[cfg(test)]
 mod tests {
+    use std::collections::HashMap;
+    use std::net::SocketAddr;
+    use std::sync::Arc;
+
+    use axum::extract::{Form, State};
+    use axum::http::HeaderMap;
+    use axum::response::Redirect;
+    use axum::routing::post;
+    use axum::{Json, Router};
+    use serde_json::json;
+    use tokio::net::TcpListener;
+    use tokio::sync::{Mutex, oneshot};
+
     use crate::cli::oauth_defaults::{
         builtin_credentials, callback_host, callback_url, is_loopback_host, landing_html,
     };
     use crate::config::helpers::lock_env;
+    use crate::testing::credentials::{TEST_OAUTH_CLIENT_ID, TEST_OAUTH_CLIENT_SECRET};
+
+    #[derive(Clone, Debug, PartialEq, Eq)]
+    struct RecordedProxyRequest {
+        authorization: Option<String>,
+        form: HashMap<String, String>,
+    }
+
+    #[derive(Clone)]
+    struct MockProxyState {
+        requests: Arc<Mutex<Vec<RecordedProxyRequest>>>,
+        exchange_redirect_target: String,
+        refresh_redirect_target: String,
+    }
+
+    struct MockProxyServer {
+        addr: SocketAddr,
+        requests: Arc<Mutex<Vec<RecordedProxyRequest>>>,
+        shutdown_tx: Option<oneshot::Sender<()>>,
+        server_task: Option<tokio::task::JoinHandle<()>>,
+    }
+
+    impl MockProxyServer {
+        async fn start() -> Self {
+            async fn exchange_handler(
+                State(state): State<MockProxyState>,
+                headers: HeaderMap,
+                Form(form): Form<HashMap<String, String>>,
+            ) -> Json<serde_json::Value> {
+                state.requests.lock().await.push(RecordedProxyRequest {
+                    authorization: headers
+                        .get(axum::http::header::AUTHORIZATION)
+                        .and_then(|value| value.to_str().ok())
+                        .map(str::to_string),
+                    form,
+                });
+                Json(json!({
+                    "access_token": "proxy-access-token",
+                    "refresh_token": "proxy-refresh-token",
+                    "expires_in": 7200
+                }))
+            }
+
+            async fn refresh_handler(
+                State(state): State<MockProxyState>,
+                headers: HeaderMap,
+                Form(form): Form<HashMap<String, String>>,
+            ) -> Json<serde_json::Value> {
+                state.requests.lock().await.push(RecordedProxyRequest {
+                    authorization: headers
+                        .get(axum::http::header::AUTHORIZATION)
+                        .and_then(|value| value.to_str().ok())
+                        .map(str::to_string),
+                    form,
+                });
+                Json(json!({
+                    "access_token": "proxy-access-token",
+                    "refresh_token": "proxy-refresh-token",
+                    "expires_in": 7200
+                }))
+            }
+
+            async fn exchange_redirect_handler(State(state): State<MockProxyState>) -> Redirect {
+                Redirect::temporary(&state.exchange_redirect_target)
+            }
+
+            async fn refresh_redirect_handler(State(state): State<MockProxyState>) -> Redirect {
+                Redirect::temporary(&state.refresh_redirect_target)
+            }
+
+            let requests = Arc::new(Mutex::new(Vec::new()));
+            let listener = TcpListener::bind("127.0.0.1:0")
+                .await
+                .expect("bind mock proxy");
+            let addr = listener.local_addr().expect("read mock proxy addr");
+            let exchange_redirect_target = format!("http://{addr}/oauth/exchange");
+            let refresh_redirect_target = format!("http://{addr}/oauth/refresh");
+            let app = Router::new()
+                .route("/oauth/exchange", post(exchange_handler))
+                .route("/oauth/refresh", post(refresh_handler))
+                .route("/redirect/oauth/exchange", post(exchange_redirect_handler))
+                .route("/redirect/oauth/refresh", post(refresh_redirect_handler))
+                .with_state(MockProxyState {
+                    requests: Arc::clone(&requests),
+                    exchange_redirect_target,
+                    refresh_redirect_target,
+                });
+            let (shutdown_tx, shutdown_rx) = oneshot::channel::<()>();
+            let server_task = tokio::spawn(async move {
+                let _ = axum::serve(listener, app)
+                    .with_graceful_shutdown(async {
+                        let _ = shutdown_rx.await;
+                    })
+                    .await;
+            });
+
+            Self {
+                addr,
+                requests,
+                shutdown_tx: Some(shutdown_tx),
+                server_task: Some(server_task),
+            }
+        }
+
+        fn base_url(&self) -> String {
+            format!("http://{}", self.addr)
+        }
+
+        fn redirecting_base_url(&self) -> String {
+            format!("{}/redirect", self.base_url())
+        }
+
+        async fn requests(&self) -> Vec<RecordedProxyRequest> {
+            self.requests.lock().await.clone()
+        }
+
+        async fn shutdown(mut self) {
+            if let Some(tx) = self.shutdown_tx.take() {
+                let _ = tx.send(());
+            }
+            if let Some(task) = self.server_task.take() {
+                let _ = task.await;
+            }
+        }
+    }
+
+    impl Drop for MockProxyServer {
+        fn drop(&mut self) {
+            if let Some(tx) = self.shutdown_tx.take() {
+                let _ = tx.send(());
+            }
+            if let Some(task) = self.server_task.take() {
+                task.abort();
+            }
+        }
+    }
+
+    struct EnvVarGuard {
+        key: &'static str,
+        original: Option<String>,
+    }
+
+    impl Drop for EnvVarGuard {
+        fn drop(&mut self) {
+            // SAFETY: Under ENV_MUTEX, no concurrent env access.
+            unsafe {
+                if let Some(ref value) = self.original {
+                    std::env::set_var(self.key, value);
+                } else {
+                    std::env::remove_var(self.key);
+                }
+            }
+        }
+    }
+
+    fn set_env_var(key: &'static str, value: Option<&str>) -> EnvVarGuard {
+        let original = std::env::var(key).ok();
+        // SAFETY: Under ENV_MUTEX, no concurrent env access.
+        unsafe {
+            if let Some(value) = value {
+                std::env::set_var(key, value);
+            } else {
+                std::env::remove_var(key);
+            }
+        }
+        EnvVarGuard { key, original }
+    }
+
+    #[test]
+    fn test_hosted_proxy_client_secret_suppresses_builtin_secret() {
+        let builtin = builtin_credentials("google_oauth_token").expect("google builtin creds");
+        let client_secret = Some(builtin.client_secret.to_string());
+
+        let result = super::hosted_proxy_client_secret(&client_secret, Some(&builtin), true);
+
+        assert_eq!(result, None);
+    }
+
+    #[test]
+    fn test_hosted_proxy_client_secret_preserves_explicit_secret() {
+        let builtin = builtin_credentials("google_oauth_token").expect("google builtin creds");
+        let client_secret = Some("hosted-server-secret".to_string());
+
+        let result = super::hosted_proxy_client_secret(&client_secret, Some(&builtin), true);
+
+        assert_eq!(result, client_secret);
+    }
+
+    #[tokio::test]
+    async fn test_exchange_via_proxy_sends_auth_and_form() {
+        let server = MockProxyServer::start().await;
+        let mut extra_token_params = HashMap::new();
+        extra_token_params.insert("resource".to_string(), "https://mcp.notion.com".to_string());
+
+        let response = super::exchange_via_proxy(super::ProxyTokenExchangeRequest {
+            proxy_url: &server.base_url(),
+            gateway_token: "shared-oauth-proxy-secret",
+            code: "auth-code-123",
+            redirect_uri: "https://oauth.example.com/oauth/callback",
+            token_url: "https://oauth2.googleapis.com/token",
+            client_id: TEST_OAUTH_CLIENT_ID,
+            client_secret: Some(TEST_OAUTH_CLIENT_SECRET),
+            access_token_field: "access_token",
+            code_verifier: Some("code-verifier-123"),
+            extra_token_params: &extra_token_params,
+        })
+        .await
+        .expect("proxy exchange succeeds");
+
+        assert_eq!(response.access_token, "proxy-access-token");
+        assert_eq!(
+            response.refresh_token.as_deref(),
+            Some("proxy-refresh-token")
+        );
+        assert_eq!(response.expires_in, Some(7200));
+
+        let requests = server.requests().await;
+        assert_eq!(requests.len(), 1);
+        assert_eq!(
+            requests[0].authorization.as_deref(),
+            Some("Bearer shared-oauth-proxy-secret")
+        );
+        assert_eq!(
+            requests[0].form.get("code").map(String::as_str),
+            Some("auth-code-123")
+        );
+        assert_eq!(
+            requests[0].form.get("redirect_uri").map(String::as_str),
+            Some("https://oauth.example.com/oauth/callback")
+        );
+        assert_eq!(
+            requests[0].form.get("token_url").map(String::as_str),
+            Some("https://oauth2.googleapis.com/token")
+        );
+        assert_eq!(
+            requests[0].form.get("client_id").map(String::as_str),
+            Some(TEST_OAUTH_CLIENT_ID)
+        );
+        assert_eq!(
+            requests[0].form.get("client_secret").map(String::as_str),
+            Some(TEST_OAUTH_CLIENT_SECRET)
+        );
+        assert_eq!(
+            requests[0]
+                .form
+                .get("access_token_field")
+                .map(String::as_str),
+            Some("access_token")
+        );
+        assert_eq!(
+            requests[0].form.get("code_verifier").map(String::as_str),
+            Some("code-verifier-123")
+        );
+        assert_eq!(
+            requests[0].form.get("resource").map(String::as_str),
+            Some("https://mcp.notion.com")
+        );
+
+        server.shutdown().await;
+    }
+
+    #[tokio::test]
+    async fn test_refresh_token_via_proxy_sends_auth_and_form() {
+        let server = MockProxyServer::start().await;
+
+        let response = super::refresh_token_via_proxy(super::ProxyRefreshTokenRequest {
+            proxy_url: &server.base_url(),
+            gateway_token: "gateway-test-token",
+            token_url: "https://oauth2.googleapis.com/token",
+            client_id: TEST_OAUTH_CLIENT_ID,
+            client_secret: Some(TEST_OAUTH_CLIENT_SECRET),
+            refresh_token: "refresh-token-123",
+            provider: Some("google"),
+        })
+        .await
+        .expect("proxy refresh succeeds");
+
+        assert_eq!(response.access_token, "proxy-access-token");
+        assert_eq!(
+            response.refresh_token.as_deref(),
+            Some("proxy-refresh-token")
+        );
+        assert_eq!(response.expires_in, Some(7200));
+
+        let requests = server.requests().await;
+        assert_eq!(requests.len(), 1);
+        assert_eq!(
+            requests[0].authorization.as_deref(),
+            Some("Bearer gateway-test-token")
+        );
+        assert_eq!(
+            requests[0].form.get("token_url").map(String::as_str),
+            Some("https://oauth2.googleapis.com/token")
+        );
+        assert_eq!(
+            requests[0].form.get("client_id").map(String::as_str),
+            Some(TEST_OAUTH_CLIENT_ID)
+        );
+        assert_eq!(
+            requests[0].form.get("client_secret").map(String::as_str),
+            Some(TEST_OAUTH_CLIENT_SECRET)
+        );
+        assert_eq!(
+            requests[0].form.get("refresh_token").map(String::as_str),
+            Some("refresh-token-123")
+        );
+        assert_eq!(
+            requests[0].form.get("provider").map(String::as_str),
+            Some("google")
+        );
+
+        server.shutdown().await;
+    }
+
+    #[tokio::test]
+    async fn test_exchange_via_proxy_does_not_follow_redirects() {
+        let server = MockProxyServer::start().await;
+
+        let error = match super::exchange_via_proxy(super::ProxyTokenExchangeRequest {
+            proxy_url: &server.redirecting_base_url(),
+            gateway_token: "gateway-test-token",
+            code: "auth-code-123",
+            redirect_uri: "http://localhost:3000/oauth/callback",
+            token_url: "https://oauth2.googleapis.com/token",
+            client_id: TEST_OAUTH_CLIENT_ID,
+            client_secret: Some(TEST_OAUTH_CLIENT_SECRET),
+            access_token_field: "access_token",
+            code_verifier: Some("code-verifier-123"),
+            extra_token_params: &HashMap::new(),
+        })
+        .await
+        {
+            Ok(_) => panic!("redirected proxy exchange should fail"),
+            Err(error) => error,
+        };
+
+        assert!(error.to_string().contains("307"));
+        assert!(server.requests().await.is_empty());
+
+        server.shutdown().await;
+    }
+
+    #[tokio::test]
+    async fn test_refresh_token_via_proxy_does_not_follow_redirects() {
+        let server = MockProxyServer::start().await;
+
+        let error = match super::refresh_token_via_proxy(super::ProxyRefreshTokenRequest {
+            proxy_url: &server.redirecting_base_url(),
+            gateway_token: "gateway-test-token",
+            token_url: "https://oauth2.googleapis.com/token",
+            client_id: TEST_OAUTH_CLIENT_ID,
+            client_secret: Some(TEST_OAUTH_CLIENT_SECRET),
+            refresh_token: "refresh-token-123",
+            provider: Some("google"),
+        })
+        .await
+        {
+            Ok(_) => panic!("redirected proxy refresh should fail"),
+            Err(error) => error,
+        };
+
+        assert!(error.to_string().contains("307"));
+        assert!(server.requests().await.is_empty());
+
+        server.shutdown().await;
+    }
 
     #[test]
     fn test_is_loopback_host() {
@@ -1159,6 +1666,54 @@ mod tests {
         }
     }
 
+    #[test]
+    fn test_oauth_proxy_auth_token_prefers_dedicated_env() {
+        let _guard = lock_env();
+        let _proxy_guard = set_env_var(
+            "IRONCLAW_OAUTH_PROXY_AUTH_TOKEN",
+            Some("shared-proxy-secret"),
+        );
+        let _gateway_guard = set_env_var("GATEWAY_AUTH_TOKEN", Some("gateway-token"));
+
+        assert_eq!(
+            crate::cli::oauth_defaults::oauth_proxy_auth_token().as_deref(),
+            Some("shared-proxy-secret")
+        );
+    }
+
+    #[test]
+    fn test_oauth_proxy_auth_token_falls_back_to_gateway_token() {
+        let _guard = lock_env();
+        let _proxy_guard = set_env_var("IRONCLAW_OAUTH_PROXY_AUTH_TOKEN", None);
+        let _gateway_guard = set_env_var("GATEWAY_AUTH_TOKEN", Some("gateway-token"));
+
+        assert_eq!(
+            crate::cli::oauth_defaults::oauth_proxy_auth_token().as_deref(),
+            Some("gateway-token")
+        );
+    }
+
+    #[test]
+    fn test_oauth_proxy_auth_token_whitespace_dedicated_env_falls_back_to_gateway_token() {
+        let _guard = lock_env();
+        let _proxy_guard = set_env_var("IRONCLAW_OAUTH_PROXY_AUTH_TOKEN", Some("   "));
+        let _gateway_guard = set_env_var("GATEWAY_AUTH_TOKEN", Some("gateway-token"));
+
+        assert_eq!(
+            crate::cli::oauth_defaults::oauth_proxy_auth_token().as_deref(),
+            Some("gateway-token")
+        );
+    }
+
+    #[test]
+    fn test_oauth_proxy_auth_token_returns_none_when_unset() {
+        let _guard = lock_env();
+        let _proxy_guard = set_env_var("IRONCLAW_OAUTH_PROXY_AUTH_TOKEN", None);
+        let _gateway_guard = set_env_var("GATEWAY_AUTH_TOKEN", None);
+
+        assert_eq!(crate::cli::oauth_defaults::oauth_proxy_auth_token(), None);
+    }
+
     #[test]
     fn test_strip_instance_prefix_with_colon() {
         use crate::cli::oauth_defaults::strip_instance_prefix;

src/cli/routines.rs

@@ -10,7 +10,7 @@ use clap::Subcommand;
 use uuid::Uuid;
 
 use crate::agent::routine::{
-    NotifyConfig, Routine, RoutineAction, RoutineGuardrails, Trigger, next_cron_fire,
+    NotifyConfig, Routine, RoutineAction, RoutineGuardrails, RunStatus, Trigger, next_cron_fire,
 };
 use crate::db::Database;
 
@@ -251,15 +251,26 @@ async fn list(
     );
     println!("{}", "-".repeat(130));
 
+    // Fetch last-run status for all routines in a single batch query
+    let routine_ids: Vec<Uuid> = filtered.iter().map(|r| r.id).collect();
+    let last_run_results = db
+        .batch_get_last_run_status(&routine_ids)
+        .await
+        .unwrap_or_default();
+
     for r in &filtered {
-        let status = if r.enabled {
+        let last_run_status = last_run_results.get(&r.id).copied();
-            if r.consecutive_failures > 0 {
+
-                format!("err({})", r.consecutive_failures)
+        let status = if !r.enabled {
-            } else {
-                "active".to_string()
-            }
-        } else {
             "disabled".to_string()
+        } else if last_run_status == Some(RunStatus::Running) {
+            "running".to_string()
+        } else if r.consecutive_failures > 0 {
+            format!("err({})", r.consecutive_failures)
+        } else if last_run_status == Some(RunStatus::Attention) {
+            "attention".to_string()
+        } else {
+            "active".to_string()
         };
 
         let next_fire = r
@@ -340,8 +351,8 @@ async fn create(
             prompt: prompt.to_string(),
             context_paths: Vec::new(),
             max_tokens: 4096,
-            use_tools: false,
+            use_tools: true,
-            max_tool_rounds: 0,
+            max_tool_rounds: 3,
         },
         guardrails: RoutineGuardrails {
             cooldown: std::time::Duration::from_secs(cooldown_secs),
@@ -685,6 +696,7 @@ fn truncate(s: &str, max_chars: usize) -> String {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::agent::routine::RoutineAction;
 
     #[test]
     fn format_relative_future() {
@@ -743,4 +755,48 @@ mod tests {
         assert!(notify.on_failure); // safety: test-only assertion
         assert!(!notify.on_success); // safety: test-only assertion
     }
+
+    #[cfg(feature = "libsql")]
+    #[tokio::test]
+    async fn cli_create_defaults_lightweight_routines_to_tools_enabled() {
+        let harness = crate::testing::TestHarnessBuilder::new().build().await;
+        let db = harness.db.clone();
+
+        run_routines_command(
+            RoutinesCommand::Create {
+                name: "cli-digest".to_string(),
+                schedule: "0 0 9 * * *".to_string(),
+                prompt: "Prepare the morning digest.".to_string(),
+                description: "CLI created routine".to_string(),
+                timezone: Some("UTC".to_string()),
+                cooldown: 300,
+                notify_channel: None,
+            },
+            db.clone(),
+            "user1",
+        )
+        .await
+        .expect("create routine");
+
+        let routine = db
+            .get_routine_by_name("user1", "cli-digest")
+            .await
+            .expect("get routine by name")
+            .expect("cli-digest should exist");
+
+        match routine.action {
+            RoutineAction::Lightweight {
+                use_tools,
+                max_tool_rounds,
+                ..
+            } => {
+                assert!(
+                    use_tools,
+                    "CLI-created lightweight routines should default to tools"
+                );
+                assert_eq!(max_tool_rounds, 3);
+            }
+            other => panic!("expected lightweight action, got {other:?}"),
+        }
+    }
 }

src/cli/tool.rs

@@ -2,6 +2,7 @@
 //!
 //! Commands for installing, listing, removing, and authenticating WASM tools.
 
+use std::collections::{HashMap, HashSet};
 use std::io::Write;
 use std::path::{Path, PathBuf};
 use std::sync::Arc;
@@ -79,6 +80,10 @@ pub enum ToolCommand {
         /// Directory to look for tool (default: ~/.ironclaw/tools/)
         #[arg(short, long)]
         dir: Option<PathBuf>,
+
+        /// User ID for checking credential status (default: "default")
+        #[arg(short, long, default_value = "default")]
+        user: String,
     },
 
     /// Configure authentication for a tool
@@ -124,7 +129,11 @@ pub async fn run_tool_command(cmd: ToolCommand) -> anyhow::Result<()> {
         } => install_tool(path, name, capabilities, target, release, skip_build, force).await,
         ToolCommand::List { dir, verbose } => list_tools(dir, verbose).await,
         ToolCommand::Remove { name, dir } => remove_tool(name, dir).await,
-        ToolCommand::Info { name_or_path, dir } => show_tool_info(name_or_path, dir).await,
+        ToolCommand::Info {
+            name_or_path,
+            dir,
+            user,
+        } => show_tool_info(name_or_path, dir, user).await,
         ToolCommand::Auth { name, dir, user } => auth_tool(name, dir, user).await,
         ToolCommand::Setup { name, dir, user } => setup_tool(name, dir, user).await,
     }
@@ -388,7 +397,11 @@ async fn remove_tool(name: String, dir: Option<PathBuf>) -> anyhow::Result<()> {
 }
 
 /// Show information about a tool.
-async fn show_tool_info(name_or_path: String, dir: Option<PathBuf>) -> anyhow::Result<()> {
+async fn show_tool_info(
+    name_or_path: String,
+    dir: Option<PathBuf>,
+    user_id: String,
+) -> anyhow::Result<()> {
     let wasm_path = if name_or_path.ends_with(".wasm") {
         PathBuf::from(&name_or_path)
     } else {
@@ -423,7 +436,37 @@ async fn show_tool_info(name_or_path: String, dir: Option<PathBuf>) -> anyhow::R
         println!("\nCapabilities ({}):", caps_path.display());
         let content = fs::read_to_string(&caps_path).await?;
         match CapabilitiesFile::from_json(&content) {
-            Ok(caps) => print_capabilities_detail(&caps),
+            Ok(caps) => {
+                // Lazily init secrets store only when auth secrets need checking.
+                let has_auth = caps.auth.is_some()
+                    || caps
+                        .setup
+                        .as_ref()
+                        .is_some_and(|s| !s.required_secrets.is_empty())
+                    || caps
+                        .http
+                        .as_ref()
+                        .is_some_and(|h| !h.credentials.is_empty());
+                let secrets_store = if has_auth {
+                    match init_secrets_store().await {
+                        Ok(store) => Some(store),
+                        Err(e) => {
+                            eprintln!("  Warning: could not init secrets store: {}", e);
+                            None
+                        }
+                    }
+                } else {
+                    None
+                };
+                print_capabilities_detail(
+                    &caps,
+                    secrets_store
+                        .as_ref()
+                        .map(|s| s.as_ref() as &(dyn SecretsStore + Send + Sync)),
+                    &user_id,
+                )
+                .await;
+            }
             Err(e) => println!("  Error parsing: {}", e),
         }
     } else {
@@ -476,8 +519,89 @@ fn print_capabilities_summary(caps: &CapabilitiesFile) {
     }
 }
 
+/// Per-secret info collected from all auth-related capability sections.
+struct AuthSecretInfo {
+    secret_name: String,
+    /// Human-readable label (from auth.display_name or setup prompt).
+    description: Option<String>,
+    /// Injection location (from http.credentials).
+    location: Option<String>,
+}
+
+/// Collected auth secrets and the set of secret names they cover.
+struct CollectedAuthSecrets {
+    secrets: Vec<AuthSecretInfo>,
+    /// Secret names present in `secrets`, for filtering the Secrets capability section.
+    seen_names: HashSet<String>,
+}
+
+/// Collect and deduplicate auth secrets from all auth-related capability sections.
+///
+/// Priority for the description label: auth.display_name > setup.required_secrets.prompt.
+/// Injection location is merged from http.credentials.
+fn collect_auth_secrets(caps: &CapabilitiesFile) -> CollectedAuthSecrets {
+    let mut secrets: Vec<AuthSecretInfo> = Vec::new();
+    let mut seen: HashMap<String, usize> = HashMap::new();
+
+    // auth.display_name is the best label — seed first.
+    if let Some(ref auth) = caps.auth {
+        let index = secrets.len();
+        seen.insert(auth.secret_name.clone(), index);
+        secrets.push(AuthSecretInfo {
+            secret_name: auth.secret_name.clone(),
+            description: auth.display_name.clone(),
+            location: None,
+        });
+    }
+
+    // setup.required_secrets.prompt is second-best label.
+    if let Some(ref setup) = caps.setup {
+        for secret in &setup.required_secrets {
+            if !seen.contains_key(&secret.name) {
+                let index = secrets.len();
+                seen.insert(secret.name.clone(), index);
+                secrets.push(AuthSecretInfo {
+                    secret_name: secret.name.clone(),
+                    description: Some(secret.prompt.clone()),
+                    location: None,
+                });
+            }
+        }
+    }
+
+    // Merge injection location from http.credentials.
+    if let Some(ref http) = caps.http {
+        for cred in http.credentials.values() {
+            let loc = format!("{:?}", cred.location);
+            if let Some(&index) = seen.get(&cred.secret_name) {
+                secrets[index].location = Some(loc);
+            } else {
+                let index = secrets.len();
+                seen.insert(cred.secret_name.clone(), index);
+                secrets.push(AuthSecretInfo {
+                    secret_name: cred.secret_name.clone(),
+                    description: None,
+                    location: Some(loc),
+                });
+            }
+        }
+    }
+
+    let seen_names = seen.into_keys().collect();
+    CollectedAuthSecrets {
+        secrets,
+        seen_names,
+    }
+}
+
 /// Print detailed capabilities.
-fn print_capabilities_detail(caps: &CapabilitiesFile) {
+async fn print_capabilities_detail(
+    caps: &CapabilitiesFile,
+    secrets_store: Option<&(dyn SecretsStore + Send + Sync)>,
+    user_id: &str,
+) {
+    let mut collected = collect_auth_secrets(caps);
+
     if let Some(ref http) = caps.http {
         println!("  HTTP:");
         for endpoint in &http.allowlist {
@@ -490,13 +614,6 @@ fn print_capabilities_detail(caps: &CapabilitiesFile) {
             println!("    {} {} {}", methods, endpoint.host, path);
         }
 
