feat(workspace): multi-scope workspace reads (#1117)

* feat(workspace): multi-scope workspace reads

Adds the ability for a workspace to read from multiple user scopes
while keeping writes isolated to the primary scope. Configuration
via WORKSPACE_READ_SCOPES env var (comma-separated user IDs).

Includes identity file isolation (read_primary), multi-scope search,
list, and read operations, WorkspaceConfig refactor, and comprehensive
integration tests.

* fix: address review feedback for multi-scope workspace reads

- fix(memory): deduplicate timezone parsing for daily_log target
  parse_timezone was called twice when target was "daily_log" without a
  layer — once in path resolution, again in the fallback. Now computed
  once and reused.

- fix(config): add character validation for WORKSPACE_READ_SCOPES and
  layer scopes — both enforce [a-zA-Z0-9_-] to prevent path traversal
  or injection via scope strings used as user_id in SQL queries.

- fix(config): use chars().take(32) instead of byte-index slicing for
  scope length error messages (UTF-8 safety).

- fix(error): remove unused WorkspaceError::NotFound variant

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* style: downgrade search log to debug, add comments on list iteration

- Downgrade hybrid_search_multi tracing::info! to debug! — fires on
  every multi-scope search with the default backend, too noisy for info
- Add comments explaining why list/list_all iterate per-scope instead
  of using _multi trait methods (identity path filtering needs scope
  attribution that merged results lose)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