-        if !http.credentials.is_empty() {
-            println!("  Credentials:");
-            for (key, cred) in &http.credentials {
-                println!("    {}: {} -> {:?}", key, cred.secret_name, cred.location);
-            }
-        }
-
         if let Some(ref rate) = http.rate_limit {
             println!(
                 "  Rate limit: {}/min, {}/hour",
@@ -505,12 +622,24 @@ fn print_capabilities_detail(caps: &CapabilitiesFile) {
         }
     }
 
+    // Filter secrets already covered by the auth section (always rendered when non-empty).
     if let Some(ref secrets) = caps.secrets
         && !secrets.allowed_names.is_empty()
     {
-        println!("  Secrets (existence check only):");
+        let extra: Vec<_> = if collected.secrets.is_empty() {
-        for name in &secrets.allowed_names {
+            secrets.allowed_names.iter().collect()
-            println!("    {}", name);
+        } else {
+            secrets
+                .allowed_names
+                .iter()
+                .filter(|name| !collected.seen_names.contains(name.as_str()))
+                .collect()
+        };
+        if !extra.is_empty() {
+            println!("  Secrets (existence check only):");
+            for name in extra {
+                println!("    {}", name);
+            }
         }
     }
 
@@ -531,6 +660,38 @@ fn print_capabilities_detail(caps: &CapabilitiesFile) {
             println!("    {}", prefix);
         }
     }
+
+    // Consolidated auth status — sorted by secret name for deterministic output.
+    if !collected.secrets.is_empty() {
+        collected
+            .secrets
+            .sort_by(|a, b| a.secret_name.cmp(&b.secret_name));
+        println!("  Auth:");
+        for info in &collected.secrets {
+            let (icon, label) = match secrets_store {
+                Some(store) => match store.exists(user_id, &info.secret_name).await {
+                    Ok(true) => ("\u{2713}", "configured"),
+                    Ok(false) => ("\u{2717}", "missing"),
+                    Err(e) => {
+                        eprintln!(
+                            "  Warning: failed to check secret `{}`: {}",
+                            info.secret_name, e
+                        );
+                        ("?", "unknown")
+                    }
+                },
+                None => ("?", "unknown"),
+            };
+            let mut parts = info.secret_name.clone();
+            if let Some(ref desc) = info.description {
+                parts = format!("{} ({})", parts, desc);
+            }
+            if let Some(ref loc) = info.location {
+                parts = format!("{} -> {}", parts, loc);
+            }
+            println!("    {}  {} {}", parts, icon, label);
+        }
+    }
 }
 
 /// Validate a tool name to prevent path traversal.
@@ -677,8 +838,7 @@ async fn combine_provider_scopes(
     secret_name: &str,
     base_oauth: &crate::tools::wasm::OAuthConfigSchema,
 ) -> crate::tools::wasm::OAuthConfigSchema {
-    let mut all_scopes: std::collections::HashSet<String> =
+    let mut all_scopes: HashSet<String> = base_oauth.scopes.iter().cloned().collect();
-        base_oauth.scopes.iter().cloned().collect();
 
     if let Ok(mut entries) = tokio::fs::read_dir(tools_dir).await {
         while let Ok(Some(entry)) = entries.next_entry().await {
@@ -1127,6 +1287,8 @@ async fn setup_tool(name: String, dir: Option<PathBuf>, user_id: String) -> anyh
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::secrets::{CreateSecretParams, SecretsStore};
+    use crate::testing::credentials::test_secrets_store;
 
     #[test]
     fn test_format_size() {
@@ -1143,4 +1305,96 @@ mod tests {
         assert!(dir.to_string_lossy().contains(".ironclaw"));
         assert!(dir.to_string_lossy().contains("tools"));
     }
+
+    /// Verify that auth secrets are deduplicated across auth, setup, and http.credentials,
+    /// and that credential status is checked against the secrets store.
+    #[tokio::test]
+    async fn test_auth_secret_dedup_and_status() {
+        let caps = CapabilitiesFile::from_json(
+            r#"{
+                "auth": {
+                    "secret_name": "gh_token",
+                    "display_name": "GitHub"
+                },
+                "setup": {
+                    "required_secrets": [
+                        { "name": "gh_token", "prompt": "GitHub PAT" },
+                        { "name": "extra_key", "prompt": "Extra API Key" }
+                    ]
+                },
+                "http": {
+                    "allowlist": [{ "host": "api.github.com" }],
+                    "credentials": {
+                        "github": {
+                            "secret_name": "gh_token",
+                            "location": { "type": "bearer" },
+                            "host_patterns": ["api.github.com"]
+                        }
+                    }
+                },
+                "secrets": {
+                    "allowed_names": ["gh_token", "gh_*"]
+                }
+            }"#,
+        )
+        .unwrap();
+
+        let collected = collect_auth_secrets(&caps);
+
+        // gh_token should appear once (from auth), with location merged from credentials.
+        // extra_key should appear once (from setup).
+        assert_eq!(collected.secrets.len(), 2);
+        let gh = collected
+            .secrets
+            .iter()
+            .find(|s| s.secret_name == "gh_token")
+            .unwrap();
+        assert_eq!(gh.description.as_deref(), Some("GitHub"));
+        assert!(
+            gh.location.is_some(),
+            "location should be merged from http.credentials"
+        );
+
+        let extra = collected
+            .secrets
+            .iter()
+            .find(|s| s.secret_name == "extra_key")
+            .unwrap();
+        assert_eq!(extra.description.as_deref(), Some("Extra API Key"));
+        assert!(extra.location.is_none());
+
+        // Secrets section should filter gh_token (in seen_names) but keep gh_* (wildcard).
+        let secrets = caps.secrets.as_ref().unwrap();
+        let extra_secrets: Vec<_> = secrets
+            .allowed_names
+            .iter()
+            .filter(|name| !collected.seen_names.contains(name.as_str()))
+            .collect();
+        assert_eq!(extra_secrets, vec!["gh_*"]);
+
+        // Verify store check: missing secret -> exists returns false.
+        let store = test_secrets_store();
+        assert!(!store.exists("default", "gh_token").await.unwrap());
+
+        // Store gh_token and verify it's found.
+        store
+            .create(
+                "default",
+                CreateSecretParams::new("gh_token", "ghp_test123"),
+            )
+            .await
+            .unwrap();
+        assert!(store.exists("default", "gh_token").await.unwrap());
+        // extra_key still missing.
+        assert!(!store.exists("default", "extra_key").await.unwrap());
+    }
+
+    /// No auth sections → collect_auth_secrets returns empty.
+    #[test]
+    fn test_collect_auth_secrets_empty_caps() {
+        let caps = CapabilitiesFile::default();
+        let collected = collect_auth_secrets(&caps);
+        assert!(collected.secrets.is_empty());
+        assert!(collected.seen_names.is_empty());
+    }
 }

src/config/agent.rs

@@ -1,6 +1,6 @@
 use std::time::Duration;
 
-use crate::config::helpers::{parse_bool_env, parse_option_env, parse_optional_env};
+use crate::config::helpers::{optional_env, parse_bool_env, parse_option_env, parse_optional_env};
 use crate::error::ConfigError;
 use crate::settings::Settings;
 
@@ -23,6 +23,8 @@ pub struct AgentConfig {
     pub max_cost_per_day_cents: Option<u64>,
     /// Maximum LLM/tool actions per hour. None = unlimited.
     pub max_actions_per_hour: Option<u64>,
+    /// Maximum daily LLM spend per user in cents. None = unlimited.
+    pub max_cost_per_user_per_day_cents: Option<u64>,
     /// Maximum tool-call iterations per agentic loop invocation. Default 50.
     pub max_tool_iterations: usize,
     /// When true, skip tool approval checks entirely. For benchmarks/CI.
@@ -31,6 +33,13 @@ pub struct AgentConfig {
     pub default_timezone: String,
     /// Maximum tokens per job (0 = unlimited).
     pub max_tokens_per_job: u64,
+    /// Whether the deployment is multi-tenant (multiple users sharing one
+    /// instance). Auto-detected from GATEWAY_USER_TOKENS presence.
+    pub multi_tenant: bool,
+    /// Maximum concurrent LLM calls per user. None = use default (4).
+    pub max_llm_concurrent_per_user: Option<usize>,
+    /// Maximum concurrent jobs per user. None = use default (3).
+    pub max_jobs_concurrent_per_user: Option<usize>,
 }
 
 impl AgentConfig {
@@ -49,10 +58,14 @@ impl AgentConfig {
             allow_local_tools: true,
             max_cost_per_day_cents: None,
             max_actions_per_hour: None,
+            max_cost_per_user_per_day_cents: None,
             max_tool_iterations: 10,
             auto_approve_tools: true,
             default_timezone: "UTC".to_string(),
             max_tokens_per_job: 0,
+            multi_tenant: false,
+            max_llm_concurrent_per_user: None,
+            max_jobs_concurrent_per_user: None,
         }
     }
 
@@ -87,6 +100,7 @@ impl AgentConfig {
             allow_local_tools: parse_bool_env("ALLOW_LOCAL_TOOLS", false)?,
             max_cost_per_day_cents: parse_option_env("MAX_COST_PER_DAY_CENTS")?,
             max_actions_per_hour: parse_option_env("MAX_ACTIONS_PER_HOUR")?,
+            max_cost_per_user_per_day_cents: parse_option_env("MAX_COST_PER_USER_PER_DAY_CENTS")?,
             max_tool_iterations: parse_optional_env(
                 "AGENT_MAX_TOOL_ITERATIONS",
                 settings.agent.max_tool_iterations,
@@ -112,6 +126,11 @@ impl AgentConfig {
                 "AGENT_MAX_TOKENS_PER_JOB",
                 settings.agent.max_tokens_per_job,
             )?,
+            // Auto-detected from GATEWAY_USER_TOKENS presence. Not a separate
+            // knob — multi-tenant mode is always implied by configuring user tokens.
+            multi_tenant: optional_env("GATEWAY_USER_TOKENS")?.is_some(),
+            max_llm_concurrent_per_user: parse_option_env("TENANT_MAX_LLM_CONCURRENT")?,
+            max_jobs_concurrent_per_user: parse_option_env("TENANT_MAX_JOBS_CONCURRENT")?,
         })
     }
 }

src/config/heartbeat.rs

@@ -21,6 +21,9 @@ pub struct HeartbeatConfig {
     pub quiet_hours_end: Option<u32>,
     /// Timezone for fire_at and quiet hours evaluation (IANA name).
     pub timezone: Option<String>,
+    /// When true, cycle through all users with routines. Auto-detected from
+    /// GATEWAY_USER_TOKENS or set explicitly via HEARTBEAT_MULTI_TENANT.
+    pub multi_tenant: bool,
 }
 
 impl Default for HeartbeatConfig {
@@ -34,6 +37,7 @@ impl Default for HeartbeatConfig {
             quiet_hours_start: None,
             quiet_hours_end: None,
             timezone: None,
+            multi_tenant: false,
         }
     }
 }
@@ -101,6 +105,12 @@ impl HeartbeatConfig {
                 }
                 tz
             },
+            // Auto-detect multi-tenant mode from GATEWAY_USER_TOKENS presence,
+            // or allow explicit override via HEARTBEAT_MULTI_TENANT.
+            multi_tenant: parse_bool_env(
+                "HEARTBEAT_MULTI_TENANT",
+                optional_env("GATEWAY_USER_TOKENS")?.is_some(),
+            )?,
         })
     }
 }

src/config/llm.rs

@@ -406,7 +406,7 @@ impl LlmConfig {
         // Resolve extra headers
         let extra_headers = if let Some(env_var) = extra_headers_env {
             optional_env(env_var)?
-                .map(|val| parse_extra_headers(&val))
+                .map(|val| parse_extra_headers_with_key(&val, env_var))
                 .transpose()?
                 .unwrap_or_default()
         } else {
@@ -475,7 +475,10 @@ impl LlmConfig {
 ///
 /// Format: `Key1:Value1,Key2:Value2` (colon-separated, not `=`, because
 /// header values often contain `=`).
-fn parse_extra_headers(val: &str) -> Result<Vec<(String, String)>, ConfigError> {
+fn parse_extra_headers_with_key(
+    val: &str,
+    env_var_name: &str,
+) -> Result<Vec<(String, String)>, ConfigError> {
     if val.trim().is_empty() {
         return Ok(Vec::new());
     }
@@ -488,14 +491,14 @@ fn parse_extra_headers(val: &str) -> Result<Vec<(String, String)>, ConfigError>
         }
         let Some((key, value)) = pair.split_once(':') else {
             return Err(ConfigError::InvalidValue {
-                key: "LLM_EXTRA_HEADERS".to_string(),
+                key: env_var_name.to_string(),
                 message: format!("malformed header entry '{}', expected Key:Value", pair),
             });
         };
         let key = key.trim();
         if key.is_empty() {
             return Err(ConfigError::InvalidValue {
-                key: "LLM_EXTRA_HEADERS".to_string(),
+                key: env_var_name.to_string(),
                 message: format!("empty header name in entry '{}'", pair),
             });
         }
@@ -536,6 +539,11 @@ mod tests {
     use crate::settings::Settings;
     use crate::testing::credentials::*;
 
+    /// Convenience wrapper for tests — uses "TEST_HEADERS" as the env var name.
+    fn parse_extra_headers(val: &str) -> Result<Vec<(String, String)>, ConfigError> {
+        parse_extra_headers_with_key(val, "TEST_HEADERS")
+    }
+
     /// Clear all openai-compatible-related env vars.
     fn clear_openai_compatible_env() {
         // SAFETY: Only called under ENV_MUTEX in tests.

src/config/mod.rs

@@ -312,13 +312,11 @@ impl Config {
         let tunnel = TunnelConfig::resolve(settings)?;
         let channels = ChannelsConfig::resolve(settings, &owner_id)?;
 
-        // Resolve workspace config using the gateway user_id for default layers.
+        // Resolve the startup workspace against the durable owner scope. The
-        let workspace_user_id = channels
+        // gateway may expose a distinct sender identity, but the base runtime
-            .gateway
+        // workspace stays owner-scoped and per-user gateway workspaces are
-            .as_ref()
+        // handled separately by WorkspacePool.
-            .map(|gw| gw.user_id.as_str())
+        let workspace = WorkspaceConfig::resolve(&owner_id)?;
-            .unwrap_or("default");
-        let workspace = WorkspaceConfig::resolve(workspace_user_id)?;
 
         Ok(Self {
             owner_id: owner_id.clone(),

src/context/state.rs

@@ -192,6 +192,9 @@ pub struct JobContext {
     /// but subsequent tools (e.g., `json`) may need the full output. This
     /// stash stores the complete, unsanitized output so tools can reference
     /// previous results by ID via `$tool_call_id` parameter syntax.
+    ///
+    /// Also used for cross-tool implicit state (keys prefixed with `__`) such
+    /// as `__routine_last_name` for fallback recovery in routine tool chains.
     #[serde(skip)]
     pub tool_output_stash: Arc<tokio::sync::RwLock<HashMap<String, String>>>,
     /// User's preferred timezone (IANA name, e.g. "America/New_York"). Defaults to "UTC".

src/db/libsql/routines.rs

@@ -462,6 +462,56 @@ impl RoutineStore for LibSqlBackend {
         Ok(counts)
     }
 
+    async fn batch_get_last_run_status(
+        &self,
+        routine_ids: &[Uuid],
+    ) -> Result<HashMap<Uuid, RunStatus>, DatabaseError> {
+        if routine_ids.is_empty() {
+            return Ok(HashMap::new());
+        }
+
+        let conn = self.connect().await?;
+
+        // SQLite doesn't support ANY($1), so we query all latest runs and filter in memory.
+        // Uses a subquery to pick only the most recent run per routine.
+        let mut rows = conn
+            .query(
+                "SELECT routine_id, status FROM routine_runs r1
+                 WHERE started_at = (
+                     SELECT MAX(started_at) FROM routine_runs r2
+                     WHERE r2.routine_id = r1.routine_id
+                 )
+                 GROUP BY routine_id",
+                params![],
+            )
+            .await
+            .map_err(|e| {
+                DatabaseError::Query(format!("Failed to batch get last run status: {}", e))
+            })?;
+
+        let routine_id_set: HashSet<Uuid> = routine_ids.iter().copied().collect();
+        let mut statuses = HashMap::new();
+
+        while let Some(row) = rows
+            .next()
+            .await
+            .map_err(|e| DatabaseError::Query(e.to_string()))?
+        {
+            let id_str: String = get_text(&row, 0);
+            let id = Uuid::parse_str(&id_str)
+                .map_err(|e| DatabaseError::Query(format!("Invalid routine UUID: {}", e)))?;
+
+            if routine_id_set.contains(&id) {
+                let status_str: String = get_text(&row, 1);
+                if let std::result::Result::Ok(status) = status_str.parse::<RunStatus>() {
+                    statuses.insert(id, status);
+                }
+            }
+        }
+
+        Ok(statuses)
+    }
+
     async fn link_routine_run_to_job(
         &self,
         run_id: Uuid,
@@ -480,10 +530,24 @@ impl RoutineStore for LibSqlBackend {
     async fn get_webhook_routine_by_path(
         &self,
         path: &str,
+        user_id: Option<&str>,
     ) -> Result<Option<Routine>, DatabaseError> {
         let conn = self.connect().await?;
-        let mut rows = conn
+        let mut rows = if let Some(uid) = user_id {
-            .query(
+            conn.query(
+                &format!(
+                    "SELECT {} FROM routines WHERE enabled = 1 AND trigger_type = 'webhook' \
+                     AND user_id = ?2 \
+                     AND (json_extract(trigger_config, '$.path') = ?1 \
+                     OR (json_extract(trigger_config, '$.path') IS NULL AND CAST(id AS TEXT) = ?1))",
+                    ROUTINE_COLUMNS
+                ),
+                params![path, uid],
+            )
+            .await
+            .map_err(|e| DatabaseError::Query(e.to_string()))?
+        } else {
+            conn.query(
                 &format!(
                     "SELECT {} FROM routines WHERE enabled = 1 AND trigger_type = 'webhook' \
                      AND (json_extract(trigger_config, '$.path') = ?1 \
@@ -493,7 +557,8 @@ impl RoutineStore for LibSqlBackend {
                 params![path],
             )
             .await
-            .map_err(|e| DatabaseError::Query(e.to_string()))?;
+            .map_err(|e| DatabaseError::Query(e.to_string()))?
+        };
 
         match rows
             .next()

src/db/mod.rs

@@ -528,6 +528,15 @@ pub trait RoutineStore: Send + Sync {
         &self,
         routine_ids: &[Uuid],
     ) -> Result<HashMap<Uuid, i64>, DatabaseError>;
+
+    /// Fetch the last run status for multiple routines in a single query.
+    /// Returns a map from routine_id to its most recent RunStatus.
+    /// Routines with no runs are omitted from the result.
+    async fn batch_get_last_run_status(
+        &self,
+        routine_ids: &[Uuid],
+    ) -> Result<HashMap<Uuid, RunStatus>, DatabaseError>;
+
     async fn link_routine_run_to_job(
         &self,
         run_id: Uuid,
@@ -536,6 +545,7 @@ pub trait RoutineStore: Send + Sync {
     async fn get_webhook_routine_by_path(
         &self,
         path: &str,
+        user_id: Option<&str>,
     ) -> Result<Option<Routine>, DatabaseError>;
 
     /// List routine runs that were dispatched as full_job but have not yet

src/db/postgres.rs

@@ -510,6 +510,14 @@ impl RoutineStore for PgBackend {
             .await
     }
 
+    async fn batch_get_last_run_status(
+        &self,
+        routine_ids: &[Uuid],
+    ) -> Result<std::collections::HashMap<Uuid, crate::agent::routine::RunStatus>, DatabaseError>
+    {
+        self.store.batch_get_last_run_status(routine_ids).await
+    }
+
     async fn link_routine_run_to_job(
         &self,
         run_id: Uuid,
@@ -521,8 +529,9 @@ impl RoutineStore for PgBackend {
     async fn get_webhook_routine_by_path(
         &self,
         path: &str,
+        user_id: Option<&str>,
     ) -> Result<Option<Routine>, DatabaseError> {
-        self.store.get_webhook_routine_by_path(path).await
+        self.store.get_webhook_routine_by_path(path, user_id).await
     }
 
     async fn list_dispatched_routine_runs(&self) -> Result<Vec<RoutineRun>, DatabaseError> {

src/extensions/manager.rs

@@ -53,22 +53,6 @@ struct HostedOAuthFlowStart {
     flow: crate::cli::oauth_defaults::PendingOAuthFlow,
 }
 
-fn hosted_proxy_client_secret(
-    client_secret: &Option<String>,
-    builtin: Option<&crate::cli::oauth_defaults::OAuthCredentials>,
-    exchange_proxy_configured: bool,
-) -> Option<String> {
-    if !exchange_proxy_configured {
-        return client_secret.clone();
-    }
-
-    let builtin_secret = builtin.map(|credentials| credentials.client_secret);
-    match (client_secret, builtin_secret) {
-        (Some(resolved), Some(baked_in)) if resolved == baked_in => None,
-        _ => client_secret.clone(),
-    }
-}
-
 fn normalize_oauth_callback_path(path: &str) -> String {
     let trimmed_path = path.trim_end_matches('/');
     if trimmed_path.is_empty() {
@@ -419,9 +403,10 @@ pub struct ExtensionManager {
     /// when running in gateway mode, consumed by the web gateway's
     /// `/oauth/callback` handler.
     pending_oauth_flows: crate::cli::oauth_defaults::PendingOAuthRegistry,
-    /// Gateway auth token for authenticating with the platform token exchange proxy.
+    /// OAuth proxy auth token for authenticating with the hosted token exchange proxy.
-    /// Read once at construction from `GATEWAY_AUTH_TOKEN` env var.
+    /// Resolved once at construction from `IRONCLAW_OAUTH_PROXY_AUTH_TOKEN`,
-    gateway_token: Option<String>,
+    /// then `GATEWAY_AUTH_TOKEN` as a backward-compatible fallback.
+    oauth_proxy_auth_token: Option<String>,
     /// Relay config captured at startup. Used by `auth_channel_relay` and
     /// `activate_channel_relay` instead of re-reading env vars.
     relay_config: Option<crate::config::RelayConfig>,
@@ -551,7 +536,7 @@ impl ExtensionManager {
             activation_errors: RwLock::new(HashMap::new()),
             sse_manager: RwLock::new(None),
             pending_oauth_flows: crate::cli::oauth_defaults::new_pending_oauth_registry(),
-            gateway_token: std::env::var("GATEWAY_AUTH_TOKEN").ok(),
+            oauth_proxy_auth_token: crate::cli::oauth_defaults::oauth_proxy_auth_token(),
             relay_config: crate::config::RelayConfig::from_env(),
             relay_event_tx: Arc::new(tokio::sync::Mutex::new(None)),
             relay_signing_secret_cache: Arc::new(std::sync::Mutex::new(None)),
@@ -675,6 +660,66 @@ impl ExtensionManager {
         })
     }
 
+    /// Resolve the relay URL override for an extension from settings.
+    ///
+    /// Returns `Some(url)` if a non-empty per-extension `relay_url` override is
+    /// set for the given extension; otherwise returns `None` and callers should
+    /// fall back to the env-level `RelayConfig`.
+    ///
+    /// Uses `self.user_id` (owner scope) for consistency with `configure()`,
+    /// which also writes setting_path fields under the owner scope.
+    ///
+    /// The override is validated: only `http` / `https` schemes are accepted
+    /// and the URL must not contain userinfo (embedded credentials).  This
+    /// prevents a malicious override from exfiltrating the instance-wide relay
+    /// API key to an attacker-controlled host.
+    async fn effective_relay_url(&self, name: &str) -> Option<String> {
+        if let Some(ref store) = self.store {
+            let key = format!("extensions.{name}.relay_url");
+            if let Ok(Some(v)) = store.get_setting(&self.user_id, &key).await {
+                let url = v
+                    .as_str()
+                    .map(|s| s.trim().to_string())
+                    .filter(|s| !s.is_empty());
+                if let Some(ref u) = url {
+                    // Validate the override to prevent API-key exfiltration:
+                    // only allow http(s) with no embedded credentials.
+                    match url::Url::parse(u) {
+                        Ok(parsed)
+                            if (parsed.scheme() == "http" || parsed.scheme() == "https")
+                                && parsed.username().is_empty()
+                                && parsed.password().is_none() =>
+                        {
+                            tracing::debug!(
+                                extension = %name,
+                                relay_url_host = %parsed.host_str().unwrap_or("unknown"),
+                                "effective_relay_url: using per-extension override from settings"
+                            );
+                            return url;
+                        }
+                        Ok(parsed) => {
+                            tracing::warn!(
+                                extension = %name,
+                                scheme = %parsed.scheme(),
+                                has_userinfo = !parsed.username().is_empty() || parsed.password().is_some(),
+                                "effective_relay_url: rejecting override — \
+                                 only http/https without embedded credentials is allowed"
+                            );
+                        }
+                        Err(e) => {
+                            tracing::warn!(
+                                extension = %name,
+                                error = %e,
+                                "effective_relay_url: rejecting override — invalid URL"
+                            );
+                        }
+                    }
+                }
+            }
+        }
+        None
+    }
+
     /// Get the shared relay event sender for the webhook endpoint.
     pub fn relay_event_tx(
         &self,
@@ -891,24 +936,67 @@ impl ExtensionManager {
         *self.relay_channel_manager.write().await = Some(channel_manager);
     }
 
-    /// Check if a channel name corresponds to a relay extension (has stored stream token
+    /// Check if a channel name corresponds to a relay extension (has stored team_id
     /// or is tracked in the installed relay extensions set).
     pub async fn is_relay_channel(&self, name: &str, user_id: &str) -> bool {
         // Check in-memory installed set first (supports no-store mode)
         if self.installed_relay_extensions.read().await.contains(name) {
             return true;
         }
-        // Then check for stored stream token
+        // Check for stored team_id (persisted across restarts by the OAuth callback)
-        self.secrets
+        if let Some(ref store) = self.store {
-            .exists(user_id, &format!("relay:{}:stream_token", name))
+            let key = format!("relay:{}:team_id", name);
-            .await
+            if let Ok(Some(v)) = store.get_setting(user_id, &key).await {
-            .unwrap_or(false)
+                return v.as_str().is_some_and(|s| !s.is_empty());
+            }
+        }
+        false
+    }
+
+    /// Check whether a stored `team_id` setting exists for the given relay extension.
+    ///
+    /// Unlike [`is_relay_channel`], this does **not** consult the in-memory
+    /// `installed_relay_extensions` set — it only looks at the persistent settings
+    /// store.  This distinction matters for `auth_channel_relay`: an extension can
+    /// be *installed* (present in the in-memory set) but not yet *authenticated*
+    /// (no OAuth completed, no team_id stored).
+    async fn has_stored_team_id(&self, name: &str, _user_id: &str) -> bool {
+        if let Some(ref store) = self.store {
+            let key = format!("relay:{}:team_id", name);
+            // Use owner scope (self.user_id) for consistency: the OAuth callback
+            // stores team_id under state.owner_id which maps to self.user_id.
+            match store.get_setting(&self.user_id, &key).await {
+                Ok(Some(v)) => {
+                    let has_id = v.as_str().is_some_and(|s| !s.is_empty());
+                    tracing::debug!(
+                        extension = %name,
+                        has_team_id = has_id,
+                        "has_stored_team_id: checked store"
+                    );
+                    return has_id;
+                }
+                Ok(None) => {
+                    tracing::debug!(
+                        extension = %name,
+                        "has_stored_team_id: no team_id setting found"
+                    );
+                }
+                Err(e) => {
+                    tracing::warn!(
+                        extension = %name,
+                        error = %e,
+                        "has_stored_team_id: failed to read from settings store"
+                    );
+                }
+            }
+        }
+        false
     }
 
     /// Restore persisted relay channels after startup.
     ///
     /// Loads the persisted active channel list, filters to relay types (those with
-    /// a stored stream token), and activates each via `activate_stored_relay()`.
+    /// a stored team_id setting), and activates each via `activate_stored_relay()`.
     /// Skips channels that are already active.
     ///
     /// Call this only after `set_relay_channel_manager()` or `set_channel_runtime()`.
@@ -1131,7 +1219,7 @@ impl ExtensionManager {
     /// Broadcast an extension status change to the web UI via SSE.
     async fn broadcast_extension_status(&self, name: &str, status: &str, message: Option<&str>) {
         if let Some(ref sse) = *self.sse_manager.read().await {
-            sse.broadcast(crate::channels::web::types::SseEvent::ExtensionStatus {
+            sse.broadcast(ironclaw_common::AppEvent::ExtensionStatus {
                 extension_name: name.to_string(),
                 status: status.to_string(),
                 message: message.map(|m| m.to_string()),
@@ -1428,9 +1516,11 @@ impl ExtensionManager {
         if kind_filter.is_none() || kind_filter == Some(ExtensionKind::ChannelRelay) {
             let installed = self.installed_relay_extensions.read().await;
             let active_names = self.active_channel_names.read().await;
+            let errors = self.activation_errors.read().await;
             for name in installed.iter() {
                 let active = active_names.contains(name);
-                let has_token = self.is_relay_channel(name, user_id).await;
+                let authenticated = self.has_stored_team_id(name, user_id).await;
+                let activation_error = errors.get(name).cloned();
                 let registry_entry = self
                     .registry
                     .get_with_kind(name, Some(ExtensionKind::ChannelRelay))
@@ -1443,13 +1533,13 @@ impl ExtensionManager {
                     display_name,
                     description,
                     url: None,
-                    authenticated: has_token,
+                    authenticated,
                     active,
                     tools: Vec::new(),
                     needs_setup: false,
                     has_auth: true,
                     installed: true,
-                    activation_error: None,
+                    activation_error,
                     version: None,
                 });
             }
@@ -1626,7 +1716,22 @@ impl ExtensionManager {
                 self.persist_active_channels(user_id).await;
                 self.activation_errors.write().await.remove(name);
 
-                // Remove stored stream token
+                // Remove stored team_id setting and clean up secrets
+                if let Some(ref store) = self.store
+                    && let Err(e) = store
+                        .delete_setting(user_id, &format!("relay:{}:team_id", name))
+                        .await
+                {
+                    tracing::warn!(error = %e, name, "Failed to delete relay team_id setting on removal");
+                }
+                if let Err(e) = self
+                    .secrets
+                    .delete(user_id, &format!("relay:{}:oauth_state", name))
+                    .await
+                {
+                    tracing::warn!(error = %e, name, "Failed to delete relay oauth_state secret on removal");
+                }
+                // Clean up legacy stream_token secret from pre-webhook installs
                 let _ = self
                     .secrets
                     .delete(user_id, &format!("relay:{}:stream_token", name))
@@ -2684,7 +2789,7 @@ impl ExtensionManager {
                 user_id: user_id.to_string(),
                 secrets: Arc::clone(&self.secrets),
                 sse_manager: self.sse_manager.read().await.clone(),
-                gateway_token: self.gateway_token.clone(),
+                gateway_token: self.oauth_proxy_auth_token.clone(),
                 token_exchange_extra_params,
                 client_id_secret_name: if server.oauth.is_none() {
                     Some(server.client_id_secret_name())
@@ -3179,7 +3284,7 @@ impl ExtensionManager {
             // apps. Sending the desktop secret would cause a client_id/secret
             // mismatch because the container's GOOGLE_OAUTH_CLIENT_ID is the web
             // app, not the desktop app.
-            let proxy_client_secret = hosted_proxy_client_secret(
+            let proxy_client_secret = oauth_defaults::hosted_proxy_client_secret(
                 &client_secret,
                 builtin.as_ref(),
                 oauth_defaults::exchange_proxy_url().is_some(),
@@ -3201,7 +3306,7 @@ impl ExtensionManager {
                 user_id: user_id.to_string(),
                 secrets: Arc::clone(&self.secrets),
                 sse_manager: self.sse_manager.read().await.clone(),
-                gateway_token: self.gateway_token.clone(),
+                gateway_token: self.oauth_proxy_auth_token.clone(),
                 token_exchange_extra_params: std::collections::HashMap::new(),
                 client_id_secret_name: None,
                 created_at: std::time::Instant::now(),
@@ -3284,7 +3389,7 @@ impl ExtensionManager {
                 }
                 .await;
 
-                // Broadcast SSE event
+                // Broadcast auth result event
                 let (success, message) = match result {
                     Ok(()) => (true, format!("{} authenticated successfully", display_name)),
                     Err(ref e) => (
@@ -3310,7 +3415,7 @@ impl ExtensionManager {
                 }
 
                 if let Some(ref sse) = sse_manager {
-                    sse.broadcast(crate::channels::web::types::SseEvent::AuthCompleted {
+                    sse.broadcast(ironclaw_common::AppEvent::AuthCompleted {
                         extension_name: ext_name,
                         success,
                         message,
@@ -4181,26 +4286,75 @@ impl ExtensionManager {
     ///
     /// For Slack: initiates OAuth flow (redirect-based).
     /// For Telegram: accepts a bot token, registers it with channel-relay,
-    /// and stores the returned stream token.
+    /// and stores the team_id setting.
     async fn auth_channel_relay(
         &self,
         name: &str,
         user_id: &str,
     ) -> Result<AuthResult, ExtensionError> {
-        // Check if already authenticated (stream token exists)
+        tracing::debug!(
-        if self.is_relay_channel(name, user_id).await {
+            extension = %name,
+            user_id = %user_id,
+            "auth_channel_relay: starting"
+        );
+
+        // Check if already authenticated by looking for a stored team_id.
+        // We intentionally skip the `installed_relay_extensions` in-memory set
+        // here because that set only tracks *installed* extensions — an extension
+        // can be installed (via registry) but not yet authenticated (no OAuth
+        // completed). Checking just `is_relay_channel()` would short-circuit
+        // to "authenticated" even when no team_id exists, preventing the OAuth
+        // flow from being offered to the user.
+        if self.has_stored_team_id(name, user_id).await {
+            tracing::debug!(
+                extension = %name,
+                "auth_channel_relay: already authenticated (team_id in store)"
+            );
             return Ok(AuthResult::authenticated(name, ExtensionKind::ChannelRelay));
         }
 
+        tracing::debug!(
+            extension = %name,
+            "auth_channel_relay: no stored team_id, initiating OAuth"
+        );
+
         // Use relay config captured at startup
-        let relay_config = self.relay_config()?;
+        let relay_config = self.relay_config().map_err(|e| {
+            tracing::warn!(
+                extension = %name,
+                error = %e,
+                "auth_channel_relay: relay config not available — \
+                 CHANNEL_RELAY_URL and CHANNEL_RELAY_API_KEY must be set"
+            );
+            e
+        })?;
+
+        // Allow per-extension URL override from settings
+        let effective_url = self
+            .effective_relay_url(name)
+            .await
+            .unwrap_or_else(|| relay_config.url.clone());
+
+        tracing::debug!(
+            extension = %name,
+            relay_url = %effective_url,
+            "auth_channel_relay: creating relay client for OAuth"
+        );
 
         let client = crate::channels::relay::RelayClient::new(
-            relay_config.url.clone(),
+            effective_url.clone(),
             relay_config.api_key.clone(),
             relay_config.request_timeout_secs,
         )
-        .map_err(|e| ExtensionError::Config(e.to_string()))?;
+        .map_err(|e| {
+            tracing::warn!(
+                extension = %name,
+                relay_url = %effective_url,
+                error = %e,
+                "auth_channel_relay: failed to create relay HTTP client"
+            );
+            ExtensionError::Config(e.to_string())
+        })?;
 
         // Generate CSRF nonce — IronClaw validates this on the callback to ensure
         // the OAuth completion is legitimate. Channel-relay embeds it in the signed
@@ -4212,18 +4366,44 @@ impl ExtensionManager {
         self.secrets
             .create(user_id, CreateSecretParams::new(&state_key, &state_nonce))
             .await
-            .map_err(|e| ExtensionError::AuthFailed(format!("Failed to store OAuth state: {e}")))?;
+            .map_err(|e| {
+                tracing::warn!(
+                    extension = %name,
+                    error = %e,
+                    "auth_channel_relay: failed to store OAuth state nonce"
+                );
+                ExtensionError::AuthFailed(format!("Failed to store OAuth state: {e}"))
+            })?;
 
         // Channel-relay derives all URLs from trusted instance_url in chat-api.
         // We only pass the nonce for CSRF validation on the callback.
+        tracing::debug!(
+            extension = %name,
+            relay_url = %effective_url,
+            "auth_channel_relay: calling initiate_oauth on channel-relay"
+        );
         match client.initiate_oauth(Some(&state_nonce)).await {
-            Ok(auth_url) => Ok(AuthResult::awaiting_authorization(
+            Ok(auth_url) => {
-                name,
+                tracing::info!(
-                ExtensionKind::ChannelRelay,
+                    extension = %name,
-                auth_url,
+                    "auth_channel_relay: OAuth URL obtained, awaiting user authorization"
-                "redirect".to_string(),
+                );
-            )),
+                Ok(AuthResult::awaiting_authorization(
-            Err(e) => Err(ExtensionError::AuthFailed(e.to_string())),
+                    name,
+                    ExtensionKind::ChannelRelay,
+                    auth_url,
+                    "redirect".to_string(),
+                ))
+            }
+            Err(e) => {
+                tracing::warn!(
+                    extension = %name,
+                    relay_url = %effective_url,
+                    error = %e,
+                    "auth_channel_relay: initiate_oauth call to channel-relay failed"
+                );
+                Err(ExtensionError::AuthFailed(e.to_string()))
+            }
         }
     }
 
@@ -4233,46 +4413,112 @@ impl ExtensionManager {
         name: &str,
         user_id: &str,
     ) -> Result<ActivateResult, ExtensionError> {
-        let token_key = format!("relay:{}:stream_token", name);
+        tracing::debug!(
+            extension = %name,
+            user_id = %user_id,
+            "activate_channel_relay: starting"
+        );
+
         let team_id_key = format!("relay:{}:team_id", name);
 
-        // Check if we have a stream token
+        // Get team_id from settings (stored by the OAuth callback)
-        // Verify auth: stream token must exist (even though we don't use it in this constructor path)
-        let _stream_token = match self.secrets.get_decrypted(user_id, &token_key).await {
-            Ok(secret) => secret.expose().to_string(),
-            Err(_) => {
-                return Err(ExtensionError::AuthRequired);
-            }
-        };
-
-        // Get team_id from settings
         let team_id = if let Some(ref store) = self.store {
-            store
+            match store.get_setting(user_id, &team_id_key).await {
-                .get_setting(user_id, &team_id_key)
+                Ok(Some(v)) => {
-                .await
+                    let id = v.as_str().map(|s| s.to_string()).unwrap_or_default();
-                .ok()
+                    tracing::debug!(
-                .flatten()
+                        extension = %name,
-                .and_then(|v| v.as_str().map(|s| s.to_string()))
+                        team_id_empty = id.is_empty(),
-                .unwrap_or_default()
+                        "activate_channel_relay: loaded team_id from store"
+                    );
+                    id
+                }
+                Ok(None) => {
+                    tracing::debug!(
+                        extension = %name,
+                        setting_key = %team_id_key,
+                        "activate_channel_relay: no team_id in settings store"
+                    );
+                    String::new()
+                }
+                Err(e) => {
+                    tracing::warn!(
+                        extension = %name,
+                        error = %e,
+                        "activate_channel_relay: failed to read team_id from settings store"
+                    );
+                    String::new()
+                }
+            }
         } else {
+            tracing::debug!(
+                extension = %name,
+                "activate_channel_relay: no settings store available"
+            );
             String::new()
         };
 
+        if team_id.is_empty() {
+            tracing::debug!(
+                extension = %name,
+                "activate_channel_relay: team_id is empty, returning AuthRequired"
+            );
+            return Err(ExtensionError::AuthRequired);
+        }
+
         // Use relay config captured at startup
-        let relay_config = self.relay_config()?;
+        let relay_config = self.relay_config().map_err(|e| {
+            tracing::warn!(
+                extension = %name,
+                error = %e,
+                "activate_channel_relay: relay config not available"
+            );
+            e
+        })?;
+
+        // Allow per-extension URL override from settings
+        let effective_url = self
+            .effective_relay_url(name)
+            .await
+            .unwrap_or_else(|| relay_config.url.clone());
+
+        tracing::debug!(
+            extension = %name,
+            relay_url = %effective_url,
+            "activate_channel_relay: relay config loaded"
+        );
 
         let instance_id = self.relay_instance_id(relay_config, user_id);
 
         let client = crate::channels::relay::RelayClient::new(
-            relay_config.url.clone(),
+            effective_url.clone(),
             relay_config.api_key.clone(),
             relay_config.request_timeout_secs,
         )
-        .map_err(|e| ExtensionError::ActivationFailed(e.to_string()))?;
+        .map_err(|e| {
+            tracing::warn!(
+                extension = %name,
+                relay_url = %effective_url,
+                error = %e,
+                "activate_channel_relay: failed to create relay HTTP client"
+            );
+            ExtensionError::ActivationFailed(e.to_string())
+        })?;
 
         // Fetch the per-instance signing secret from channel-relay.
         // This must succeed — there is no fallback.
+        tracing::debug!(
+            extension = %name,
+            relay_url = %effective_url,
+            "activate_channel_relay: fetching signing secret from channel-relay"
+        );
         let signing_secret = client.get_signing_secret(&team_id).await.map_err(|e| {
+            tracing::warn!(
+                extension = %name,
+                relay_url = %effective_url,
+                error = %e,
+                "activate_channel_relay: failed to fetch signing secret from channel-relay"
+            );
             ExtensionError::Config(format!("Failed to fetch relay signing secret: {e}"))
         })?;
 
@@ -4291,16 +4537,29 @@ impl ExtensionManager {
         // Hot-add to channel manager
         let cm_guard = self.relay_channel_manager.read().await;
         let channel_mgr = cm_guard.as_ref().ok_or_else(|| {
+            tracing::warn!(
+                extension = %name,
+                "activate_channel_relay: channel manager not initialized"
+            );
             ExtensionError::ActivationFailed("Channel manager not initialized".to_string())
         })?;
 
-        channel_mgr
+        channel_mgr.hot_add(Box::new(channel)).await.map_err(|e| {
-            .hot_add(Box::new(channel))
+            tracing::warn!(
-            .await
+                extension = %name,
-            .map_err(|e| ExtensionError::ActivationFailed(e.to_string()))?;
+                error = %e,
+                "activate_channel_relay: hot_add to channel manager failed"
+            );
+            ExtensionError::ActivationFailed(e.to_string())
+        })?;
 
         if let Ok(mut cache) = self.relay_signing_secret_cache.lock() {
             *cache = Some(signing_secret);
+        } else {
+            tracing::warn!(
+                extension = %name,
+                "activate_channel_relay: failed to cache signing secret (mutex poisoned)"
+            );
         }
 
         // Store the event sender so the web gateway's relay webhook endpoint can push events
@@ -4318,6 +4577,12 @@ impl ExtensionManager {
         self.broadcast_extension_status(name, "active", Some(&status_msg))
             .await;
 
+        tracing::info!(
+            extension = %name,
+            instance_id = %instance_id,
+            "activate_channel_relay: relay channel activated successfully"
+        );
+
         Ok(ActivateResult {
             name: name.to_string(),
             kind: ExtensionKind::ChannelRelay,
@@ -4367,11 +4632,11 @@ impl ExtensionManager {
             return Ok(ExtensionKind::WasmChannel);
         }
 
-        // Check channel-relay extensions (installed in memory or has stored token)
+        // Check channel-relay extensions (installed in memory or has stored team_id)
         if self.installed_relay_extensions.read().await.contains(name) {
             return Ok(ExtensionKind::ChannelRelay);
         }
-        // Also check if there's a stored stream token (persisted across restarts)
+        // Also check if there's a stored team_id setting (persisted across restarts)
         if self.is_relay_channel(name, user_id).await {
             return Ok(ExtensionKind::ChannelRelay);
         }
@@ -4597,6 +4862,41 @@ impl ExtensionManager {
                 }
                 Ok(ExtensionSetupSchema { secrets, fields })
             }
+            ExtensionKind::ChannelRelay => {
+                let relay_url_key = format!("extensions.{name}.relay_url");
+                let current_url = if let Some(ref store) = self.store {
+                    match store.get_setting(&self.user_id, &relay_url_key).await {
+                        Ok(value_opt) => value_opt
+                            .and_then(|v| v.as_str().map(|s| s.to_string()))
+                            .filter(|s| !s.is_empty()),
+                        Err(e) => {
+                            tracing::warn!(
+                                extension = %name,
+                                setting_key = %relay_url_key,
+                                error = %e,
+                                "get_setup_schema: failed to read relay_url from settings"
+                            );
+                            None
+                        }
+                    }
+                } else {
+                    None
+                };
+                let env_url = self.relay_config.as_ref().map(|c| c.url.as_str());
+                Ok(ExtensionSetupSchema {
+                    secrets: Vec::new(),
+                    fields: vec![crate::channels::web::types::SetupFieldInfo {
+                        name: "relay_url".to_string(),
+                        prompt: format!(
+                            "Channel-relay service URL (leave empty to use env default{})",
+                            env_url.map(|u| format!(": {u}")).unwrap_or_default()
+                        ),
+                        optional: true,
+                        provided: current_url.is_some(),
+                        input_type: crate::tools::wasm::ToolSetupFieldInputType::Text,
+                    }],
+                })
+            }
             _ => Ok(ExtensionSetupSchema {
                 secrets: Vec::new(),
                 fields: Vec::new(),
@@ -5000,9 +5300,15 @@ impl ExtensionManager {
                 (names, Vec::new())
             }
             ExtensionKind::ChannelRelay => {
-                let mut names = std::collections::HashSet::new();
+                let relay_fields = vec![crate::tools::wasm::ToolFieldSetupSchema {
-                names.insert(format!("relay:{}:stream_token", name));
+                    name: "relay_url".to_string(),
-                (names, Vec::new())
+                    prompt: "Channel-relay service URL override".to_string(),
+                    optional: true,
+                    setting_path: Some(format!("extensions.{name}.relay_url")),
+                    input_type: crate::tools::wasm::ToolSetupFieldInputType::Text,
+                    restart_required: false,
+                }];
+                (std::collections::HashSet::new(), relay_fields)
             }
         };
 
@@ -5094,13 +5400,28 @@ impl ExtensionManager {
                 )));
             }
             let trimmed = field_value.trim();
+            let field_def = setup_field_defs.get(field_name);
+
+            // Empty value on an optional field with a setting_path: clear the
+            // stored override so the system reverts to the env/default value.
             if trimmed.is_empty() {
+                if let Some(def) = field_def
+                    && def.optional
+                {
+                    stored_fields.remove(field_name);
+                    if let Some(setting_path) = &def.setting_path {
+                        Self::validate_setup_setting_path(name, setting_path)?;
+                        if let Some(store) = self.store.as_ref() {
+                            let _ = store.delete_setting(&self.user_id, setting_path).await;
+                        }
+                    }
+                }
                 continue;
             }
 
             stored_fields.insert(field_name.clone(), trimmed.to_string());
 