---------

Co-authored-by: [email protected] <[email protected]>
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
This commit is contained in:
standardtoaster
2026-03-22 22:56:26 -07:00
committed by GitHub
co-authored by [email protected] <[email protected]> Claude Opus 4.6
parent 8f6999a074
commit d9358b0fa9
15 changed files with 1964 additions and 64 deletions
+11
View File
@@ -325,9 +325,20 @@ impl AppBuilder {
};
let mut ws = Workspace::new_with_db(workspace_user_id, db.clone())
.with_search_config(&self.config.search);
if let Some(ref emb) = embeddings {
ws = ws.with_embeddings_cached(emb.clone(), emb_cache_config);
}
// Wire workspace-level settings (read scopes, memory layers)
if !self.config.workspace.read_scopes.is_empty() {
ws = ws.with_additional_read_scopes(self.config.workspace.read_scopes.clone());
tracing::info!(
user_id = workspace_user_id,
read_scopes = ?ws.read_user_ids(),
"Workspace configured with multi-scope reads"
);
}
ws = ws.with_memory_layers(self.config.workspace.memory_layers.clone());
let ws = Arc::new(ws);
tools.register_memory_tools(Arc::clone(&ws));
+7 -1
View File
@@ -1822,7 +1822,13 @@ async fn memory_write_handler(
"Workspace not available".to_string(),
))?;
// Route through layer-aware methods when a layer is specified
// 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
+8 -7
View File
@@ -24,7 +24,7 @@ mod skills;
mod transcription;
mod tunnel;
mod wasm;
mod workspace;
pub(crate) mod workspace;
use std::collections::HashMap;
use std::sync::{LazyLock, Mutex, Once};
@@ -178,9 +178,7 @@ impl Config {
},
transcription: TranscriptionConfig::default(),
search: WorkspaceSearchConfig::default(),
workspace: WorkspaceConfig {
memory_layers: vec![],
},
workspace: WorkspaceConfig::default(),
observability: crate::observability::ObservabilityConfig::default(),
relay: None,
}
@@ -313,11 +311,14 @@ 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.
let workspace_user_id = channels
.gateway
.as_ref()
.map(|gw| gw.user_id.clone())
.unwrap_or_else(|| "default".to_string());
.map(|gw| gw.user_id.as_str())
.unwrap_or("default");
let workspace = WorkspaceConfig::resolve(workspace_user_id)?;
Ok(Self {
owner_id: owner_id.clone(),
@@ -339,7 +340,7 @@ impl Config {
skills: SkillsConfig::resolve()?,
transcription: TranscriptionConfig::resolve(settings)?,
search: WorkspaceSearchConfig::resolve()?,
workspace: WorkspaceConfig::resolve(&workspace_user_id)?,
workspace,
observability: crate::observability::ObservabilityConfig {
backend: std::env::var("OBSERVABILITY_BACKEND").unwrap_or_else(|_| "none".into()),
},
+68 -7
View File
@@ -2,18 +2,29 @@ use crate::config::helpers::optional_env;
use crate::error::ConfigError;
use crate::workspace::layer::MemoryLayer;
/// Workspace memory configuration.
/// Workspace-level configuration (memory layers, read scopes).
///
/// Controls memory layer definitions for privacy-aware writes.
/// Layers are parsed from the `MEMORY_LAYERS` env var (JSON array)
/// or default to a single private layer scoped to the gateway user.
#[derive(Debug, Clone)]
/// Parsed from environment variables. Lives outside of `GatewayConfig`
/// so that non-gateway channels can eventually use the same settings.
#[derive(Debug, Clone, Default)]
pub struct WorkspaceConfig {
/// Memory layer definitions (JSON in `MEMORY_LAYERS` env var, or defaults).
pub memory_layers: Vec<MemoryLayer>,
/// Additional user scopes for workspace reads.
///
/// When set, the workspace can read (search, read, list) from these
/// additional user scopes while writes remain isolated to the primary
/// `user_id`. Parsed from `WORKSPACE_READ_SCOPES` (comma-separated).
pub read_scopes: Vec<String>,
}
impl WorkspaceConfig {
pub(crate) fn resolve(user_id: &str) -> Result<Self, ConfigError> {
/// Resolve workspace config from environment variables.
///
/// `user_id` is used to derive default memory layers when `MEMORY_LAYERS`
/// is not set.
pub fn resolve(user_id: &str) -> Result<Self, ConfigError> {
// --- Memory layers ---
let memory_layers: Vec<MemoryLayer> = match optional_env("MEMORY_LAYERS")? {
Some(json_str) => {
serde_json::from_str(&json_str).map_err(|e| ConfigError::InvalidValue {
@@ -57,6 +68,20 @@ impl WorkspaceConfig {
message: format!("layer '{}' has an empty scope", layer.name),
});
}