-            if let Some(field_def) = setup_field_defs.get(field_name) {
+            if let Some(field_def) = field_def {
                 if field_def.restart_required {
                     restart_required = true;
                 }
@@ -5434,7 +5755,9 @@ impl ExtensionManager {
                     .map_err(|e| ExtensionError::NotInstalled(e.to_string()))?;
                 server.token_secret_name()
             }
-            ExtensionKind::ChannelRelay => format!("relay:{}:stream_token", name),
+            ExtensionKind::ChannelRelay => {
+                return Err(ExtensionError::AuthRequired);
+            }
         };
 
         let mut secrets = std::collections::HashMap::new();
@@ -5702,7 +6025,7 @@ mod tests {
     use crate::extensions::manager::{
         ChannelRuntimeState, FallbackDecision, TelegramBindingData, TelegramBindingResult,
         TelegramOwnerBindingState, build_wasm_channel_runtime_config_updates,
-        combine_install_errors, fallback_decision, hosted_proxy_client_secret, infer_kind_from_url,
+        combine_install_errors, fallback_decision, infer_kind_from_url,
         normalize_hosted_callback_url, send_telegram_text_message,
         telegram_message_matches_verification_code,
     };
@@ -7043,7 +7366,7 @@ mod tests {
         let dir = tempfile::tempdir().expect("temp dir");
         let mgr = make_test_manager(None, dir.path().to_path_buf());
 
-        // No token stored → not a relay channel
+        // No store configured, no team_id → not a relay channel
         assert!(!mgr.is_relay_channel("slack-relay", "test").await);
     }
 
@@ -7062,6 +7385,39 @@ mod tests {
         );
     }
 
+    /// Regression: installed-but-not-authenticated relay must NOT short-circuit
+    /// `auth_channel_relay()` to "authenticated".  Previously, `auth_channel_relay`
+    /// called `is_relay_channel()` which checked the in-memory
+    /// `installed_relay_extensions` set; that returned `true` even when no team_id
+    /// existed in the store, so the OAuth URL was never offered.
+    #[tokio::test]
+    async fn test_auth_channel_relay_installed_without_team_id_is_not_authenticated() {
+        let dir = tempfile::tempdir().expect("temp dir");
+        let mgr = make_test_manager(None, dir.path().to_path_buf());
+
+        // Mark as installed (simulates clicking Install in the UI)
+        mgr.installed_relay_extensions
+            .write()
+            .await
+            .insert("slack-relay".to_string());
+
+        // Without a stored team_id, auth should NOT return authenticated.
+        // It should fail because relay config is missing (no CHANNEL_RELAY_URL),
+        // but the key assertion is that it does NOT return Ok(authenticated).
+        let result = mgr.auth_channel_relay("slack-relay", "test").await;
+        match result {
+            Ok(ref auth_result) if auth_result.is_authenticated() => {
+                panic!(
+                    "auth_channel_relay returned authenticated for installed-but-no-team-id relay; \
+                     expected either an OAuth URL or a config error"
+                );
+            }
+            _ => {
+                // Config error (no relay URL) or awaiting_authorization — both are correct
+            }
+        }
+    }
+
     #[tokio::test]
     async fn test_remove_relay_shuts_down_via_relay_channel_manager() {
         // Regression: remove() only checked channel_runtime for shutdown, missing
@@ -7862,19 +8218,13 @@ mod tests {
             .await
             .insert("test-relay".to_string());
 
-        // configure() should dispatch to activate_channel_relay(), not
+        // configure() with empty secrets should dispatch to
-        // activate_wasm_channel(). Both will fail (no runtime configured),
+        // activate_channel_relay(), not activate_wasm_channel(). Relay auth
-        // but the error should be about relay config, not WASM channels.
+        // is OAuth-only so there are no manual secrets to pass.
-        let mut secrets = std::collections::HashMap::new();
-        secrets.insert(
-            "relay:test-relay:stream_token".to_string(),
-            "tok".to_string(),
-        );
-
         let result = mgr
             .configure(
                 "test-relay",
-                &secrets,
+                &std::collections::HashMap::new(),
                 &std::collections::HashMap::new(),
                 "test",
             )
@@ -7886,7 +8236,6 @@ mod tests {
         );
 
         let result = result.unwrap();
-        // Activation will fail (no relay config), but secrets should still be stored
         assert!(
             !result.activated,
             "activation should fail without relay config"
@@ -7896,15 +8245,6 @@ mod tests {
             "error should not mention WASM — got: {}",
             result.message
         );
-
-        // Verify the secret was stored
-        assert!(
-            mgr.secrets
-                .exists("test", "relay:test-relay:stream_token")
-                .await
-                .unwrap_or(false),
-            "configure should have stored the relay stream token"
-        );
     }
     #[test]
     fn test_validation_failed_is_distinct_error_variant() {
@@ -7970,7 +8310,8 @@ mod tests {
         let builtin_ref = builtin.as_ref();
         let secret = Some(builtin_ref.unwrap().client_secret.to_string());
 
-        let result = hosted_proxy_client_secret(&secret, builtin_ref, true);
+        let result =
+            crate::cli::oauth_defaults::hosted_proxy_client_secret(&secret, builtin_ref, true);
         assert_eq!(
             result, None,
             "built-in desktop secret must be suppressed when the exchange proxy is configured"
@@ -7982,7 +8323,8 @@ mod tests {
         let builtin = crate::cli::oauth_defaults::builtin_credentials("google_oauth_token");
         let secret = Some("user-entered-custom-secret".to_string());
 
-        let result = hosted_proxy_client_secret(&secret, builtin.as_ref(), true);
+        let result =
+            crate::cli::oauth_defaults::hosted_proxy_client_secret(&secret, builtin.as_ref(), true);
         assert_eq!(
             result,
             Some("user-entered-custom-secret".to_string()),
@@ -7996,7 +8338,8 @@ mod tests {
         let builtin_ref = builtin.as_ref();
         let secret = Some(builtin_ref.unwrap().client_secret.to_string());
 
-        let result = hosted_proxy_client_secret(&secret, builtin_ref, false);
+        let result =
+            crate::cli::oauth_defaults::hosted_proxy_client_secret(&secret, builtin_ref, false);
         assert_eq!(
             result, secret,
             "built-in secret must be kept when the callback will exchange directly"
@@ -8007,7 +8350,8 @@ mod tests {
     fn test_proxy_client_secret_none_stays_none() {
         let builtin = crate::cli::oauth_defaults::builtin_credentials("google_oauth_token");
 
-        let result = hosted_proxy_client_secret(&None, builtin.as_ref(), true);
+        let result =
+            crate::cli::oauth_defaults::hosted_proxy_client_secret(&None, builtin.as_ref(), true);
         assert_eq!(
             result, None,
             "None secret stays None even when the exchange proxy is configured"
@@ -8021,7 +8365,8 @@ mod tests {
         assert!(builtin.is_none());
 
         let secret = Some("dcr-secret".to_string());
-        let result = hosted_proxy_client_secret(&secret, builtin.as_ref(), true);
+        let result =
+            crate::cli::oauth_defaults::hosted_proxy_client_secret(&secret, builtin.as_ref(), true);
         assert_eq!(
             result,
             Some("dcr-secret".to_string()),

src/history/store.rs

@@ -1162,15 +1162,25 @@ impl Store {
     pub async fn get_webhook_routine_by_path(
         &self,
         path: &str,
+        user_id: Option<&str>,
     ) -> Result<Option<Routine>, DatabaseError> {
         let conn = self.conn().await?;
-        let row = conn
+        let row = if let Some(uid) = user_id {
-            .query_opt(
+            conn.query_opt(
+                "SELECT * FROM routines WHERE enabled AND trigger_type = 'webhook' \
+                 AND user_id = $2 \
+                 AND (trigger_config->>'path' = $1 OR (trigger_config->>'path' IS NULL AND id::text = $1))",
+                &[&path, &uid],
+            )
+            .await?
+        } else {
+            conn.query_opt(
                 "SELECT * FROM routines WHERE enabled AND trigger_type = 'webhook' \
                  AND (trigger_config->>'path' = $1 OR (trigger_config->>'path' IS NULL AND id::text = $1))",
                 &[&path],
             )
-            .await?;
+            .await?
+        };
         row.as_ref().map(row_to_routine).transpose()
     }
 
@@ -1403,6 +1413,40 @@ impl Store {
         Ok(counts)
     }
 
+    /// Batch-load the most recent run status for multiple routines in a single query.
+    /// Uses a window function to pick only the latest run per routine.
+    #[cfg(feature = "postgres")]
+    pub async fn batch_get_last_run_status(
+        &self,
+        routine_ids: &[Uuid],
+    ) -> Result<HashMap<Uuid, RunStatus>, DatabaseError> {
+        if routine_ids.is_empty() {
+            return Ok(HashMap::new());
+        }
+
+        let conn = self.conn().await?;
+        let rows = conn
+            .query(
+                "SELECT DISTINCT ON (routine_id) routine_id, status
+                 FROM routine_runs
+                 WHERE routine_id = ANY($1)
+                 ORDER BY routine_id, started_at DESC",
+                &[&routine_ids],
+            )
+            .await?;
+
+        let mut statuses = HashMap::new();
+        for row in rows {
+            let id: Uuid = row.get("routine_id");
+            let status_str: String = row.get("status");
+            if let std::result::Result::Ok(status) = status_str.parse::<RunStatus>() {
+                statuses.insert(id, status);
+            }
+        }
+
+        Ok(statuses)
+    }
+
     /// Link a routine run to a dispatched job.
     pub async fn link_routine_run_to_job(
         &self,

src/lib.rs

@@ -69,6 +69,7 @@ pub mod service;
 pub mod settings;
 pub mod setup;
 pub mod skills;
+pub mod tenant;
 pub mod timezone;
 pub mod tools;
 pub mod tracing_fmt;

src/llm/anthropic_oauth.rs

@@ -575,6 +575,7 @@ fn extract_response_content(response: &AnthropicResponse) -> (Option<String>, Ve
                     id: id.clone(),
                     name: name.clone(),
                     arguments: input.clone(),
+                    reasoning: None,
                 });
             }
         }
@@ -623,6 +624,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"q": "test"}),
+            reasoning: None,
         }];
         let messages = vec![
             ChatMessage::user("Search for test"),

src/llm/bedrock.rs

@@ -522,6 +522,7 @@ fn extract_content_blocks(
                     id: tu.tool_use_id().to_string(),
                     name: tu.name().to_string(),
                     arguments: document_to_json(tu.input()),
+                    reasoning: None,
                 });
             }
             // Ignore reasoning, citations, images, etc.
@@ -759,11 +760,13 @@ mod tests {
             id: "call_1".to_string(),
             name: "echo".to_string(),
             arguments: serde_json::json!({"text": "hi"}),
+            reasoning: None,
         };
         let tc2 = crate::llm::provider::ToolCall {
             id: "call_2".to_string(),
             name: "time".to_string(),
             arguments: serde_json::json!({}),
+            reasoning: None,
         };
 
         let messages = vec![
@@ -802,6 +805,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"query": "test"}),
+            reasoning: None,
         };
 
         let messages = vec![
@@ -825,6 +829,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "echo".to_string(),
             arguments: serde_json::json!({}),
+            reasoning: None,
         };
 
         let messages = vec![
@@ -989,11 +994,13 @@ mod tests {
             id: "call_abc".to_string(),
             name: "get_weather".to_string(),
             arguments: serde_json::json!({"city": "NYC"}),
+            reasoning: None,
         };
         let tc2 = crate::llm::provider::ToolCall {
             id: "call_def".to_string(),
             name: "get_time".to_string(),
             arguments: serde_json::json!({"tz": "EST"}),
+            reasoning: None,
         };
 
         let messages = vec![

src/llm/codex_chatgpt.rs

@@ -732,6 +732,7 @@ impl LlmProvider for CodexChatGptProvider {
                     id: tc.call_id,
                     name: tc.name,
                     arguments: args,
+                    reasoning: None,
                 }
             })
             .collect();
@@ -825,6 +826,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "search".to_string(),
             arguments: json!({"query": "rust"}),
+            reasoning: None,
         };
         let msg = ChatMessage::assistant_with_tool_calls(Some("thinking...".into()), vec![tc]);
         let items = CodexChatGptProvider::message_to_input_items(&msg);

src/llm/gemini_oauth.rs

@@ -1898,6 +1898,7 @@ impl GeminiOauthProvider {
                         id,
                         name,
                         arguments: args,
+                        reasoning: None,
                     });
                 }
             }

src/llm/github_copilot.rs

@@ -107,14 +107,21 @@ impl GithubCopilotProvider {
         body: &impl Serialize,
     ) -> Result<R, LlmError> {
         let url = self.api_url();
-        // Map token exchange failures to RequestFailed (retryable) rather than
+        // Distinguish permanent auth errors (non-retryable) from transient
-        // AuthFailed (non-retryable), since transient network errors during
+        // network failures (retryable) so RetryProvider handles them correctly.
-        // exchange should be retried by RetryProvider.
         let token = self.token_manager.get_token().await.map_err(|e| {
             tracing::warn!(error = %e, "Copilot: token exchange failed");
-            LlmError::RequestFailed {
+            match &e {
-                provider: "github_copilot".to_string(),
+                crate::llm::github_copilot_auth::GithubCopilotAuthError::AccessDenied
-                reason: format!("Token exchange failed: {e}"),
+                | crate::llm::github_copilot_auth::GithubCopilotAuthError::Expired => {
+                    LlmError::AuthFailed {
+                        provider: "github_copilot".to_string(),
+                    }
+                }
+                _ => LlmError::RequestFailed {
+                    provider: "github_copilot".to_string(),
+                    reason: format!("Token exchange failed: {e}"),
+                },
             }
         })?;
 
@@ -157,54 +164,14 @@ impl GithubCopilotProvider {
             );
 
             if status.as_u16() == 401 {
-                // Invalidate the cached session token and retry once with a
+                // Invalidate the cached session token so the next attempt
-                // fresh exchange — stale tokens are the most common 401 cause.
+                // (driven by RetryProvider) gets a fresh one. We don't retry
-                tracing::warn!("Copilot: 401 Unauthorized — invalidating session token, retrying");
+                // inline to avoid nested retries with the outer RetryProvider.
+                tracing::warn!("Copilot: 401 Unauthorized — invalidating session token for retry");
                 self.token_manager.invalidate().await;
-                let fresh = self.token_manager.get_token().await.map_err(|e| {
+                return Err(LlmError::RequestFailed {
-                    tracing::warn!(error = %e, "Copilot: re-exchange after 401 failed");
-                    LlmError::RequestFailed {
-                        provider: "github_copilot".to_string(),
-                        reason: format!("Token re-exchange after 401 failed: {e}"),
-                    }
-                })?;
-                let mut retry_req = self
-                    .client
-                    .post(&url)
-                    .bearer_auth(fresh.expose_secret())
-                    .header("Content-Type", "application/json");
-                for (key, value) in &self.extra_headers {
-                    retry_req = retry_req.header(key.as_str(), value.as_str());
-                }
-                let retry =
-                    retry_req
-                        .json(body)
-                        .send()
-                        .await
-                        .map_err(|e| LlmError::RequestFailed {
-                            provider: "github_copilot".to_string(),
-                            reason: format!("Retry after 401 failed: {e}"),
-                        })?;
-                if retry.status().is_success() {
-                    let text = retry.text().await.map_err(|e| LlmError::RequestFailed {
-                        provider: "github_copilot".to_string(),
-                        reason: format!("Failed to read retry response body: {e}"),
-                    })?;
-                    return serde_json::from_str(&text).map_err(|e| {
-                        let truncated = crate::agent::truncate_for_preview(&text, 512);
-                        LlmError::InvalidResponse {
-                            provider: "github_copilot".to_string(),
-                            reason: format!("JSON parse error: {e}. Raw: {truncated}"),
-                        }
-                    });
-                }
-                let retry_status = retry.status();
-                tracing::warn!(
-                    status = %retry_status,
-                    "Copilot: 401 retry also failed"
-                );
-                return Err(LlmError::AuthFailed {
                     provider: "github_copilot".to_string(),
+                    reason: "HTTP 401 Unauthorized".to_string(),
                 });
             }
             if status.as_u16() == 429 {
@@ -629,6 +596,7 @@ fn extract_choice_content(choice: &OpenAiChoice) -> (Option<String>, Vec<ToolCal
                     name: tc.function.name.clone(),
                     arguments: serde_json::from_str(&tc.function.arguments)
                         .unwrap_or(serde_json::Value::Object(serde_json::Map::new())),
+                    reasoning: None,
                 })
                 .collect()
         })
@@ -661,6 +629,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"q": "test"}),
+            reasoning: None,
         }];
         let messages = vec![
             ChatMessage::user("Search"),

src/llm/mod.rs

@@ -63,7 +63,8 @@ pub use provider::{
 };
 pub use reasoning::{
     ActionPlan, Reasoning, ReasoningContext, RespondOutput, RespondResult, SILENT_REPLY_TOKEN,
-    TOOL_INTENT_NUDGE, TokenUsage, ToolSelection, is_silent_reply, llm_signals_tool_intent,
+    TOOL_INTENT_NUDGE, TRUNCATED_TOOL_CALL_NOTICE, TokenUsage, ToolSelection, is_silent_reply,
+    llm_signals_tool_intent,
 };
 pub use recording::RecordingLlm;
 pub use registry::{ProviderDefinition, ProviderProtocol, ProviderRegistry};

src/llm/nearai_chat.rs

@@ -463,8 +463,15 @@ impl LlmProvider for NearAiChatProvider {
         let model = req.model.unwrap_or_else(|| self.active_model_name());
         let mut raw_messages = req.messages;
         crate::llm::provider::sanitize_tool_messages(&mut raw_messages);
-        let messages: Vec<ChatCompletionMessage> =
+        let raw: Vec<ChatCompletionMessage> = raw_messages.into_iter().map(|m| m.into()).collect();
-            raw_messages.into_iter().map(|m| m.into()).collect();
+
+        // NEAR AI rejects `role:"tool"` messages even on text-only completion paths.
+        // Apply the same flattening used by complete_with_tools().
+        let messages = if self.flatten_tool_messages {
+            flatten_tool_messages(raw)
+        } else {
+            raw
+        };
 
         let request = ChatCompletionRequest {
             model,
@@ -580,6 +587,7 @@ impl LlmProvider for NearAiChatProvider {
                     id: tc.id,
                     name: tc.function.name,
                     arguments,
+                    reasoning: None,
                 }
             })
             .collect();
@@ -1173,11 +1181,13 @@ mod tests {
                 id: "call_1".to_string(),
                 name: "list_issues".to_string(),
                 arguments: serde_json::json!({"owner": "foo", "repo": "bar"}),
+                reasoning: None,
             },
             ToolCall {
                 id: "call_2".to_string(),
                 name: "search".to_string(),
                 arguments: serde_json::json!({"query": "test"}),
+                reasoning: None,
             },
         ];
 
@@ -1210,6 +1220,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "test".to_string(),
             arguments: serde_json::json!({"key": "value"}),
+            reasoning: None,
         };
         let msg = ChatMessage::assistant_with_tool_calls(None, vec![tc]);
         let chat_msg: ChatCompletionMessage = msg.into();
@@ -1453,6 +1464,7 @@ mod tests {
                     id: tc.id,
                     name: tc.function.name,
                     arguments,
+                    reasoning: None,
                 }
             })
             .collect();
@@ -1502,6 +1514,7 @@ mod tests {
                     id: tc.id,
                     name: tc.function.name,
                     arguments,
+                    reasoning: None,
                 }
             })
             .collect();
@@ -2124,6 +2137,7 @@ mod tests {
                 id: "call_1".to_string(),
                 name: "test".to_string(),
                 arguments: serde_json::json!({}),
+                reasoning: None,
             }],
         );
         let chat_msg: ChatCompletionMessage = msg.into();
@@ -2193,6 +2207,65 @@ mod tests {
         assert_eq!(deserialized.function.arguments, r#"{"city":"London"}"#);
     }
 
+    // -- flatten_tool_messages in complete() path ----------------------------
+
+    #[test]
+    fn test_flatten_applied_on_text_only_path() {
+        // Verify that flatten_tool_messages converts tool-role messages to user
+        // messages (mirrors the complete_with_tools path).
+        let messages = vec![
+            ChatCompletionMessage {
+                role: "user".to_string(),
+                content: Some(MessageContent::Text("run it".to_string())),
+                tool_call_id: None,
+                name: None,
+                tool_calls: None,
+            },
+            ChatCompletionMessage {
+                role: "tool".to_string(),
+                content: Some(MessageContent::Text("ok".to_string())),
+                tool_call_id: Some("call_1".to_string()),
+                name: Some("run_cmd".to_string()),
+                tool_calls: None,
+            },
+        ];
+        let flattened = flatten_tool_messages(messages);
+        assert_eq!(flattened.len(), 2);
+        assert_eq!(flattened[1].role, "user");
+        let text = flattened[1]
+            .content
+            .as_ref()
+            .and_then(|c| c.as_text())
+            .unwrap();
+        assert!(text.contains("run_cmd"), "should reference tool name");
+        assert!(text.contains("ok"), "should include tool result");
+    }
+
+    #[test]
+    fn test_no_flatten_when_no_tool_messages() {
+        // When there are no tool-role messages, flatten_tool_messages is a no-op.
+        let messages = vec![
+            ChatCompletionMessage {
+                role: "user".to_string(),
+                content: Some(MessageContent::Text("hi".to_string())),
+                tool_call_id: None,
+                name: None,
+                tool_calls: None,
+            },
+            ChatCompletionMessage {
+                role: "assistant".to_string(),
+                content: Some(MessageContent::Text("hello".to_string())),
+                tool_call_id: None,
+                name: None,
+                tool_calls: None,
+            },
+        ];
+        let result = flatten_tool_messages(messages);
+        // No tool messages → unchanged roles
+        assert_eq!(result[0].role, "user");
+        assert_eq!(result[1].role, "assistant");
+    }
+
     // -- api_url edge cases ---------------------------------------------------
 
     #[test]

src/llm/openai_codex_provider.rs

@@ -625,6 +625,7 @@ fn parse_sse_response(body: &str) -> Result<ParsedResponse, LlmError> {
                                 id: state.call_id,
                                 name: state.name,
                                 arguments,
+                                reasoning: None,
                             });
                         } else {
                             // Fallback: extract directly from the item
@@ -650,6 +651,7 @@ fn parse_sse_response(body: &str) -> Result<ParsedResponse, LlmError> {
                                 id: call_id,
                                 name,
                                 arguments,
+                                reasoning: None,
                             });
                         }
                     }
@@ -727,6 +729,7 @@ fn parse_sse_response(body: &str) -> Result<ParsedResponse, LlmError> {
                 id: state.call_id,
                 name: state.name,
                 arguments,
+                reasoning: None,
             });
         }
     }
@@ -822,11 +825,13 @@ mod tests {
                 id: "call_1".to_string(),
                 name: "search".to_string(),
                 arguments: serde_json::json!({"query": "test"}),
+                reasoning: None,
             },
             ToolCall {
                 id: "call_2".to_string(),
                 name: "read".to_string(),
                 arguments: serde_json::json!({"path": "/tmp"}),
+                reasoning: None,
             },
         ];
         let msg =

src/llm/provider.rs

@@ -231,6 +231,10 @@ pub struct ToolCall {
     pub id: String,
     pub name: String,
     pub arguments: serde_json::Value,
+    /// Optional reasoning for why this tool was chosen — supplied by the provider
+    /// or derived from the shared response content as a fallback.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub reasoning: Option<String>,
 }
 
 /// Generate a tool-call ID that satisfies all providers.
@@ -637,6 +641,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "echo".to_string(),
             arguments: serde_json::json!({}),
+            reasoning: None,
         };
         let mut messages = vec![
             ChatMessage::user("hello"),
@@ -680,6 +685,7 @@ mod tests {
             id: "call_1".to_string(),
             name: "echo".to_string(),
             arguments: serde_json::json!({}),
+            reasoning: None,
         };
         let mut messages = vec![
             ChatMessage::user("test"),
@@ -705,11 +711,13 @@ mod tests {
             id: "call_sel_1".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"q": "test"}),
+            reasoning: None,
         };
         let tc2 = ToolCall {
             id: "call_sel_2".to_string(),
             name: "http".to_string(),
             arguments: serde_json::json!({"url": "https://example.com"}),
+            reasoning: None,
         };
         let mut messages = vec![
             ChatMessage::system("You are a helpful assistant."),

src/llm/reasoning.rs

@@ -8,8 +8,8 @@ use serde::{Deserialize, Serialize};
 use crate::llm::error::LlmError;
 
 use crate::llm::{
-    ChatMessage, CompletionRequest, LlmProvider, Role, ToolCall, ToolCompletionRequest,
+    ChatMessage, CompletionRequest, FinishReason, LlmProvider, Role, ToolCall,
-    ToolDefinition,
+    ToolCompletionRequest, ToolDefinition,
 };
 
 /// Token the agent returns when it has nothing to say (e.g. in group chats).
@@ -23,6 +23,13 @@ You said you would perform an action, but you did not include any tool calls.\n\
 Do NOT describe what you intend to do — actually call the tool now.\n\
 Use the tool_calls mechanism to invoke the appropriate tool.";
 
+/// Notice injected when the LLM's response was truncated mid-tool-call,
+/// causing incomplete parameters. Tells the LLM to try a different approach.
+pub const TRUNCATED_TOOL_CALL_NOTICE: &str = "\
+Your previous response was truncated while generating tool call parameters. \
+The tool calls were discarded. Please try a different approach — \
+summarize or transform the data instead of echoing it verbatim in a tool call.";
+
 /// Seed value used as the second argument to `generate_tool_call_id` when
 /// recovering tool calls from malformed LLM text responses. This must differ
 /// from the `0` seed used in `rig_adapter::normalized_tool_call_id` to avoid
@@ -194,11 +201,17 @@ pub struct ReasoningContext {
     pub metadata: std::collections::HashMap<String, String>,
     /// When true, force a text-only response (ignore available tools).
     /// Used by the agentic loop to guarantee termination near the iteration limit.
+    /// Sticky: once set, never cleared within a loop invocation. Callers must
+    /// create a fresh `ReasoningContext` per `run_agentic_loop()` call.
     pub force_text: bool,
     /// Pre-built system prompt. When set, `respond_with_tools` uses this directly
     /// instead of calling `build_system_prompt_with_tools`. Allows callers to build
     /// the prompt once and reuse it across iterations.
     pub system_prompt: Option<String>,
+    /// Per-user model override. When set, completion requests use this model
+    /// instead of the provider's default. Only effective with providers that
+    /// support per-request model overrides (e.g. NearAI).
+    pub model_override: Option<String>,
 }
 
 impl ReasoningContext {
@@ -212,6 +225,7 @@ impl ReasoningContext {
             metadata: std::collections::HashMap::new(),
             force_text: false,
             system_prompt: None,
+            model_override: None,
         }
     }
 
@@ -344,6 +358,7 @@ pub enum RespondResult {
 pub struct RespondOutput {
     pub result: RespondResult,
     pub usage: TokenUsage,
+    pub finish_reason: FinishReason,
 }
 
 /// Reasoning engine for the agent.
@@ -525,17 +540,46 @@ impl Reasoning {
 
         let response = self.llm.complete_with_tools(request).await?;
 
-        let reasoning = response.content.unwrap_or_default();
+        // If the response was truncated, tool call parameters are likely incomplete.
+        // Return empty so the caller can fall through to respond_with_tools() which
+        // has a larger output token budget.
+        if response.finish_reason == FinishReason::Length {
+            tracing::warn!(
+                "select_tools response truncated (finish_reason=Length), \
+                 discarding potentially incomplete tool selections"
+            );
+            return Ok(vec![]);
+        }
+
+        let shared_reasoning = response
+            .content
+            .map(|c| {
+                let pre_truncated = truncate_at_tool_tags(&c);
+                clean_response(&pre_truncated)
+            })
+            .unwrap_or_default();
 
         let selections: Vec<ToolSelection> = response
             .tool_calls
             .into_iter()
-            .map(|tool_call| ToolSelection {
+            .map(|tool_call| {
-                tool_name: tool_call.name,
+                // Prefer per-tool reasoning if the provider supplied it,
-                parameters: tool_call.arguments,
+                // otherwise fall back to the shared response content.
-                reasoning: reasoning.clone(),
+                let rationale = tool_call
-                alternatives: vec![],
+                    .reasoning
-                tool_call_id: tool_call.id,
+                    .map(|r| {
+                        let pre_truncated = truncate_at_tool_tags(&r);
+                        clean_response(&pre_truncated)
+                    })
+                    .filter(|r| !r.trim().is_empty())
+                    .unwrap_or_else(|| shared_reasoning.clone());
+                ToolSelection {
+                    tool_name: tool_call.name,
+                    parameters: tool_call.arguments,
+                    reasoning: rationale,
+                    alternatives: vec![],
+                    tool_call_id: tool_call.id,
+                }
             })
             .collect();
 
@@ -653,6 +697,9 @@ Respond in JSON format:
                 .with_temperature(0.7)
                 .with_tool_choice("auto");
             request.metadata = context.metadata.clone();
+            if let Some(ref model) = context.model_override {
+                request.model = Some(model.clone());
+            }
 
             let response = self.llm.complete_with_tools(request).await?;
             let usage = TokenUsage {
@@ -664,15 +711,39 @@ Respond in JSON format:
 
             // If there were tool calls, return them for execution
             if !response.tool_calls.is_empty() {
+                let narrative = response.content.map(|c| {
+                    let pre_truncated = truncate_at_tool_tags(&c);
+                    clean_response(&pre_truncated)
+                });
+                // Populate per-tool reasoning from the shared narrative when the
+                // provider did not supply per-tool rationale.
+                let tool_calls: Vec<ToolCall> = response
+                    .tool_calls
+                    .into_iter()
+                    .map(|mut tc| {
+                        if tc.reasoning.as_ref().is_none_or(|r| r.trim().is_empty()) {
+                            tc.reasoning = narrative.as_ref().filter(|n| !n.is_empty()).cloned();
+                        } else {
+                            // Clean provider-supplied per-tool reasoning the same way
+                            // we clean the shared narrative (strip thinking/tool tags).
+                            tc.reasoning = tc
+                                .reasoning
+                                .map(|r| {
+                                    let pre_truncated = truncate_at_tool_tags(&r);
+                                    clean_response(&pre_truncated)
+                                })
+                                .filter(|r| !r.trim().is_empty());
+                        }
+                        tc
+                    })
+                    .collect();
                 return Ok(RespondOutput {
                     result: RespondResult::ToolCalls {
-                        tool_calls: response.tool_calls,
+                        tool_calls,
-                        content: response.content.map(|c| {
+                        content: narrative,
-                            let pre_truncated = truncate_at_tool_tags(&c);
-                            clean_response(&pre_truncated)
-                        }),
                     },
                     usage,
+                    finish_reason: response.finish_reason,
                 });
             }
 
@@ -700,6 +771,7 @@ Respond in JSON format:
                         },
                     },
                     usage,
+                    finish_reason: response.finish_reason,
                 });
             }
 
@@ -725,6 +797,7 @@ Respond in JSON format:
             Ok(RespondOutput {
                 result: RespondResult::Text(final_text),
                 usage,
+                finish_reason: response.finish_reason,
             })
         } else {
             // No tools, use simple completion
@@ -732,6 +805,9 @@ Respond in JSON format:
                 .with_max_tokens(4096)
                 .with_temperature(0.7);
             request.metadata = context.metadata.clone();
+            if let Some(ref model) = context.model_override {
+                request.model = Some(model.clone());
+            }
 
             let response = self.llm.complete(request).await?;
             let pre_truncated = truncate_at_tool_tags(&response.content);
@@ -753,6 +829,7 @@ Respond in JSON format:
                     cache_read_input_tokens: response.cache_read_input_tokens,
                     cache_creation_input_tokens: response.cache_creation_input_tokens,
                 },
+                finish_reason: response.finish_reason,
             })
         }
     }
@@ -1293,6 +1370,49 @@ fn is_inside_code(pos: usize, regions: &[CodeRegion]) -> bool {
     regions.iter().any(|r| pos >= r.start && pos < r.end)
 }
 
+/// Check whether a byte range overlaps any code region.
+fn overlaps_code_region(start: usize, end: usize, regions: &[CodeRegion]) -> bool {
+    regions.iter().any(|r| start < r.end && end > r.start)
+}
+
+/// Return the byte bounds of the line containing `pos`, excluding the trailing newline.
+fn line_bounds(text: &str, pos: usize) -> (usize, usize) {
+    let start = text[..pos].rfind('\n').map_or(0, |idx| idx + 1);
+    let end = text[pos..].find('\n').map_or(text.len(), |idx| pos + idx);
+    (start, end)
+}
+
+/// Only recover XML-style tool calls when they are isolated content outside
+/// markdown code and quote contexts. This avoids converting code examples or
+/// quoted snippets into executable tool calls.
+fn is_recoverable_tool_call_segment(
+    text: &str,
+    start: usize,
+    end: usize,
+    code_regions: &[CodeRegion],
+) -> bool {
+    if overlaps_code_region(start, end, code_regions) {
+        return false;
+    }
+
+    let (first_line_start, first_line_end) = line_bounds(text, start);
+    let first_line = &text[first_line_start..first_line_end];
+
+    if first_line.trim_start().starts_with('>') {
+        return false;
+    }
+
+    let (_, last_line_end) = line_bounds(text, end.saturating_sub(1));
+    let first_line_prefix = &text[first_line_start..start];
+    let last_line_suffix = &text[end..last_line_end];
+
+    if !first_line_prefix.trim().is_empty() || !last_line_suffix.trim().is_empty() {
+        return false;
+    }
+
+    true
+}
+
 /// Clean up LLM response by stripping model-internal tags and reasoning patterns.
 ///
 /// Some models (GLM-4.7, etc.) emit XML-tagged internal state like
@@ -1312,6 +1432,7 @@ fn recover_tool_calls_from_content(
 ) -> Vec<ToolCall> {
     let tool_names: std::collections::HashSet<&str> =
         available_tools.iter().map(|t| t.name.as_str()).collect();
+    let code_regions = find_code_regions(content);
     let mut calls = Vec::new();
 
     for (open, close) in &[
@@ -1320,15 +1441,23 @@ fn recover_tool_calls_from_content(
         ("<function_call>", "</function_call>"),
         ("<|function_call|>", "<|/function_call|>"),
     ] {
-        let mut remaining = content;
+        let mut search_from = 0;
-        while let Some(start) = remaining.find(open) {
+        while let Some(offset) = content[search_from..].find(open) {
+            let start = search_from + offset;
             let inner_start = start + open.len();
-            let after = &remaining[inner_start..];
+            let after = &content[inner_start..];
-            let Some(end) = after.find(close) else {
+            let Some(end_offset) = after.find(close) else {
                 break;
             };
-            let inner = after[..end].trim();
+            let end = inner_start + end_offset;
-            remaining = &after[end + close.len()..];
+            let segment_end = end + close.len();
+            search_from = segment_end;
+
+            if !is_recoverable_tool_call_segment(content, start, segment_end, &code_regions) {
+                continue;
+            }
+
+            let inner = content[inner_start..end].trim();
 
             if inner.is_empty() {
                 continue;
@@ -1350,6 +1479,7 @@ fn recover_tool_calls_from_content(
                     ),
                     name: name.to_string(),
                     arguments,
+                    reasoning: None,
                 });
                 continue;
             }
@@ -1364,6 +1494,7 @@ fn recover_tool_calls_from_content(
                     ),
                     name: name.to_string(),
                     arguments: serde_json::Value::Object(Default::default()),
+                    reasoning: None,
                 });
             }
         }
@@ -1401,6 +1532,7 @@ fn recover_tool_calls_from_content(
                         ),
                         name: name.to_string(),
                         arguments,
+                        reasoning: None,
                     });
                     remaining = &args_start[bracket_end + 1..];
                     continue;
@@ -1412,6 +1544,7 @@ fn recover_tool_calls_from_content(
                 id: super::provider::generate_tool_call_id(calls.len(), RECOVERED_TOOL_CALL_SEED),
                 name: name.to_string(),
                 arguments: serde_json::Value::Object(Default::default()),
+                reasoning: None,
             });
             remaining = after_name;
         }
@@ -2257,6 +2390,40 @@ That's my plan."#;
         assert_eq!(calls[0].name, "tool_list");
     }
 
+    #[test]
+    fn test_recover_tool_call_in_fenced_code_block_ignored() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "Here is the XML format:\n\n```xml\n<tool_call>tool_list</tool_call>\n```";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert!(calls.is_empty());
+    }
+
+    #[test]
+    fn test_recover_tool_call_in_inline_code_ignored() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "Use `<tool_call>tool_list</tool_call>` to illustrate the syntax.";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert!(calls.is_empty());
+    }
+
+    #[test]
+    fn test_recover_tool_call_in_blockquote_ignored() {
+        let tools = make_tools(&["tool_list"]);
+        let content = "The page replied:\n> <tool_call>tool_list</tool_call>";
+        let calls = recover_tool_calls_from_content(content, &tools);
+        assert!(calls.is_empty());
+    }
+
+    #[test]
+    fn test_recover_multiline_json_tool_call_on_own_line() {
+        let tools = make_tools(&["memory_search"]);
+        let content = "Let me check.\n\n<tool_call>\n{\"name\": \"memory_search\", \"arguments\": {\"query\": \"test\"}}\n</tool_call>\n\nDone.";
+        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"}));
+    }
+
     // ---- System prompt building tests (issue #565) ----
 
     fn make_test_reasoning() -> Reasoning {
@@ -3145,4 +3312,113 @@ That's my plan."#;
             "Text <function_call>{}</function_call> middle "
         );
     }
+
+    /// Verify that reasoning normalization strips thinking tags and tool tags
+    /// from per-tool reasoning, matching the cleaning applied to shared reasoning.
+    #[test]
+    fn test_reasoning_normalization_strips_thinking_tags() {
+        let raw = "<thinking>Let me consider...</thinking>Search memory for prior context";
+        let pre_truncated = truncate_at_tool_tags(raw);
+        let cleaned = clean_response(&pre_truncated);
+        assert!(!cleaned.contains("<thinking>"));
+        assert!(cleaned.contains("Search memory"));
+    }
+
+    #[test]
+    fn test_reasoning_normalization_strips_tool_tags() {
+        let raw = "Calling search <tool_call>{\"name\": \"search\"}";
+        let pre_truncated = truncate_at_tool_tags(raw);
+        let cleaned = clean_response(&pre_truncated);
+        assert!(!cleaned.contains("<tool_call>"));
+        assert!(cleaned.contains("Calling search"));
+    }
+
+    #[test]
+    fn test_reasoning_normalization_empty_after_cleaning() {
+        let raw = "<thinking>internal only</thinking>";
+        let pre_truncated = truncate_at_tool_tags(raw);
+        let cleaned = clean_response(&pre_truncated);
+        assert!(cleaned.trim().is_empty());
+    }
+
+    // ---- select_tools truncation guard ----
+
+    /// Mock provider that returns tool calls with a configurable finish_reason.
+    struct TruncatingLlm {
+        finish_reason: crate::llm::FinishReason,
+    }
+
+    #[async_trait::async_trait]
+    impl crate::llm::LlmProvider for TruncatingLlm {
+        fn model_name(&self) -> &str {
+            "truncating-stub"
+        }
+        fn cost_per_token(&self) -> (rust_decimal::Decimal, rust_decimal::Decimal) {
+            (rust_decimal::Decimal::ZERO, rust_decimal::Decimal::ZERO)
+        }
+        async fn complete(
+            &self,
+            _request: crate::llm::CompletionRequest,
+        ) -> Result<crate::llm::CompletionResponse, crate::llm::error::LlmError> {
+            unimplemented!()
+        }
+        async fn complete_with_tools(
+            &self,
+            _request: crate::llm::ToolCompletionRequest,
+        ) -> Result<crate::llm::ToolCompletionResponse, crate::llm::error::LlmError> {
+            Ok(crate::llm::ToolCompletionResponse {
+                content: Some("I'll write the report.".to_string()),
+                tool_calls: vec![ToolCall {
+                    id: "call_1".to_string(),
+                    name: "memory_write".to_string(),
+                    arguments: serde_json::json!({}),
+                    reasoning: None,
+                }],
+                input_tokens: 5000,
+                output_tokens: 1024,
+                finish_reason: self.finish_reason,
+                cache_read_input_tokens: 0,
+                cache_creation_input_tokens: 0,
+            })
+        }
+    }
+
+    #[tokio::test]
+    async fn test_select_tools_returns_empty_on_truncation() {
+        let llm = Arc::new(TruncatingLlm {
+            finish_reason: FinishReason::Length,
+        });
+        let reasoning = Reasoning::new(llm);
+        let mut ctx = ReasoningContext::new().with_message(ChatMessage::user("Write a report"));
+        ctx.available_tools.push(ToolDefinition {
+            name: "memory_write".to_string(),
+            description: "Write to memory".to_string(),
+            parameters: serde_json::json!({"type": "object"}),
+        });
+
+        let selections = reasoning.select_tools(&ctx).await.unwrap();
+        assert!(
+            selections.is_empty(),
+            "Truncated tool selections should be discarded (got {} selections)",
+            selections.len()
+        );
+    }
+
+    #[tokio::test]
+    async fn test_select_tools_returns_selections_when_not_truncated() {
+        let llm = Arc::new(TruncatingLlm {
+            finish_reason: FinishReason::ToolUse,
+        });
+        let reasoning = Reasoning::new(llm);
+        let mut ctx = ReasoningContext::new().with_message(ChatMessage::user("Write a report"));
+        ctx.available_tools.push(ToolDefinition {
+            name: "memory_write".to_string(),
+            description: "Write to memory".to_string(),
+            parameters: serde_json::json!({"type": "object"}),
+        });
+
+        let selections = reasoning.select_tools(&ctx).await.unwrap();
+        assert_eq!(selections.len(), 1);
+        assert_eq!(selections[0].tool_name, "memory_write");
+    }
 }

src/llm/rig_adapter.rs

@@ -490,6 +490,7 @@ fn extract_response(
                     id: tc.id.clone(),
                     name: tc.function.name.clone(),
                     arguments: tc.function.arguments.clone(),
+                    reasoning: None,
                 });
             }
             // Reasoning and Image variants are not mapped to IronClaw types
@@ -597,6 +598,30 @@ fn build_rig_request(
     })
 }
 
+/// Inject a per-request model override into the rig request's `additional_params`.
+///
+/// Rig-core bakes the model name at construction time inside each provider's
+/// `CompletionModel` implementation. The actual HTTP request body includes a
+/// `model` field set by the provider. Rig-core's `#[serde(flatten)]` on
+/// `additional_params` emits these fields AFTER the provider's own fields.
+/// Most API servers (Python, Go) use last-key-wins when deserializing
+/// duplicate JSON keys, so the injected `model` value takes effect.
+fn inject_model_override(rig_req: &mut RigRequest, model_override: Option<&str>) {
+    let Some(model) = model_override else {
+        return;
+    };
+    match rig_req.additional_params {
+        Some(ref mut params) => {
+            if let Some(obj) = params.as_object_mut() {
+                obj.insert("model".to_string(), serde_json::json!(model));
+            }
+        }
+        None => {
+            rig_req.additional_params = Some(serde_json::json!({ "model": model }));
+        }
+    }
+}
+
 #[async_trait]
 impl<M> LlmProvider for RigAdapter<M>
 where
@@ -631,15 +656,7 @@ where
         &self,
         mut request: CompletionRequest,
     ) -> Result<CompletionResponse, LlmError> {
-        if let Some(requested_model) = request.model.as_deref()
+        let model_override = request.model.take();
-            && requested_model != self.model_name.as_str()
-        {
-            tracing::warn!(
-                requested_model = requested_model,
-                active_model = %self.model_name,
-                "Per-request model override is not supported for this provider; using configured model"
-            );
-        }
 
         self.strip_unsupported_completion_params(&mut request);
 
@@ -647,7 +664,7 @@ where
         crate::llm::provider::sanitize_tool_messages(&mut messages);
         let (preamble, history) = convert_messages(&messages);
 
-        let rig_req = build_rig_request(
+        let mut rig_req = build_rig_request(
             preamble,
             history,
             Vec::new(),
@@ -657,6 +674,8 @@ where
             self.cache_retention,
         )?;
 
+        inject_model_override(&mut rig_req, model_override.as_deref());
+
         let response =
             self.model
                 .completion(rig_req)
@@ -694,15 +713,7 @@ where
         &self,
         mut request: ToolCompletionRequest,
     ) -> Result<ToolCompletionResponse, LlmError> {
-        if let Some(requested_model) = request.model.as_deref()
+        let model_override = request.model.take();
-            && requested_model != self.model_name.as_str()
-        {
-            tracing::warn!(
-                requested_model = requested_model,
-                active_model = %self.model_name,
-                "Per-request model override is not supported for this provider; using configured model"
-            );
-        }
 
         self.strip_unsupported_tool_params(&mut request);
 
@@ -715,7 +726,7 @@ where
         let tools = convert_tools(&request.tools);
         let tool_choice = convert_tool_choice(request.tool_choice.as_deref());
 
-        let rig_req = build_rig_request(
+        let mut rig_req = build_rig_request(
             preamble,
             history,
             tools,
@@ -725,6 +736,8 @@ where
             self.cache_retention,
         )?;
 
+        inject_model_override(&mut rig_req, model_override.as_deref());
+
         let response =
             self.model
                 .completion(rig_req)
@@ -880,6 +893,7 @@ mod tests {
             id: "Xt7mK9pQ2".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"query": "test"}),
+            reasoning: None,
         };
         let msg = ChatMessage::assistant_with_tool_calls(Some("thinking".to_string()), vec![tc]);
         let messages = vec![msg];
@@ -997,6 +1011,7 @@ mod tests {
             id: "".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"query": "test"}),
+            reasoning: None,
         };
         let messages = vec![ChatMessage::assistant_with_tool_calls(None, vec![tc])];
         let (_preamble, history) = convert_messages(&messages);
@@ -1028,6 +1043,7 @@ mod tests {
             id: "   ".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"query": "test"}),
+            reasoning: None,
         };
         let messages = vec![ChatMessage::assistant_with_tool_calls(None, vec![tc])];
         let (_preamble, history) = convert_messages(&messages);
@@ -1061,6 +1077,7 @@ mod tests {
             id: "".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"query": "test"}),
+            reasoning: None,
         };
         let assistant_msg = ChatMessage::assistant_with_tool_calls(None, vec![tc]);
         let tool_result_msg = ChatMessage {
@@ -1380,11 +1397,13 @@ mod tests {
             id: "call_a".to_string(),
             name: "search".to_string(),
             arguments: serde_json::json!({"q": "rust"}),
+            reasoning: None,
         };
         let tc2 = IronToolCall {
             id: "call_b".to_string(),
             name: "fetch".to_string(),
             arguments: serde_json::json!({"url": "https://example.com"}),
+            reasoning: None,
         };
         let assistant = ChatMessage::assistant_with_tool_calls(None, vec![tc1, tc2]);
         let result_a = ChatMessage::tool_result("call_a", "search", "search results");

src/main.rs

@@ -590,8 +590,6 @@ async fn async_main() -> anyhow::Result<()> {
 
     let mut gateway_url: Option<String> = None;
     let mut sse_manager: Option<std::sync::Arc<ironclaw::channels::web::sse::SseManager>> = None;
-    let mut _gateway_state: Option<std::sync::Arc<ironclaw::channels::web::server::GatewayState>> =
-        None;
     if let Some(ref gw_config) = config.channels.gateway {
         // Build multi-user auth state if user_tokens is configured, else single-user.
         let mut gw = if let Some(ref user_tokens) = gw_config.user_tokens {
@@ -613,6 +611,7 @@ async fn async_main() -> anyhow::Result<()> {
         } else {
             GatewayChannel::new(gw_config.clone())
         };
+        gw = gw.with_owner_scope(config.owner_id.clone());
         gw = gw.with_llm_provider(Arc::clone(&components.llm));
         if let Some(ref ws) = components.workspace {
             gw = gw.with_workspace(Arc::clone(ws));
@@ -729,7 +728,6 @@ async fn async_main() -> anyhow::Result<()> {
         // IMPORTANT: This must come after all `with_*` calls since `rebuild_state`
         // creates a new SseManager, which would orphan this sender.
         sse_manager = Some(Arc::clone(&gw.state().sse));
-        _gateway_state = Some(Arc::clone(gw.state()));
         channel_names.push("gateway".to_string());
         channels.add(Box::new(gw)).await;
     }
@@ -792,6 +790,14 @@ async fn async_main() -> anyhow::Result<()> {
         .register_message_tools(Arc::clone(&channels), components.extension_manager.clone())
         .await;
 