if !layer
.scope
.chars()
.all(|c| c.is_ascii_alphanumeric() || c == '_' || c == '-')
{
return Err(ConfigError::InvalidValue {
key: "MEMORY_LAYERS".to_string(),
message: format!(
"layer '{}' scope '{}' contains invalid characters \
(allowed: a-z, A-Z, 0-9, _, -)",
layer.name, layer.scope
),
});
}
}
// Check for duplicate layer names
@@ -72,7 +97,43 @@ impl WorkspaceConfig {
}
}
Ok(Self { memory_layers })
// --- Read scopes ---
let read_scopes: Vec<String> = optional_env("WORKSPACE_READ_SCOPES")?
.map(|s| {
s.split(',')
.map(|s| s.trim().to_string())
.filter(|s| !s.is_empty())
.collect()
})
.unwrap_or_default();
for scope in &read_scopes {
if scope.len() > 128 {
let prefix: String = scope.chars().take(32).collect();
return Err(ConfigError::InvalidValue {
key: "WORKSPACE_READ_SCOPES".to_string(),
message: format!("scope '{prefix}...' exceeds 128 characters"),
});
}
if !scope
.chars()
.all(|c| c.is_ascii_alphanumeric() || c == '_' || c == '-')
{
return Err(ConfigError::InvalidValue {
key: "WORKSPACE_READ_SCOPES".to_string(),
message: format!(
"scope '{}' contains invalid characters \
(allowed: a-z, A-Z, 0-9, _, -)",
scope
),
});
}
}
Ok(Self {
memory_layers,
read_scopes,
})
}
}
+97
View File
@@ -644,6 +644,103 @@ pub trait WorkspaceStore: Send + Sync {
embedding: Option<&[f32]>,
config: &SearchConfig,
) -> Result<Vec<SearchResult>, WorkspaceError>;
// ==================== Multi-scope read methods ====================
//
// Default implementations loop over user_ids calling single-scope methods,
// then merge results. Backends can override with efficient SQL (e.g.,
// `WHERE user_id = ANY($1::text[])`).
/// Hybrid search across multiple user scopes, merging results by score.
///
/// **Note:** The default implementation calls `hybrid_search` per scope and
/// merges by raw score. Because RRF scores are normalized independently
/// within each scope, scores are not directly comparable across scopes.
/// The Postgres backend overrides this with a single combined query that
/// applies RRF once to the unified result set.
async fn hybrid_search_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
query: &str,
embedding: Option<&[f32]>,
config: &SearchConfig,
) -> Result<Vec<SearchResult>, WorkspaceError> {
if user_ids.len() > 1 {
tracing::debug!(
scope_count = user_ids.len(),
"hybrid_search_multi: using default per-scope RRF merge; \
cross-scope score comparison may be unreliable"
);
}
let mut all_results = Vec::new();
for uid in user_ids {
let results = self
.hybrid_search(uid, agent_id, query, embedding, config)
.await?;
all_results.extend(results);
}
// Re-sort by score descending and truncate to limit
all_results.sort_by(|a, b| {
b.score
.partial_cmp(&a.score)
.unwrap_or(std::cmp::Ordering::Equal)
});
all_results.truncate(config.limit);
Ok(all_results)
}
/// List all file paths across multiple user scopes.
async fn list_all_paths_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
) -> Result<Vec<String>, WorkspaceError> {
let mut all_paths = Vec::new();
for uid in user_ids {
let paths = self.list_all_paths(uid, agent_id).await?;
all_paths.extend(paths);
}
all_paths.sort();
all_paths.dedup();
Ok(all_paths)
}
/// Get a document by path, searching across multiple user scopes.
///
/// Returns the first match found (tries each user_id in order).
async fn get_document_by_path_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
path: &str,
) -> Result<MemoryDocument, WorkspaceError> {
for uid in user_ids {
match self.get_document_by_path(uid, agent_id, path).await {
Ok(doc) => return Ok(doc),
Err(WorkspaceError::DocumentNotFound { .. }) => continue,
Err(e) => return Err(e),
}
}
Err(WorkspaceError::DocumentNotFound {
doc_type: path.to_string(),
user_id: format!("[{}]", user_ids.join(", ")),
})
}
/// List directory contents across multiple user scopes.
async fn list_directory_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
directory: &str,
) -> Result<Vec<WorkspaceEntry>, WorkspaceError> {
let mut all_entries = Vec::new();
for uid in user_ids {
all_entries.extend(self.list_directory(uid, agent_id, directory).await?);
}
Ok(crate::workspace::merge_workspace_entries(all_entries))
}
}
/// Backend-agnostic database supertrait.
+45
View File
@@ -717,4 +717,49 @@ impl WorkspaceStore for PgBackend {