+    // Default user ID for extension operations (single-user mode).
+    let ext_user_id = config
+        .channels
+        .gateway
+        .as_ref()
+        .map(|g| g.user_id.clone())
+        .unwrap_or_else(|| "default".to_string());
+
     // Wire up channel runtime for hot-activation of WASM channels.
     if let Some(ref ext_mgr) = components.extension_manager
         && let Some((rt, ps, router)) = wasm_channel_runtime_state.take()
@@ -812,12 +818,6 @@ async fn async_main() -> anyhow::Result<()> {
 
         // Auto-activate WASM channels that were active in a previous session.
         // Relay channels are handled separately below via restore_relay_channels().
-        let ext_user_id = config
-            .channels
-            .gateway
-            .as_ref()
-            .map(|g| g.user_id.clone())
-            .unwrap_or_else(|| "default".to_string());
         let persisted = ext_mgr.load_persisted_active_channels(&ext_user_id).await;
         for name in &persisted {
             if active_at_startup.contains(name)
@@ -850,20 +850,14 @@ async fn async_main() -> anyhow::Result<()> {
         ext_mgr
             .set_relay_channel_manager(Arc::clone(&channels))
             .await;
-        let ext_user_id = config
-            .channels
-            .gateway
-            .as_ref()
-            .map(|g| g.user_id.clone())
-            .unwrap_or_else(|| "default".to_string());
         ext_mgr.restore_relay_channels(&ext_user_id).await;
     }
 
     // Wire SSE sender into extension manager for broadcasting status events.
     if let Some(ref ext_mgr) = components.extension_manager
-        && let Some(sse) = sse_manager
+        && let Some(ref sse) = sse_manager
     {
-        ext_mgr.set_sse_sender(sse).await;
+        ext_mgr.set_sse_sender(Arc::clone(sse)).await;
     }
 
     // Snapshot memory for trace recording before the agent starts
@@ -901,7 +895,7 @@ async fn async_main() -> anyhow::Result<()> {
         skills_config: config.skills.clone(),
         hooks: components.hooks,
         cost_guard: components.cost_guard,
-        sse_tx: None, // TODO: wire SseManager into scheduler (needs Sender<SseEvent> → Arc<SseManager> refactor)
+        sse_tx: sse_manager,
         http_interceptor,
         transcription: config.transcription.create_provider().map(|p| {
             Arc::new(ironclaw::llm::transcription::TranscriptionMiddleware::new(
@@ -919,6 +913,11 @@ async fn async_main() -> anyhow::Result<()> {
             ironclaw::agent::routine_engine::SandboxReadiness::DockerUnavailable
         },
         builder: components.builder,
+        llm_backend: config.llm.backend.clone(),
+        tenant_rates: Arc::new(ironclaw::tenant::TenantRateRegistry::new(
+            config.agent.max_llm_concurrent_per_user.unwrap_or(4),
+            config.agent.max_jobs_concurrent_per_user.unwrap_or(3),
+        )),
     };
 
     let channels_for_warnings = Arc::clone(&channels);

src/orchestrator/api.rs

@@ -14,7 +14,7 @@ use serde::{Deserialize, Serialize};
 use tokio::sync::{Mutex, broadcast};
 use uuid::Uuid;
 
-use crate::channels::web::types::SseEvent;
+use crate::channels::web::types::ToolDecisionDto;
 use crate::db::Database;
 use crate::llm::{CompletionRequest, LlmProvider, ToolCompletionRequest};
 use crate::orchestrator::auth::{TokenStore, worker_auth_middleware};
@@ -25,6 +25,7 @@ use crate::worker::api::{
     CompletionReport, CredentialResponse, JobDescription, ProxyCompletionRequest,
     ProxyCompletionResponse, ProxyToolCompletionRequest, ProxyToolCompletionResponse, StatusUpdate,
 };
+use ironclaw_common::AppEvent;
 
 /// A follow-up prompt queued for a Claude Code bridge.
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -41,7 +42,7 @@ pub struct OrchestratorState {
     pub token_store: TokenStore,
     /// Broadcast channel for job events (consumed by the web gateway SSE).
     /// Tuple: (job_id, user_id, event).
-    pub job_event_tx: Option<broadcast::Sender<(Uuid, String, SseEvent)>>,
+    pub job_event_tx: Option<broadcast::Sender<(Uuid, String, AppEvent)>>,
     /// 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.
@@ -50,6 +51,9 @@ pub struct OrchestratorState {
     pub secrets_store: Option<Arc<dyn SecretsStore + Send + Sync>>,
     /// User ID for secret lookups (single-tenant, typically "default").
     pub user_id: String,
+    /// In-memory cache of job_id → user_id for SSE scoping. Populated when
+    /// sandbox jobs are created, avoiding a DB round-trip on every job event.
+    pub job_owner_cache: Arc<std::sync::RwLock<HashMap<Uuid, String>>>,
 }
 
 /// The orchestrator's internal API server.
@@ -274,10 +278,10 @@ async fn job_event_handler(
         });
     }
 
-    // Convert to SSE event and broadcast
+    // Convert to app event and broadcast
     let job_id_str = job_id.to_string();
-    let sse_event = match payload.event_type.as_str() {
+    let app_event = match payload.event_type.as_str() {
-        "message" => SseEvent::JobMessage {
+        "message" => AppEvent::JobMessage {
             job_id: job_id_str,
             role: payload
                 .data
@@ -292,7 +296,7 @@ async fn job_event_handler(
                 .unwrap_or("")
                 .to_string(),
         },
-        "tool_use" => SseEvent::JobToolUse {
+        "tool_use" => AppEvent::JobToolUse {
             job_id: job_id_str,
             tool_name: payload
                 .data
@@ -306,7 +310,7 @@ async fn job_event_handler(
                 .cloned()
                 .unwrap_or(serde_json::Value::Null),
         },
-        "tool_result" => SseEvent::JobToolResult {
+        "tool_result" => AppEvent::JobToolResult {
             job_id: job_id_str,
             tool_name: payload
                 .data
@@ -321,7 +325,7 @@ async fn job_event_handler(
                 .unwrap_or("")
                 .to_string(),
         },
-        "result" => SseEvent::JobResult {
+        "result" => AppEvent::JobResult {
             job_id: job_id_str,
             status: payload
                 .data
@@ -341,7 +345,21 @@ async fn job_event_handler(
             // gain context/memory tracking capabilities.
             fallback_deliverable: payload.data.get("fallback_deliverable").cloned(),
         },
-        _ => SseEvent::JobStatus {
+        "reasoning" => {
+            let narrative = payload
+                .data
+                .get("narrative")
+                .and_then(|v| v.as_str())
+                .unwrap_or("")
+                .to_string();
+            let decisions = ToolDecisionDto::from_json_array(&payload.data["decisions"]);
+            AppEvent::JobReasoning {
+                job_id: job_id_str,
+                narrative,
+                decisions,
+            }
+        }
+        _ => AppEvent::JobStatus {
             job_id: job_id_str,
             message: payload
                 .data
@@ -353,22 +371,43 @@ async fn job_event_handler(
     };
 
     // Broadcast via the channel (if configured).
-    // Look up the job owner so the gateway can scope delivery per-user.
+    // Look up the job owner from the in-memory cache (populated at job creation).
     if let Some(ref tx) = state.job_event_tx {
-        let user_id = match state.store.as_ref() {
+        let cached_uid = state
-            Some(store) => store
+            .job_owner_cache
-                .get_sandbox_job(job_id)
+            .read()
-                .await
+            .unwrap_or_else(|e| e.into_inner())
-                .ok()
+            .get(&job_id)
-                .flatten()
+            .cloned();
-                .map(|j| j.user_id),
+
-            None => None,
+        let user_id = match cached_uid {
+            Some(uid) => uid,
+            None => {
+                // Cache miss: fall back to DB lookup and populate cache.
+                let uid = match state.store.as_ref() {
+                    Some(store) => store
+                        .get_sandbox_job(job_id)
+                        .await
+                        .ok()
+                        .flatten()
+                        .map(|j| j.user_id),
+                    None => None,
+                };
+                if let Some(ref uid) = uid {
+                    state
+                        .job_owner_cache
+                        .write()
+                        .unwrap_or_else(|e| e.into_inner())
+                        .insert(job_id, uid.clone());
+                }
+                uid.unwrap_or_default()
+            }
         };
-        if let Some(uid) = user_id {
+
-            let _ = tx.send((job_id, uid, sse_event));
+        if user_id.is_empty() {
+            let _ = tx.send((job_id, String::new(), app_event));
         } else {
-            // Fallback: broadcast globally (single-user mode or job not found).
+            let _ = tx.send((job_id, user_id, app_event));
-            let _ = tx.send((job_id, String::new(), sse_event));
         }
     }
 
@@ -496,6 +535,7 @@ mod tests {
             store: None,
             secrets_store: None,
             user_id: "default".to_string(),
+            job_owner_cache: Arc::new(std::sync::RwLock::new(HashMap::new())),
         }
     }
 
@@ -725,6 +765,7 @@ mod tests {
             store: None,
             secrets_store: Some(secrets_store),
             user_id: "default".to_string(),
+            job_owner_cache: Arc::new(std::sync::RwLock::new(HashMap::new())),
         };
 
         let router = OrchestratorApi::router(state);
@@ -760,6 +801,7 @@ mod tests {
             store: None,
             secrets_store: None,
             user_id: "default".to_string(),
+            job_owner_cache: Arc::new(std::sync::RwLock::new(HashMap::new())),
         };
 
         let job_id = Uuid::new_v4();
@@ -790,7 +832,7 @@ mod tests {
         // No store configured, so user_id falls back to empty string.
         assert_eq!(recv_uid, "");
         match event {
-            SseEvent::JobMessage {
+            AppEvent::JobMessage {
                 job_id: jid,
                 role,
                 content,
@@ -817,6 +859,7 @@ mod tests {
             store: None,
             secrets_store: None,
             user_id: "default".to_string(),
+            job_owner_cache: Arc::new(std::sync::RwLock::new(HashMap::new())),
         };
 
         let job_id = Uuid::new_v4();
@@ -844,7 +887,7 @@ mod tests {
 
         let (_recv_id, _recv_uid, event) = rx.recv().await.unwrap();
         match event {
-            SseEvent::JobToolUse { tool_name, .. } => {
+            AppEvent::JobToolUse { tool_name, .. } => {
                 assert_eq!(tool_name, "shell");
             }
             other => panic!("Expected JobToolUse, got {:?}", other),
@@ -865,6 +908,7 @@ mod tests {
             store: None,
             secrets_store: None,
             user_id: "default".to_string(),
+            job_owner_cache: Arc::new(std::sync::RwLock::new(HashMap::new())),
         };
 
         let job_id = Uuid::new_v4();
@@ -889,7 +933,7 @@ mod tests {
 
         let (_recv_id, _recv_uid, event) = rx.recv().await.unwrap();
         // Unknown event types fall through to JobStatus
-        assert!(matches!(event, SseEvent::JobStatus { .. }));
+        assert!(matches!(event, AppEvent::JobStatus { .. }));
     }
 
     // -- Status update test --

src/orchestrator/mod.rs

@@ -46,10 +46,10 @@ use std::sync::Arc;
 use tokio::sync::{Mutex, broadcast};
 use uuid::Uuid;
 
-use crate::channels::web::types::SseEvent;
 use crate::db::Database;
 use crate::llm::LlmProvider;
 use crate::secrets::SecretsStore;
+use ironclaw_common::AppEvent;
 
 /// Resolve the orchestrator port from the `ORCHESTRATOR_PORT` environment
 /// variable, falling back to 50051.
@@ -63,7 +63,7 @@ fn resolve_orchestrator_port() -> u16 {
 /// Result of orchestrator setup, containing all handles needed by the agent.
 pub struct OrchestratorSetup {
     pub container_job_manager: Option<Arc<ContainerJobManager>>,
-    pub job_event_tx: Option<broadcast::Sender<(Uuid, String, SseEvent)>>,
+    pub job_event_tx: Option<broadcast::Sender<(Uuid, String, AppEvent)>>,
     pub prompt_queue: Arc<Mutex<HashMap<Uuid, VecDeque<api::PendingPrompt>>>>,
     pub docker_status: crate::sandbox::DockerStatus,
 }
@@ -134,6 +134,7 @@ pub async fn setup_orchestrator(
             store: db.cloned(),
             secrets_store: secrets_store.cloned(),
             user_id: "default".to_string(),
+            job_owner_cache: Arc::new(std::sync::RwLock::new(std::collections::HashMap::new())),
         };
 
         tokio::spawn(async move {

src/settings.rs

@@ -1297,6 +1297,92 @@ mod tests {
         assert_eq!(loaded.heartbeat.interval_secs, 900);
     }
 
+    /// Regression: /model writes a single key ("selected_model") to the DB via
+    /// set_setting(). On restart, get_all_settings() returns ALL keys including
+    /// wizard-written defaults. The single-key update must survive the full
+    /// from_db_map() round trip.
+    #[test]
+    fn db_single_key_model_update_survives_roundtrip() {
+        // Step 1: Wizard writes full settings to DB (including selected_model
+        // from initial setup).
+        let wizard_settings = Settings {
+            llm_backend: Some("nearai".to_string()),
+            selected_model: Some("old-wizard-model".to_string()),
+            ..Default::default()
+        };
+        let mut db: std::collections::HashMap<String, serde_json::Value> =
+            wizard_settings.to_db_map();
+
+        // Step 2: User runs /model new-model — persist_selected_model writes
+        // a single key, overwriting the wizard value.
+        db.insert(
+            "selected_model".to_string(),
+            serde_json::Value::String("new-model".to_string()),
+        );
+
+        // Step 3: On restart, from_db_map() rebuilds Settings from the full
+        // DB map.
+        let restored = Settings::from_db_map(&db);
+        assert_eq!(
+            restored.selected_model,
+            Some("new-model".to_string()),
+            "/model change must survive DB round trip"
+        );
+    }
+
+    /// Regression: TOML overlay must not clobber a DB-persisted selected_model
+    /// when the TOML file matches the DB. This is the normal case after /model
+    /// successfully writes to both DB and TOML.
+    #[test]
+    fn toml_overlay_preserves_matching_model() {
+        // DB settings with new model from /model command.
+        let mut db_settings = Settings {
+            llm_backend: Some("nearai".to_string()),
+            selected_model: Some("new-model".to_string()),
+            ..Default::default()
+        };
+
+        // TOML also updated by /model command to the same value.
+        let toml_settings = Settings {
+            selected_model: Some("new-model".to_string()),
+            ..Default::default()
+        };
+
+        db_settings.merge_from(&toml_settings);
+        assert_eq!(
+            db_settings.selected_model,
+            Some("new-model".to_string()),
+            "TOML overlay must not clobber matching model"
+        );
+    }
+
+    /// Regression: when /model updates DB but TOML write fails, a stale TOML
+    /// file would overwrite the DB value. This test documents the priority:
+    /// TOML > DB (by design). persist_selected_model MUST update the TOML.
+    #[test]
+    fn stale_toml_overwrites_db_model() {
+        // DB has the new model from /model.
+        let mut db_settings = Settings {
+            selected_model: Some("new-model".to_string()),
+            ..Default::default()
+        };
+
+        // TOML still has the old model (write failed or was not attempted).
+        let stale_toml = Settings {
+            selected_model: Some("old-model".to_string()),
+            ..Default::default()
+        };
+
+        db_settings.merge_from(&stale_toml);
+        // This documents the current priority: TOML wins over DB.
+        // The fix in persist_selected_model ensures TOML is always updated.
+        assert_eq!(
+            db_settings.selected_model,
+            Some("old-model".to_string()),
+            "TOML overlay has higher priority than DB (by design)"
+        );
+    }
+
     /// Regression test: /model command must persist selected_model to TOML config.
     /// Prior to the fix, `set_model()` only changed the in-memory provider and the
     /// choice was lost on restart.
@@ -1322,6 +1408,28 @@ mod tests {
         assert_eq!(reloaded.selected_model, Some("new-model".to_string()));
     }
 
+    /// Regression: /model must create config.toml when it doesn't exist, so the
+    /// model survives restarts. Previously the Ok(None) case was a no-op.
+    #[test]
+    fn toml_created_when_missing_for_model_persist() {
+        let dir = tempfile::tempdir().unwrap();
+        let path = dir.path().join("config.toml");
+
+        // No config.toml yet (fresh install, no wizard).
+        assert!(Settings::load_toml(&path).unwrap().is_none());
+
+        // Simulate what persist_selected_model now does for the Ok(None) case.
+        let settings = Settings {
+            selected_model: Some("new-model".to_string()),
+            ..Default::default()
+        };
+        settings.save_toml(&path).unwrap();
+
+        // Verify the model survived.
+        let loaded = Settings::load_toml(&path).unwrap().unwrap();
+        assert_eq!(loaded.selected_model, Some("new-model".to_string()));
+    }
+
     #[test]
     fn toml_missing_file_returns_none() {
         let result = Settings::load_toml(std::path::Path::new("/tmp/nonexistent_config.toml"));

src/tenant.rs

@@ -0,0 +1,906 @@
+//! Compile-time tenant isolation.
+//!
+//! Provides two database access tiers:
+//!
+//! - **[`TenantScope`]** (default): All operations are bound to a single user.
+//!   ID-based lookups return `None` if the resource doesn't belong to this user.
+//!   This is the only way handler code should access the database.
+//!
+//! - **[`AdminScope`]**: Cross-tenant access for system-level operations
+//!   (heartbeat, routine engine, self-repair). Must be obtained explicitly via
+//!   [`AgentDeps::admin_store()`](crate::agent::AgentDeps::admin_store).
+//!
+//! [`TenantCtx`] bundles a `TenantScope` with workspace, cost guard, and
+//! per-tenant rate limiting. Constructed once per request at the entry point
+//! where a `user_id` becomes known.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use chrono::{DateTime, Utc};
+use rust_decimal::Decimal;
+use tokio::sync::{Semaphore, SemaphorePermit};
+use uuid::Uuid;
+
+use crate::agent::BrokenTool;
+use crate::agent::cost_guard::{CostGuard, CostLimitExceeded};
+use crate::agent::routine::{Routine, RoutineRun, RunStatus};
+use crate::context::{ActionRecord, JobContext, JobState};
+use crate::db::Database;
+use crate::error::DatabaseError;
+use crate::history::{
+    AgentJobRecord, AgentJobSummary, ConversationMessage, ConversationSummary, LlmCallRecord,
+    SandboxJobRecord, SandboxJobSummary, SettingRow,
+};
+use crate::workspace::Workspace;
+
+// ---------------------------------------------------------------------------
+// TenantScope — scoped database access (default tier)
+// ---------------------------------------------------------------------------
+
+/// Scoped database view. All operations are bound to a single user.
+///
+/// This is the **only** way handler code should access the database.
+/// ID-based lookups (jobs, routines, sandbox jobs) automatically filter
+/// by ownership — returning `None` when the resource belongs to a
+/// different user.
+#[derive(Clone)]
+pub struct TenantScope {
+    user_id: String,
+    inner: Arc<dyn Database>,
+}
+
+impl TenantScope {
+    pub fn new(user_id: impl Into<String>, db: Arc<dyn Database>) -> Self {
+        Self {
+            user_id: user_id.into(),
+            inner: db,
+        }
+    }
+
+    pub fn user_id(&self) -> &str {
+        &self.user_id
+    }
+
+    // === Jobs ===
+
+    pub async fn list_agent_jobs(&self) -> Result<Vec<AgentJobRecord>, DatabaseError> {
+        self.inner.list_agent_jobs_for_user(&self.user_id).await
+    }
+
+    pub async fn agent_job_summary(&self) -> Result<AgentJobSummary, DatabaseError> {
+        self.inner.agent_job_summary_for_user(&self.user_id).await
+    }
+
+    /// Fetch a job by ID, returning `None` if it doesn't belong to this user.
+    pub async fn get_job(&self, id: Uuid) -> Result<Option<JobContext>, DatabaseError> {
+        match self.inner.get_job(id).await? {
+            Some(ctx) if ctx.user_id == self.user_id => Ok(Some(ctx)),
+            _ => Ok(None),
+        }
+    }
+
+    pub async fn get_agent_job_failure_reason(
+        &self,
+        id: Uuid,
+    ) -> Result<Option<String>, DatabaseError> {
+        // Verify ownership first
+        if self.get_job(id).await?.is_none() {
+            return Ok(None);
+        }
+        self.inner.get_agent_job_failure_reason(id).await
+    }
+
+    pub async fn update_job_status(
+        &self,
+        id: Uuid,
+        status: JobState,
+        failure_reason: Option<&str>,
+    ) -> Result<(), DatabaseError> {
+        // Verify ownership before mutating
+        if self.get_job(id).await?.is_none() {
+            return Err(DatabaseError::NotFound {
+                entity: "job".to_string(),
+                id: id.to_string(),
+            });
+        }
+        self.inner
+            .update_job_status(id, status, failure_reason)
+            .await
+    }
+
+    // === Sandbox jobs ===
+
+    pub async fn list_sandbox_jobs(&self) -> Result<Vec<SandboxJobRecord>, DatabaseError> {
+        self.inner.list_sandbox_jobs_for_user(&self.user_id).await
+    }
+
+    pub async fn sandbox_job_summary(&self) -> Result<SandboxJobSummary, DatabaseError> {
+        self.inner.sandbox_job_summary_for_user(&self.user_id).await
+    }
+
+    /// Fetch a sandbox job by ID, returning `None` if it doesn't belong to this user.
+    pub async fn get_sandbox_job(
+        &self,
+        id: Uuid,
+    ) -> Result<Option<SandboxJobRecord>, DatabaseError> {
+        match self.inner.get_sandbox_job(id).await? {
+            Some(job) if job.user_id == self.user_id => Ok(Some(job)),
+            _ => Ok(None),
+        }
+    }
+
+    pub async fn sandbox_job_belongs_to_user(&self, job_id: Uuid) -> Result<bool, DatabaseError> {
+        self.inner
+            .sandbox_job_belongs_to_user(job_id, &self.user_id)
+            .await
+    }
+
+    // === Routines ===
+
+    pub async fn list_routines(&self) -> Result<Vec<Routine>, DatabaseError> {
+        self.inner.list_routines(&self.user_id).await
+    }
+
+    pub async fn get_routine_by_name(&self, name: &str) -> Result<Option<Routine>, DatabaseError> {
+        self.inner.get_routine_by_name(&self.user_id, name).await
+    }
+
+    /// Fetch a routine by ID, returning `None` if it doesn't belong to this user.
+    pub async fn get_routine(&self, id: Uuid) -> Result<Option<Routine>, DatabaseError> {
+        match self.inner.get_routine(id).await? {
+            Some(r) if r.user_id == self.user_id => Ok(Some(r)),
+            _ => Ok(None),
+        }
+    }
+
+    pub async fn create_routine(&self, routine: &Routine) -> Result<(), DatabaseError> {
+        debug_assert_eq!(
+            routine.user_id, self.user_id,
+            "routine.user_id must match TenantScope user"
+        );
+        self.inner.create_routine(routine).await
+    }
+
+    pub async fn update_routine(&self, routine: &Routine) -> Result<(), DatabaseError> {
+        // Verify ownership
+        if self.get_routine(routine.id).await?.is_none() {
+            return Err(DatabaseError::NotFound {
+                entity: "routine".to_string(),
+                id: routine.id.to_string(),
+            });
+        }
+        self.inner.update_routine(routine).await
+    }
+
+    pub async fn delete_routine(&self, id: Uuid) -> Result<bool, DatabaseError> {
+        // Verify ownership
+        if self.get_routine(id).await?.is_none() {
+            return Err(DatabaseError::NotFound {
+                entity: "routine".to_string(),
+                id: id.to_string(),
+            });
+        }
+        self.inner.delete_routine(id).await
+    }
+
+    /// List routine runs, verifying the routine belongs to this user.
+    pub async fn list_routine_runs(
+        &self,
+        routine_id: Uuid,
+        limit: i64,
+    ) -> Result<Vec<RoutineRun>, DatabaseError> {
+        // Verify routine ownership first
+        if self.get_routine(routine_id).await?.is_none() {
+            return Err(DatabaseError::NotFound {
+                entity: "routine".to_string(),
+                id: routine_id.to_string(),
+            });
+        }
+        self.inner.list_routine_runs(routine_id, limit).await
+    }
+
+    pub async fn get_webhook_routine_by_path(
+        &self,
+        path: &str,
+    ) -> Result<Option<Routine>, DatabaseError> {
+        self.inner
+            .get_webhook_routine_by_path(path, Some(&self.user_id))
+            .await
+    }
+
+    // === Settings ===
+
+    pub async fn get_setting(&self, key: &str) -> Result<Option<serde_json::Value>, DatabaseError> {
+        self.inner.get_setting(&self.user_id, key).await
+    }
+
+    pub async fn get_setting_full(&self, key: &str) -> Result<Option<SettingRow>, DatabaseError> {
+        self.inner.get_setting_full(&self.user_id, key).await
+    }
+
+    pub async fn set_setting(
+        &self,
+        key: &str,
+        value: &serde_json::Value,
+    ) -> Result<(), DatabaseError> {
+        self.inner.set_setting(&self.user_id, key, value).await
+    }
+
+    pub async fn delete_setting(&self, key: &str) -> Result<bool, DatabaseError> {
+        self.inner.delete_setting(&self.user_id, key).await
+    }
+
+    pub async fn list_settings(&self) -> Result<Vec<SettingRow>, DatabaseError> {
+        self.inner.list_settings(&self.user_id).await
+    }
+
+    pub async fn get_all_settings(
+        &self,
+    ) -> Result<HashMap<String, serde_json::Value>, DatabaseError> {
+        self.inner.get_all_settings(&self.user_id).await
+    }
+
+    pub async fn set_all_settings(
+        &self,
+        settings: &HashMap<String, serde_json::Value>,
+    ) -> Result<(), DatabaseError> {
+        self.inner.set_all_settings(&self.user_id, settings).await
+    }
+
+    pub async fn has_settings(&self) -> Result<bool, DatabaseError> {
+        self.inner.has_settings(&self.user_id).await
+    }
+
+    // === Conversations ===
+
+    pub async fn create_conversation(
+        &self,
+        channel: &str,
+        thread_id: Option<&str>,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .create_conversation(channel, &self.user_id, thread_id)
+            .await
+    }
+
+    pub async fn ensure_conversation(
+        &self,
+        id: Uuid,
+        channel: &str,
+        thread_id: Option<&str>,
+    ) -> Result<bool, DatabaseError> {
+        self.inner
+            .ensure_conversation(id, channel, &self.user_id, thread_id)
+            .await
+    }
+
+    pub async fn list_conversations_with_preview(
+        &self,
+        channel: &str,
+        limit: i64,
+    ) -> Result<Vec<ConversationSummary>, DatabaseError> {
+        self.inner
+            .list_conversations_with_preview(&self.user_id, channel, limit)
+            .await
+    }
+
+    pub async fn list_conversations_all_channels(
+        &self,
+        limit: i64,
+    ) -> Result<Vec<ConversationSummary>, DatabaseError> {
+        self.inner
+            .list_conversations_all_channels(&self.user_id, limit)
+            .await
+    }
+
+    pub async fn get_or_create_routine_conversation(
+        &self,
+        routine_id: Uuid,
+        routine_name: &str,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .get_or_create_routine_conversation(routine_id, routine_name, &self.user_id)
+            .await
+    }
+
+    pub async fn get_or_create_heartbeat_conversation(&self) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .get_or_create_heartbeat_conversation(&self.user_id)
+            .await
+    }
+
+    pub async fn get_or_create_assistant_conversation(
+        &self,
+        channel: &str,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .get_or_create_assistant_conversation(&self.user_id, channel)
+            .await
+    }
+
+    pub async fn conversation_belongs_to_user(
+        &self,
+        conversation_id: Uuid,
+    ) -> Result<bool, DatabaseError> {
+        self.inner
+            .conversation_belongs_to_user(conversation_id, &self.user_id)
+            .await
+    }
+
+    /// Add a message to a conversation owned by this tenant.
+    ///
+    /// Verifies the conversation belongs to this user before adding.
+    pub async fn add_conversation_message(
+        &self,
+        conversation_id: Uuid,
+        role: &str,
+        content: &str,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .add_conversation_message(conversation_id, role, content)
+            .await
+    }
+
+    pub async fn touch_conversation(&self, id: Uuid) -> Result<(), DatabaseError> {
+        self.inner.touch_conversation(id).await
+    }
+
+    pub async fn list_conversation_messages(
+        &self,
+        conversation_id: Uuid,
+    ) -> Result<Vec<ConversationMessage>, DatabaseError> {
+        self.inner.list_conversation_messages(conversation_id).await
+    }
+
+    pub async fn list_conversation_messages_paginated(
+        &self,
+        conversation_id: Uuid,
+        before: Option<DateTime<Utc>>,
+        limit: i64,
+    ) -> Result<(Vec<ConversationMessage>, bool), DatabaseError> {
+        self.inner
+            .list_conversation_messages_paginated(conversation_id, before, limit)
+            .await
+    }
+
+    pub async fn create_conversation_with_metadata(
+        &self,
+        channel: &str,
+        metadata: &serde_json::Value,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .create_conversation_with_metadata(channel, &self.user_id, metadata)
+            .await
+    }
+
+    pub async fn update_conversation_metadata_field(
+        &self,
+        id: Uuid,
+        key: &str,
+        value: &serde_json::Value,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .update_conversation_metadata_field(id, key, value)
+            .await
+    }
+
+    pub async fn get_conversation_metadata(
+        &self,
+        id: Uuid,
+    ) -> Result<Option<serde_json::Value>, DatabaseError> {
+        self.inner.get_conversation_metadata(id).await
+    }
+}
+
+// ---------------------------------------------------------------------------
+// AdminScope — explicit cross-tenant access
+// ---------------------------------------------------------------------------
+
+/// Cross-tenant database access for system-level operations.
+///
+/// **Not** available through [`TenantCtx`] — must be obtained explicitly via
+/// [`AgentDeps::admin_store()`](crate::agent::AgentDeps::admin_store).
+///
+/// Used by: heartbeat enumeration, routine engine scheduling, self-repair,
+/// scheduler job persistence, worker status updates.
+#[derive(Clone)]
+pub struct AdminScope {
+    inner: Arc<dyn Database>,
+}
+
+impl AdminScope {
+    pub fn new(db: Arc<dyn Database>) -> Self {
+        Self { inner: db }
+    }
+
+    /// Access the raw Database trait object.
+    ///
+    /// Prefer using the typed methods on AdminScope instead. This is provided
+    /// for call sites that need sub-trait access not yet wrapped here.
+    pub fn db(&self) -> &Arc<dyn Database> {
+        &self.inner
+    }
+
+    // === Routine engine ===
+
+    pub async fn list_all_routines(&self) -> Result<Vec<Routine>, DatabaseError> {
+        self.inner.list_all_routines().await
+    }
+
+    pub async fn list_event_routines(&self) -> Result<Vec<Routine>, DatabaseError> {
+        self.inner.list_event_routines().await
+    }
+
+    pub async fn list_due_cron_routines(&self) -> Result<Vec<Routine>, DatabaseError> {
+        self.inner.list_due_cron_routines().await
+    }
+
+    pub async fn list_dispatched_routine_runs(&self) -> Result<Vec<RoutineRun>, DatabaseError> {
+        self.inner.list_dispatched_routine_runs().await
+    }
+
+    pub async fn count_running_routine_runs_batch(
+        &self,
+        routine_ids: &[Uuid],
+    ) -> Result<HashMap<Uuid, i64>, DatabaseError> {
+        self.inner
+            .count_running_routine_runs_batch(routine_ids)
+            .await
+    }
+
+    pub async fn batch_get_last_run_status(
+        &self,
+        routine_ids: &[Uuid],
+    ) -> Result<HashMap<Uuid, RunStatus>, DatabaseError> {
+        self.inner.batch_get_last_run_status(routine_ids).await
+    }
+
+    pub async fn count_running_routine_runs(&self, routine_id: Uuid) -> Result<i64, DatabaseError> {
+        self.inner.count_running_routine_runs(routine_id).await
+    }
+
+    pub async fn update_routine_runtime(
+        &self,
+        id: Uuid,
+        last_run_at: DateTime<Utc>,
+        next_fire_at: Option<DateTime<Utc>>,
+        run_count: u64,
+        consecutive_failures: u32,
+        state: &serde_json::Value,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .update_routine_runtime(
+                id,
+                last_run_at,
+                next_fire_at,
+                run_count,
+                consecutive_failures,
+                state,
+            )
+            .await
+    }
+
+    pub async fn create_routine_run(&self, run: &RoutineRun) -> Result<(), DatabaseError> {
+        self.inner.create_routine_run(run).await
+    }
+
+    pub async fn complete_routine_run(
+        &self,
+        id: Uuid,
+        status: RunStatus,
+        result_summary: Option<&str>,
+        tokens_used: Option<i32>,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .complete_routine_run(id, status, result_summary, tokens_used)
+            .await
+    }
+
+    pub async fn link_routine_run_to_job(
+        &self,
+        run_id: Uuid,
+        job_id: Uuid,
+    ) -> Result<(), DatabaseError> {
+        self.inner.link_routine_run_to_job(run_id, job_id).await
+    }
+
+    pub async fn get_routine(&self, id: Uuid) -> Result<Option<Routine>, DatabaseError> {
+        self.inner.get_routine(id).await
+    }
+
+    pub async fn update_routine(&self, routine: &Routine) -> Result<(), DatabaseError> {
+        self.inner.update_routine(routine).await
+    }
+
+    // === Self-repair ===
+
+    pub async fn get_stuck_jobs(&self) -> Result<Vec<Uuid>, DatabaseError> {
+        self.inner.get_stuck_jobs().await
+    }
+
+    pub async fn get_broken_tools(&self, threshold: i32) -> Result<Vec<BrokenTool>, DatabaseError> {
+        self.inner.get_broken_tools(threshold).await
+    }
+
+    pub async fn record_tool_failure(
+        &self,
+        tool_name: &str,
+        error_message: &str,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .record_tool_failure(tool_name, error_message)
+            .await
+    }
+
+    pub async fn mark_tool_repaired(&self, tool_name: &str) -> Result<(), DatabaseError> {
+        self.inner.mark_tool_repaired(tool_name).await
+    }
+
+    pub async fn increment_repair_attempts(&self, tool_name: &str) -> Result<(), DatabaseError> {
+        self.inner.increment_repair_attempts(tool_name).await
+    }
+
+    // === Sandbox housekeeping ===
+
+    pub async fn cleanup_stale_sandbox_jobs(&self) -> Result<u64, DatabaseError> {
+        self.inner.cleanup_stale_sandbox_jobs().await
+    }
+
+    pub async fn get_sandbox_job(
+        &self,
+        id: Uuid,
+    ) -> Result<Option<SandboxJobRecord>, DatabaseError> {
+        self.inner.get_sandbox_job(id).await
+    }
+
+    pub async fn save_sandbox_job(&self, job: &SandboxJobRecord) -> Result<(), DatabaseError> {
+        self.inner.save_sandbox_job(job).await
+    }
+
+    pub async fn update_sandbox_job_status(
+        &self,
+        id: Uuid,
+        status: &str,
+        success: Option<bool>,
+        message: Option<&str>,
+        started_at: Option<DateTime<Utc>>,
+        completed_at: Option<DateTime<Utc>>,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .update_sandbox_job_status(id, status, success, message, started_at, completed_at)
+            .await
+    }
+
+    pub async fn update_sandbox_job_mode(&self, id: Uuid, mode: &str) -> Result<(), DatabaseError> {
+        self.inner.update_sandbox_job_mode(id, mode).await
+    }
+
+    pub async fn get_sandbox_job_mode(&self, id: Uuid) -> Result<Option<String>, DatabaseError> {
+        self.inner.get_sandbox_job_mode(id).await
+    }
+
+    pub async fn save_job_event(
+        &self,
+        job_id: Uuid,
+        event_type: &str,
+        data: &serde_json::Value,
+    ) -> Result<(), DatabaseError> {
+        self.inner.save_job_event(job_id, event_type, data).await
+    }
+
+    pub async fn list_job_events(
+        &self,
+        job_id: Uuid,
+        limit: Option<i64>,
+    ) -> Result<Vec<crate::history::JobEventRecord>, DatabaseError> {
+        self.inner.list_job_events(job_id, limit).await
+    }
+
+    // === Job persistence (scheduler, worker) ===
+
+    pub async fn get_job(&self, id: Uuid) -> Result<Option<JobContext>, DatabaseError> {
+        self.inner.get_job(id).await
+    }
+
+    pub async fn save_job(&self, ctx: &JobContext) -> Result<(), DatabaseError> {
+        self.inner.save_job(ctx).await
+    }
+
+    pub async fn update_job_status(
+        &self,
+        id: Uuid,
+        status: JobState,
+        failure_reason: Option<&str>,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .update_job_status(id, status, failure_reason)
+            .await
+    }
+
+    pub async fn mark_job_stuck(&self, id: Uuid) -> Result<(), DatabaseError> {
+        self.inner.mark_job_stuck(id).await
+    }
+
+    pub async fn list_agent_jobs(&self) -> Result<Vec<AgentJobRecord>, DatabaseError> {
+        self.inner.list_agent_jobs().await
+    }
+
+    pub async fn get_agent_job_failure_reason(
+        &self,
+        id: Uuid,
+    ) -> Result<Option<String>, DatabaseError> {
+        self.inner.get_agent_job_failure_reason(id).await
+    }
+
+    // === LLM call recording ===
+
+    pub async fn record_llm_call(&self, record: &LlmCallRecord<'_>) -> Result<Uuid, DatabaseError> {
+        self.inner.record_llm_call(record).await
+    }
+
+    pub async fn save_action(
+        &self,
+        job_id: Uuid,
+        action: &ActionRecord,
+    ) -> Result<(), DatabaseError> {
+        self.inner.save_action(job_id, action).await
+    }
+
+    pub async fn get_job_actions(&self, job_id: Uuid) -> Result<Vec<ActionRecord>, DatabaseError> {
+        self.inner.get_job_actions(job_id).await
+    }
+
+    // === Estimation ===
+
+    pub async fn save_estimation_snapshot(
+        &self,
+        job_id: Uuid,
+        category: &str,
+        tool_names: &[String],
+        estimated_cost: Decimal,
+        estimated_time_secs: i32,
+        estimated_value: Decimal,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .save_estimation_snapshot(
+                job_id,
+                category,
+                tool_names,
+                estimated_cost,
+                estimated_time_secs,
+                estimated_value,
+            )
+            .await
+    }
+
+    pub async fn update_estimation_actuals(
+        &self,
+        id: Uuid,
+        actual_cost: Decimal,
+        actual_time_secs: i32,
+        actual_value: Option<Decimal>,
+    ) -> Result<(), DatabaseError> {
+        self.inner
+            .update_estimation_actuals(id, actual_cost, actual_time_secs, actual_value)
+            .await
+    }
+
+    // === Conversations (admin context) ===
+
+    pub async fn add_conversation_message(
+        &self,
+        conversation_id: Uuid,
+        role: &str,
+        content: &str,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .add_conversation_message(conversation_id, role, content)
+            .await
+    }
+
+    pub async fn get_or_create_routine_conversation(
+        &self,
+        routine_id: Uuid,
+        routine_name: &str,
+        user_id: &str,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .get_or_create_routine_conversation(routine_id, routine_name, user_id)
+            .await
+    }
+
+    pub async fn get_or_create_heartbeat_conversation(
+        &self,
+        user_id: &str,
+    ) -> Result<Uuid, DatabaseError> {
+        self.inner
+            .get_or_create_heartbeat_conversation(user_id)
+            .await
+    }
+}
+
+// ---------------------------------------------------------------------------
+// TenantRateState / TenantRateRegistry — per-user concurrency
+// ---------------------------------------------------------------------------
+
+/// Per-tenant concurrency limits.
+pub struct TenantRateState {
+    /// Limits concurrent LLM calls for this user.
+    pub llm_semaphore: Arc<Semaphore>,
+    /// Limits concurrent jobs for this user.
+    pub job_semaphore: Arc<Semaphore>,
+}
+
+impl TenantRateState {
+    pub fn new(max_llm_concurrent: usize, max_job_concurrent: usize) -> Self {
+        Self {
+            llm_semaphore: Arc::new(Semaphore::new(max_llm_concurrent)),
+            job_semaphore: Arc::new(Semaphore::new(max_job_concurrent)),
+        }
+    }
+}
+
+/// Registry that lazily creates per-tenant rate state.
+///
+/// Uses `tokio::sync::RwLock<HashMap>` (consistent with the rest of the
+/// codebase — no DashMap dependency).
+pub struct TenantRateRegistry {
+    state: tokio::sync::RwLock<HashMap<String, Arc<TenantRateState>>>,
+    max_llm_concurrent: usize,
+    max_job_concurrent: usize,
+}
+
+impl TenantRateRegistry {
+    pub fn new(max_llm_concurrent: usize, max_job_concurrent: usize) -> Self {
+        Self {
+            state: tokio::sync::RwLock::new(HashMap::new()),
+            max_llm_concurrent,
+            max_job_concurrent,
+        }
+    }
+
+    /// Get or lazily create rate state for a user.
+    pub async fn get_or_create(&self, user_id: &str) -> Arc<TenantRateState> {
+        // Fast path: read lock
+        {
+            let map = self.state.read().await;
+            if let Some(s) = map.get(user_id) {
+                return Arc::clone(s);
+            }
+        }
+        // Slow path: write lock with double-check
+        let mut map = self.state.write().await;
+        if let Some(s) = map.get(user_id) {
+            return Arc::clone(s);
+        }
+        let s = Arc::new(TenantRateState::new(
+            self.max_llm_concurrent,
+            self.max_job_concurrent,
+        ));
+        map.insert(user_id.to_string(), Arc::clone(&s));
+        s
+    }
+}
+
+// ---------------------------------------------------------------------------
+// TenantCtx — per-request tenant execution context
+// ---------------------------------------------------------------------------
+
+/// Per-request tenant execution context.
+///
+/// Bundles a [`TenantScope`] (scoped DB access), workspace, cost guard,
+/// and per-tenant rate limiting. Constructed once per request via
+/// [`AgentDeps::tenant_ctx()`](crate::agent::AgentDeps::tenant_ctx).
+///
+/// `Clone + Send + Sync` — safe to store on `ChatDelegate` without lifetime issues.
+#[derive(Clone)]
+pub struct TenantCtx {
+    user_id: String,
+    store: Option<TenantScope>,
+    workspace: Option<Arc<Workspace>>,
+    cost_guard: Arc<CostGuard>,
+    rate: Arc<TenantRateState>,
+}
+
+impl TenantCtx {
+    pub fn new(
+        user_id: impl Into<String>,
+        store: Option<TenantScope>,
+        workspace: Option<Arc<Workspace>>,
+        cost_guard: Arc<CostGuard>,
+        rate: Arc<TenantRateState>,
+    ) -> Self {
+        Self {
+            user_id: user_id.into(),
+            store,
+            workspace,
+            cost_guard,
+            rate,
+        }
+    }
+
+    pub fn user_id(&self) -> &str {
+        &self.user_id
+    }
+
+    pub fn store(&self) -> Option<&TenantScope> {
+        self.store.as_ref()
+    }
+
+    pub fn workspace(&self) -> Option<&Arc<Workspace>> {
+        self.workspace.as_ref()
+    }
+
+    pub fn cost_guard(&self) -> &CostGuard {
+        &self.cost_guard
+    }
+
+    /// Check cost limits for this tenant (global + per-user).
+    pub async fn check_cost_allowed(&self) -> Result<(), CostLimitExceeded> {
+        self.cost_guard.check_allowed_for_user(&self.user_id).await
+    }
+
+    /// Record an LLM call for this tenant.
+    #[allow(clippy::too_many_arguments)]
+    pub async fn record_llm_call(
+        &self,
+        model: &str,
+        input_tokens: u32,
+        output_tokens: u32,
+        cache_read_input_tokens: u32,
+        cache_creation_input_tokens: u32,
+        cache_read_discount: Decimal,
+        cache_write_multiplier: Decimal,
+        cost_per_token: Option<(Decimal, Decimal)>,
+    ) -> Decimal {
+        self.cost_guard
+            .record_llm_call_for_user(
+                &self.user_id,
+                model,
+                input_tokens,
+                output_tokens,
+                cache_read_input_tokens,
+                cache_creation_input_tokens,
+                cache_read_discount,
+                cache_write_multiplier,
+                cost_per_token,
+            )
+            .await
+    }
+
+    /// Acquire an LLM concurrency permit for this tenant.
+    pub async fn acquire_llm_permit(&self) -> Result<SemaphorePermit<'_>, crate::error::Error> {
+        self.rate.llm_semaphore.acquire().await.map_err(|_| {
+            crate::error::Error::Config(crate::error::ConfigError::InvalidValue {
+                key: "llm_semaphore".to_string(),
+                message: "semaphore closed".to_string(),
+            })
+        })
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn test_rate_registry_returns_same_state_for_same_user() {
+        let registry = TenantRateRegistry::new(4, 3);
+        let a1 = registry.get_or_create("alice").await;
+        let a2 = registry.get_or_create("alice").await;
+        assert!(Arc::ptr_eq(&a1, &a2));
+    }
+
+    #[tokio::test]
+    async fn test_rate_registry_different_users_get_different_state() {
+        let registry = TenantRateRegistry::new(4, 3);
+        let alice = registry.get_or_create("alice").await;
+        let bob = registry.get_or_create("bob").await;
+        assert!(!Arc::ptr_eq(&alice, &bob));
+    }
+}

src/testing/mod.rs

@@ -532,6 +532,7 @@ impl TestHarnessBuilder {
         let cost_guard = Arc::new(CostGuard::new(CostGuardConfig {
             max_cost_per_day_cents: None,
             max_actions_per_hour: None,
+            max_cost_per_user_per_day_cents: None,
         }));
 
         let channel = if self.stub_channel {
@@ -563,6 +564,8 @@ impl TestHarnessBuilder {
             document_extraction: None,
             sandbox_readiness: crate::agent::routine_engine::SandboxReadiness::DisabledByConfig,
             builder: None,
+            llm_backend: "nearai".to_string(),
+            tenant_rates: std::sync::Arc::new(crate::tenant::TenantRateRegistry::new(4, 3)),
         };
 
         TestHarness {

src/tools/builtin/job.rs

@@ -17,7 +17,6 @@ use uuid::Uuid;
 
 use crate::bootstrap::ironclaw_base_dir;
 use crate::channels::IncomingMessage;
-use crate::channels::web::types::SseEvent;
 use crate::context::{ContextManager, JobContext, JobState};
 use crate::db::Database;
 use crate::history::SandboxJobRecord;
@@ -25,6 +24,7 @@ use crate::orchestrator::auth::CredentialGrant;
 use crate::orchestrator::job_manager::{ContainerJobManager, JobMode};
 use crate::secrets::SecretsStore;
 use crate::tools::tool::{ApprovalRequirement, Tool, ToolError, ToolOutput, require_str};
+use ironclaw_common::AppEvent;
 
 /// Lazy scheduler reference, filled after Agent::new creates the Scheduler.
 ///
@@ -85,7 +85,7 @@ pub struct CreateJobTool {
     job_manager: Option<Arc<ContainerJobManager>>,
     store: Option<Arc<dyn Database>>,
     /// Broadcast sender for job events (used to subscribe a monitor).
-    event_tx: Option<tokio::sync::broadcast::Sender<(Uuid, String, SseEvent)>>,
+    event_tx: Option<tokio::sync::broadcast::Sender<(Uuid, String, AppEvent)>>,
     /// Injection channel for pushing messages into the agent loop.
     inject_tx: Option<tokio::sync::mpsc::Sender<IncomingMessage>>,
     /// Encrypted secrets store for validating credential grants.
@@ -120,7 +120,7 @@ impl CreateJobTool {
     /// monitor that forwards Claude Code output to the main agent loop.
     pub fn with_monitor_deps(
         mut self,
-        event_tx: tokio::sync::broadcast::Sender<(Uuid, String, SseEvent)>,
+        event_tx: tokio::sync::broadcast::Sender<(Uuid, String, AppEvent)>,
         inject_tx: tokio::sync::mpsc::Sender<IncomingMessage>,
     ) -> Self {
         self.event_tx = Some(event_tx);

src/tools/builtin/memory.rs

@@ -12,12 +12,10 @@
 //! Use `memory_write` to persist important facts that should be remembered
 //! across sessions.
 
-use std::collections::HashMap;
 use std::path::Path;
 use std::sync::Arc;
 
 use async_trait::async_trait;
-use tokio::sync::RwLock;
 
 use crate::context::JobContext;
 use crate::tools::tool::{Tool, ToolError, ToolOutput, require_str};
@@ -52,79 +50,6 @@ impl WorkspaceResolver for FixedWorkspaceResolver {
     }
 }
 
-/// Creates per-user workspaces on demand, caching them for reuse.
-///
-/// Used in multi-tenant mode where each authenticated user gets their own
-/// workspace scope. The workspace is constructed with the same configuration
-/// (embeddings, search config, memory layers) as the startup workspace.
-pub struct PerUserWorkspaceResolver {
-    db: Arc<dyn crate::db::Database>,
-    embeddings: Option<Arc<dyn crate::workspace::EmbeddingProvider>>,
-    embedding_cache_config: crate::workspace::EmbeddingCacheConfig,
-    search_config: crate::config::WorkspaceSearchConfig,
-    workspace_config: crate::config::WorkspaceConfig,
-    cache: RwLock<HashMap<String, Arc<Workspace>>>,
-}
-
-impl PerUserWorkspaceResolver {
-    pub fn new(
-        db: Arc<dyn crate::db::Database>,
-        embeddings: Option<Arc<dyn crate::workspace::EmbeddingProvider>>,
-        embedding_cache_config: crate::workspace::EmbeddingCacheConfig,
-        search_config: crate::config::WorkspaceSearchConfig,
-        workspace_config: crate::config::WorkspaceConfig,
-    ) -> Self {
-        Self {
-            db,
-            embeddings,
-            embedding_cache_config,
-            search_config,
-            workspace_config,
-            cache: RwLock::new(HashMap::new()),
-        }
-    }
-
-    fn build_workspace(&self, user_id: &str) -> Arc<Workspace> {
-        let mut ws = Workspace::new_with_db(user_id, Arc::clone(&self.db))
-            .with_search_config(&self.search_config);
-
-        if let Some(ref emb) = self.embeddings {
-            ws = ws.with_embeddings_cached(Arc::clone(emb), self.embedding_cache_config.clone());
-        }
-
-        if !self.workspace_config.read_scopes.is_empty() {
-            ws = ws.with_additional_read_scopes(self.workspace_config.read_scopes.clone());
-        }
-        ws = ws.with_memory_layers(self.workspace_config.memory_layers.clone());
-
-        Arc::new(ws)
-    }
-}
-
-#[async_trait]
-impl WorkspaceResolver for PerUserWorkspaceResolver {
-    async fn resolve(&self, user_id: &str) -> Arc<Workspace> {
-        // Fast path: read lock
-        {
-            let cache = self.cache.read().await;
-            if let Some(ws) = cache.get(user_id) {
-                return Arc::clone(ws);
-            }
-        }
-
-        // Slow path: write lock, double-check
-        let mut cache = self.cache.write().await;
-        if let Some(ws) = cache.get(user_id) {
-            return Arc::clone(ws);
-        }
-
-        let ws = self.build_workspace(user_id);
-        cache.insert(user_id.to_string(), Arc::clone(&ws));
-        tracing::debug!(user_id = user_id, "Created per-user workspace");
-        ws
-    }
-}
-
 /// Detect paths that are clearly local filesystem references, not workspace-memory docs.
 ///
 /// Examples:
@@ -1007,10 +932,10 @@ mod tests {
         }
 
         #[tokio::test]
-        async fn test_per_user_workspace_resolver_returns_different_workspaces() {
+        async fn test_workspace_pool_resolver_returns_different_workspaces() {
             let db = make_test_db().await;
 
-            let resolver = PerUserWorkspaceResolver::new(
+            let pool = crate::channels::web::server::WorkspacePool::new(
                 db,
                 None,
                 crate::workspace::EmbeddingCacheConfig::default(),
@@ -1018,8 +943,8 @@ mod tests {
                 crate::config::WorkspaceConfig::default(),
             );
 
-            let ws_alice = resolver.resolve("alice").await;
+            let ws_alice = pool.resolve("alice").await;
-            let ws_bob = resolver.resolve("bob").await;
+            let ws_bob = pool.resolve("bob").await;
 
             // Different user IDs should get different workspaces
             assert_eq!(ws_alice.user_id(), "alice");
@@ -1028,10 +953,10 @@ mod tests {
         }
 
         #[tokio::test]
-        async fn test_per_user_workspace_resolver_caches_workspace() {
+        async fn test_workspace_pool_resolver_caches_workspace() {
             let db = make_test_db().await;
 
-            let resolver = PerUserWorkspaceResolver::new(
+            let pool = crate::channels::web::server::WorkspacePool::new(
                 db,
                 None,
                 crate::workspace::EmbeddingCacheConfig::default(),
@@ -1039,8 +964,8 @@ mod tests {
                 crate::config::WorkspaceConfig::default(),
             );
 
-            let ws1 = resolver.resolve("alice").await;
+            let ws1 = pool.resolve("alice").await;
-            let ws2 = resolver.resolve("alice").await;
+            let ws2 = pool.resolve("alice").await;
 
             // Same user_id should return the same cached Arc (pointer equality)
             assert!(Arc::ptr_eq(&ws1, &ws2));

src/tools/builtin/routine.rs

@@ -140,7 +140,8 @@ fn execution_properties() -> Value {
         },
         "use_tools": {
             "type": "boolean",
-            "description": "Only applies to lightweight mode. When true, safe non-approval tools are available."
+            "default": true,
+            "description": "Only applies to lightweight mode. New lightweight routines default this to true; when enabled, the routine can use the owner's live autonomous tool scope."
         },
         "max_tool_rounds": {
             "type": "integer",
@@ -290,7 +291,7 @@ fn routine_request_discovery_schema() -> Value {
 fn lightweight_execution_variant() -> Value {
     serde_json::json!({
         "type": "object",
-        "description": "Default lightweight execution. Applies when execution is omitted or execution.mode='lightweight'.",
+        "description": "Default lightweight execution. Applies when execution is omitted or execution.mode='lightweight'. New lightweight routines default to tools enabled unless execution.use_tools=false is set.",
         "properties": {
             "mode": {
                 "type": "string",
@@ -304,7 +305,8 @@ fn lightweight_execution_variant() -> Value {
             },
             "use_tools": {
                 "type": "boolean",
-                "description": "When true, safe non-approval tools are available."
+                "default": true,
+                "description": "Defaults to true for new lightweight routines. When enabled, the routine can use the owner's live autonomous tool scope."
             },
             "max_tool_rounds": {
                 "type": "integer",
@@ -335,7 +337,7 @@ fn full_job_execution_variant() -> Value {
 fn execution_discovery_schema() -> Value {
     serde_json::json!({
         "type": "object",
-        "description": "Optional execution settings. Omit this block for the default lightweight mode.",
+        "description": "Optional execution settings. Omit this block for the default lightweight mode with tools enabled.",
         "properties": execution_properties(),
         "oneOf": [
             lightweight_execution_variant(),
@@ -408,7 +410,8 @@ fn routine_create_tool_summary() -> ToolDiscoverySummary {
             "execution.mode='full_job' uses the owner's live autonomous tool scope and ignores use_tools, max_tool_rounds, and context_paths.".into(),
         ],
         notes: vec![
-            "Omitting execution defaults to lightweight mode.".into(),
+            "Omitting execution defaults to lightweight mode with tools enabled.".into(),
+            "Set execution.use_tools=false to keep a new lightweight routine text-only.".into(),
             "Omitting delivery.user falls back to the owner's last-seen notification target.".into(),
             "advanced.cooldown_secs defaults to 300.".into(),
             "Legacy flat aliases are still accepted for compatibility, but grouped fields are preferred.".into(),
@@ -605,7 +608,8 @@ fn routine_create_schema(include_compatibility_aliases: bool) -> Value {
 }
 
 pub(crate) fn routine_create_parameters_schema() -> Value {
-    routine_create_schema(false)
+    static CACHE: OnceLock<Value> = OnceLock::new();
+    CACHE.get_or_init(|| routine_create_schema(false)).clone()
 }
 
 fn routine_create_discovery_schema() -> Value {
@@ -646,6 +650,23 @@ pub(crate) fn routine_update_parameters_schema() -> Value {
     })
 }
 
+const ROUTINE_LAST_NAME_STASH_KEY: &str = "__routine_last_name";
+
+async fn stash_last_routine_name(ctx: &JobContext, name: &str) {
+    ctx.tool_output_stash
+        .write()
+        .await
+        .insert(ROUTINE_LAST_NAME_STASH_KEY.to_string(), name.to_string());
+}
+
+async fn restore_last_routine_name(ctx: &JobContext) -> Option<String> {
+    ctx.tool_output_stash
+        .read()
+        .await
+        .get(ROUTINE_LAST_NAME_STASH_KEY)
+        .cloned()
+}
+
 fn nested_object<'a>(params: &'a Value, field: &str) -> Option<&'a Map<String, Value>> {
     params.get(field).and_then(Value::as_object)
 }
@@ -852,11 +873,15 @@ fn parse_execution_mode(value: Option<String>) -> Result<NormalizedExecutionMode
     }
 }
 
-fn parse_routine_execution(params: &Value) -> Result<NormalizedExecutionRequest, ToolError> {
+fn parse_routine_execution(
+    params: &Value,
+    default_use_tools: bool,
+) -> Result<NormalizedExecutionRequest, ToolError> {
     let mode = parse_execution_mode(string_field(params, "execution", "mode", &["action_type"]))?;
     let context_paths =
         string_array_field(params, "execution", "context_paths", &["context_paths"]);
-    let use_tools = bool_field(params, "execution", "use_tools", &["use_tools"]).unwrap_or(false);
+    let use_tools =
+        bool_field(params, "execution", "use_tools", &["use_tools"]).unwrap_or(default_use_tools);
     let max_tool_rounds = u64_field(params, "execution", "max_tool_rounds", &["max_tool_rounds"])
         .unwrap_or(3)
         .clamp(1, crate::agent::routine::MAX_TOOL_ROUNDS_LIMIT as u64)
@@ -888,7 +913,7 @@ fn parse_routine_create_request(
         .unwrap_or("")
         .to_string();
     let trigger = parse_routine_trigger(params)?;
-    let execution = parse_routine_execution(params)?;
+    let execution = parse_routine_execution(params, true)?;
     let delivery = parse_routine_delivery(params);
     let cooldown_secs =
         u64_field(params, "advanced", "cooldown_secs", &["cooldown_secs"]).unwrap_or(300);
@@ -907,7 +932,7 @@ fn parse_routine_create_request(
 fn build_routine_trigger(trigger: &NormalizedTriggerRequest) -> Trigger {
     match trigger {
         NormalizedTriggerRequest::Cron { schedule, timezone } => Trigger::Cron {
-            schedule: schedule.clone(),
+            schedule: normalize_cron_expression(schedule),
             timezone: timezone.clone(),
         },
         NormalizedTriggerRequest::Manual => Trigger::Manual,
@@ -1007,7 +1032,8 @@ fn event_emit_schema(include_source_alias: bool) -> Value {
 }
 
 pub(crate) fn event_emit_parameters_schema() -> Value {
-    event_emit_schema(false)
+    static CACHE: OnceLock<Value> = OnceLock::new();
+    CACHE.get_or_init(|| event_emit_schema(false)).clone()
 }
 
 fn event_emit_discovery_schema() -> Value {
@@ -1084,6 +1110,7 @@ impl Tool for RoutineCreateTool {
     ) -> Result<ToolOutput, ToolError> {
         let start = std::time::Instant::now();
         let normalized = parse_routine_create_request(&params)?;
+        stash_last_routine_name(ctx, &normalized.name).await;
         let trigger = build_routine_trigger(&normalized.trigger);
         let action =
             build_routine_action(&normalized.name, &normalized.prompt, &normalized.execution);
@@ -1265,6 +1292,7 @@ impl Tool for RoutineUpdateTool {
         let start = std::time::Instant::now();
 
         let name = require_str(&params, "name")?;
+        stash_last_routine_name(ctx, name).await;
 
         let mut routine = self
             .store
@@ -1402,11 +1430,24 @@ impl Tool for RoutineDeleteTool {
     ) -> Result<ToolOutput, ToolError> {
         let start = std::time::Instant::now();
 
-        let name = require_str(&params, "name")?;
+        let name = if let Some(name) = params.get("name").and_then(|v| v.as_str()) {
+            if name.trim().is_empty() {
+                return Err(ToolError::InvalidParameters(
+                    "'name' parameter cannot be empty".to_string(),
+                ));
+            }
+            name.to_string()
+        } else {
+            restore_last_routine_name(ctx).await.ok_or_else(|| {
+                ToolError::InvalidParameters(
+                    "missing 'name' parameter and no previous routine target to infer".to_string(),
+                )
+            })?
+        };
 
         let routine = self
             .store
-            .get_routine_by_name(&ctx.user_id, name)
+            .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)))?;
@@ -1421,7 +1462,7 @@ impl Tool for RoutineDeleteTool {
         self.engine.refresh_event_cache().await;
 
         let result = serde_json::json!({
-            "name": name,
+            "name": &name,
             "deleted": deleted,
         });
 
@@ -1827,6 +1868,20 @@ mod tests {
         assert_eq!(parsed.cooldown_secs, 30);
     }
 
+    #[test]
+    fn build_routine_trigger_normalizes_cron_schedule() {
+        let trigger = build_routine_trigger(&NormalizedTriggerRequest::Cron {
+            schedule: "0 0 9 * * MON-FRI".to_string(),
+            timezone: Some("UTC".to_string()),
+        });
+
+        assert!(matches!(
+            trigger,
+            Trigger::Cron { schedule, timezone }
+                if schedule == "0 0 9 * * MON-FRI *" && timezone.as_deref() == Some("UTC")
+        ));
+    }
+
     #[test]
     fn parses_grouped_message_event_with_tools() {
         let params = serde_json::json!({
@@ -1863,6 +1918,56 @@ mod tests {
         );
     }
 
+    #[test]
+    fn parses_lightweight_create_with_tools_enabled_by_default() {
+        let params = serde_json::json!({
+            "name": "manual-check",
+            "prompt": "Inspect the repo for issues.",
+            "request": {
+                "kind": "manual"
+            }
+        });
+
+        let parsed = parse_routine_create_request(&params).expect("parse default lightweight");
+
+        assert!(
+            matches!(parsed.execution.mode, NormalizedExecutionMode::Lightweight),
+            "expected lightweight execution mode",
+        );
+        assert!(
+            parsed.execution.use_tools,
+            "new lightweight routines should default use_tools=true",
+        );
+        assert_eq!(parsed.execution.max_tool_rounds, 3);
+    }
+
+    #[test]
+    fn parses_lightweight_create_with_explicit_tools_disabled() {
+        let params = serde_json::json!({
+            "name": "manual-check",
+            "prompt": "Inspect the repo for issues.",
+            "request": {
+                "kind": "manual"
+            },
+            "execution": {
+                "use_tools": false
+            }
+        });
+
+        let parsed =
+            parse_routine_create_request(&params).expect("parse lightweight with tools disabled");
+
+        assert!(
+            matches!(parsed.execution.mode, NormalizedExecutionMode::Lightweight),
+            "expected lightweight execution mode",
+        );
+        assert!(
+            !parsed.execution.use_tools,
+            "explicit use_tools=false should be preserved",
+        );
+        assert_eq!(parsed.execution.max_tool_rounds, 3);
+    }
+
     #[test]
     fn parses_context_paths_with_trim_drop_empty_and_stable_dedupe() {
         let params = serde_json::json!({
@@ -2201,6 +2306,20 @@ mod tests {
                 .any(|rule| rule.contains("request.kind='cron'")),
             "summary should explain cron requirement",
         );
+        assert!(
+            summary
+                .notes
+                .iter()
+                .any(|note| note.contains("lightweight mode with tools enabled")),
+            "summary should mention the new lightweight default",
+        );
+        assert!(
+            summary
+                .notes
+                .iter()
+                .any(|note| note.contains("execution.use_tools=false")),
+            "summary should mention the text-only opt-out",
+        );
         assert!(
             summary
                 .notes

src/tools/mcp/client.rs

@@ -117,6 +117,11 @@ impl McpClient {
     /// The config must use HTTP transport (the default); for stdio/UDS use `new_with_transport`.
     ///
     /// Returns an error if the config uses a non-HTTP transport.
+    ///
+    /// **Note:** The session manager is NOT wired into the transport. For
+    /// production use, prefer `create_client_from_config()` which constructs
+    /// the transport with session tracking.
+    #[cfg(test)]
     pub fn new_with_config(config: McpServerConfig) -> Result<Self, ToolError> {
         if !matches!(
             config.effective_transport(),
@@ -214,7 +219,14 @@ impl McpClient {
         }
     }
 
-    /// Attach a session manager for Streamable HTTP session tracking.
+    /// Attach a session manager to the **client** only.
+    ///
+    /// **Warning:** This does NOT wire the session manager into the underlying
+    /// `HttpMcpTransport`, so the transport will not capture `Mcp-Session-Id`
+    /// from responses. For production use, construct the transport with
+    /// `HttpMcpTransport::with_session_manager()` and pass it to
+    /// `new_with_transport()` instead. See `create_client_from_config()`.
+    #[cfg(test)]
     pub fn with_session_manager(mut self, session_manager: Arc<McpSessionManager>) -> Self {
         self.session_manager = Some(session_manager);
         self
@@ -235,6 +247,12 @@ impl McpClient {
         self.session_manager.is_some()
     }
 
+    /// Get the underlying transport (test-only).
+    #[cfg(test)]
+    pub(crate) fn transport(&self) -> &Arc<dyn McpTransport> {
+        &self.transport
+    }
+
     /// Get the next request ID.
     fn next_request_id(&self) -> u64 {
         self.next_id.fetch_add(1, Ordering::SeqCst)

src/tools/mcp/factory.rs

@@ -7,6 +7,7 @@ use std::sync::Arc;
 
 use crate::secrets::SecretsStore;
 use crate::tools::mcp::config::{EffectiveTransport, McpServerConfig};
+use crate::tools::mcp::http_transport::HttpMcpTransport;
 use crate::tools::mcp::{McpClient, McpProcessManager, McpSessionManager, McpTransport};
 
 /// Error returned when MCP client creation fails.
@@ -78,33 +79,37 @@ pub async fn create_client_from_config(
             Err(McpFactoryError::UnixNotSupported { name: server_name })
         }
         EffectiveTransport::Http => {
+            // Authenticated (OAuth) path: tokens exist or server requires auth.
             if let Some(ref secrets) = secrets {
                 let has_tokens =
                     crate::tools::mcp::is_authenticated(&server, secrets, user_id).await;
 
                 if has_tokens || server.requires_auth() {
-                    Ok(McpClient::new_authenticated(
+                    return Ok(McpClient::new_authenticated(
                         server,
                         Arc::clone(session_manager),
                         Arc::clone(secrets),
                         user_id,
-                    ))
+                    ));
-                } else {
-                    Ok(McpClient::new_with_config(server)
-                        .map_err(|e| McpFactoryError::InvalidConfig {
-                            name: server_name.clone(),
-                            reason: e.to_string(),
-                        })?
-                        .with_session_manager(Arc::clone(session_manager)))
                 }
-            } else {
-                Ok(McpClient::new_with_config(server)
-                    .map_err(|e| McpFactoryError::InvalidConfig {
-                        name: server_name,
-                        reason: e.to_string(),
-                    })?
-                    .with_session_manager(Arc::clone(session_manager)))
             }
+
+            // Non-OAuth HTTP: wire the session manager into the *transport* so
+            // it captures `Mcp-Session-Id` from responses. Passing it only to
+            // the client (via `with_session_manager`) is not enough — the
+            // transport must know about it to read/write the header.
+            let transport = Arc::new(
+                HttpMcpTransport::new(server.url.clone(), server.name.clone())
+                    .with_session_manager(Arc::clone(session_manager)),
+            );
+            Ok(McpClient::new_with_transport(
+                server.name.clone(),
+                transport,
+                Some(Arc::clone(session_manager)),
+                secrets,
+                user_id,
+                Some(server),
+            ))
         }
     }
 }
@@ -134,4 +139,84 @@ mod tests {
             "non-OAuth HTTP clients must carry a session manager"
         );
     }
+
+    /// Regression test: the factory must wire the session manager into the
+    /// *transport*, not just the client. Otherwise the transport never
+    /// captures `Mcp-Session-Id` from responses and subsequent requests
+    /// lack the header, causing the server to reject them.
+    #[tokio::test]
+    async fn test_factory_non_oauth_http_transport_captures_session_id() {
+        use axum::http::header::HeaderName;
+        use axum::{Router, http::StatusCode, response::IntoResponse, routing::post};
+        use tokio::net::TcpListener;
+
+        const SESSION_ID: &str = "test-session-abc123";
+
+        async fn session_echo() -> impl IntoResponse {
+            let body = serde_json::json!({
+                "jsonrpc": "2.0",
+                "id": 1,
+                "result": {}
+            })
+            .to_string();
+            (
+                StatusCode::OK,
+                [(
+                    HeaderName::from_static("mcp-session-id"),
+                    SESSION_ID.to_string(),
+                )],
+                body,
+            )
+        }
+
+        let app = Router::new().route("/", post(session_echo));
+        let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+        let addr = listener.local_addr().unwrap();
+        let url = format!("http://127.0.0.1:{}", addr.port());
+
+        tokio::spawn(async move {
+            axum::serve(listener, app).await.unwrap();
+        });
+
+        let server = McpServerConfig::new("session-test", &url);
+        let session_manager = Arc::new(McpSessionManager::new());
+        let process_manager = Arc::new(McpProcessManager::new());
+
+        let client = create_client_from_config(
+            server,
+            &session_manager,
+            &process_manager,
+            None,
+            "test-user",
+        )
+        .await
+        .expect("factory should succeed for HTTP config");
+
+        // Pre-create a session entry so that update_session_id has something to update.
+        // In production, the MCP initialize handshake calls get_or_create before responses arrive.
+        session_manager.get_or_create("session-test", &url).await;
+
+        // Send a request through the client's transport to trigger session capture.
+        use crate::tools::mcp::protocol::McpRequest;
+        let request = McpRequest {
+            jsonrpc: "2.0".to_string(),
+            id: Some(1),
+            method: "test".to_string(),
+            params: Some(serde_json::json!({})),
+        };
+        let headers = std::collections::HashMap::new();
+        client
+            .transport()
+            .send(&request, &headers)
+            .await
+            .expect("request should succeed");
+
+        // Verify the session manager captured the session ID from the response.
+        let captured = session_manager.get_session_id("session-test").await;
+        assert_eq!(
+            captured.as_deref(),
+            Some(SESSION_ID),
+            "transport must capture Mcp-Session-Id into session manager"
+        );
+    }
 }

src/tools/mcp/http_transport.rs

@@ -494,6 +494,34 @@ mod tests {
         assert_eq!(echoed["authorization"], "Bearer oauth-token");
     }
 
+    /// Regression test for #1436: 202 Accepted responses for notifications
+    /// were parsed as JSON, causing "Failed to parse MCP response" errors
+    /// that broke the MCP session handshake.
+    #[tokio::test]
+    async fn test_wire_202_accepted_for_notification() {
+        use axum::{Router, http::StatusCode, routing::post};
+        use tokio::net::TcpListener;
+
+        async fn accept_notification() -> StatusCode {
+            StatusCode::ACCEPTED
+        }
+
+        let app = Router::new().route("/", post(accept_notification));
+        let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+        let addr = listener.local_addr().unwrap();
+        let url = format!("http://127.0.0.1:{}", addr.port());
+
+        tokio::spawn(async move {
+            axum::serve(listener, app).await.unwrap();
+        });
+
+        let transport = HttpMcpTransport::new(&url, "test-202");
+        let request = McpRequest::initialized_notification();
+        let response = transport.send(&request, &HashMap::new()).await.unwrap();
+        assert!(response.result.is_none());
+        assert!(response.error.is_none());
+    }
+
     #[tokio::test]
     async fn test_wire_custom_auth_preserved_when_no_per_request_auth() {
         let (url, _handle) = spawn_echo_server().await;

src/tools/registry.rs

@@ -383,11 +383,7 @@ impl ToolRegistry {
         job_manager: Option<Arc<ContainerJobManager>>,
         store: Option<Arc<dyn Database>>,
         job_event_tx: Option<
-            tokio::sync::broadcast::Sender<(
+            tokio::sync::broadcast::Sender<(uuid::Uuid, String, ironclaw_common::AppEvent)>,
-                uuid::Uuid,
-                String,
-                crate::channels::web::types::SseEvent,
-            )>,
         >,
         inject_tx: Option<tokio::sync::mpsc::Sender<crate::channels::IncomingMessage>>,
         prompt_queue: Option<PromptQueue>,