.hybrid_search(user_id, agent_id, query, embedding, config)
.await
}
// Optimized multi-scope overrides using `ANY($1::text[])` SQL.
async fn hybrid_search_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
query: &str,
embedding: Option<&[f32]>,
config: &SearchConfig,
) -> Result<Vec<SearchResult>, WorkspaceError> {
self.repo
.hybrid_search_multi(user_ids, agent_id, query, embedding, config)
.await
}
async fn list_all_paths_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
) -> Result<Vec<String>, WorkspaceError> {
self.repo.list_all_paths_multi(user_ids, agent_id).await
}
async fn get_document_by_path_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
path: &str,
) -> Result<MemoryDocument, WorkspaceError> {
self.repo
.get_document_by_path_multi(user_ids, agent_id, path)
.await
}
async fn list_directory_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
directory: &str,
) -> Result<Vec<WorkspaceEntry>, WorkspaceError> {
self.repo
.list_directory_multi(user_ids, agent_id, directory)
.await
}
}
-3
View File
@@ -304,9 +304,6 @@ pub enum WorkspaceError {
#[error("I/O error: {reason}")]
IoError { reason: String },
#[error("Not found: {path}")]
NotFound { path: String },
#[error("Layer not found: {name}")]
LayerNotFound { name: String },
+3 -4
View File
@@ -271,12 +271,13 @@ impl Tool for MemoryWriteTool {
.and_then(|v| v.as_bool())
.unwrap_or(false);
// Parse timezone once for targets that need it (daily_log).
let tz = crate::timezone::parse_timezone(&ctx.user_timezone).unwrap_or(chrono_tz::Tz::UTC);
// Resolve the target to a workspace path
let resolved_path = match target {
"memory" => paths::MEMORY.to_string(),
"daily_log" => {
let tz = crate::timezone::parse_timezone(&ctx.user_timezone)
.unwrap_or(chrono_tz::Tz::UTC);
let now = chrono::Utc::now().with_timezone(&tz);
format!("daily/{}.md", now.format("%Y-%m-%d"))
}
@@ -318,8 +319,6 @@ impl Tool for MemoryWriteTool {
}
}
"daily_log" => {
let tz = crate::timezone::parse_timezone(&ctx.user_timezone)
.unwrap_or(chrono_tz::Tz::UTC);
self.workspace
.append_daily_log_tz(content, tz)
.await
+21
View File
@@ -91,6 +91,27 @@ Default k=60. Results from both methods are combined, with documents appearing i
- **PostgreSQL:** `ts_rank_cd` for FTS, pgvector cosine distance for vectors, full RRF
- **libSQL:** FTS5 for keyword search + vector search via `libsql_vector_idx` (dimension set dynamically by `ensure_vector_index()` during startup)
## Multi-Scope Reads & Identity Isolation
When a workspace has additional read scopes (via `with_additional_read_scopes`), read operations can span multiple user scopes — a user with scopes `["alice", "shared"]` can read documents from both.
**Identity files are exempt from multi-scope reads.** The system prompt reads identity and configuration files from the **primary scope only** (`read_primary()`), never from secondary scopes:
| File | Read method | Rationale |
|------|------------|-----------|
| AGENTS.md | `read_primary()` | Agent instructions are per-user |
| SOUL.md | `read_primary()` | Core values are per-user |
| USER.md | `read_primary()` | User context is per-user |
| IDENTITY.md | `read_primary()` | Identity is per-user |
| TOOLS.md | `read_primary()` | Tool config is per-user |
| BOOTSTRAP.md | `read_primary()` | Onboarding is per-user |
| MEMORY.md | `read()` | Shared memory is a feature |
| daily/*.md | `read()` | Shared daily logs are a feature |
**Why:** Without this, a user with read access to another scope could silently inherit that scope's identity if their own copy is missing. The agent would present itself as the wrong user — a correctness and security issue.
**Design rule:** If you want shared identity across users, seed the same content into each user's scope at setup time. Don't rely on multi-scope fallback for identity files.
## Heartbeat System
Proactive periodic execution (default: 30 minutes):
+167 -4
View File
@@ -37,6 +37,25 @@ pub mod paths {
pub const ASSISTANT_DIRECTIVES: &str = "context/assistant-directives.md";
}
/// Paths treated as identity documents for multi-scope isolation.
///
/// These files are always read from the primary scope only — never from
/// secondary read scopes. This prevents silent identity inheritance
/// (e.g., user A accidentally presenting as user B).
pub const IDENTITY_PATHS: &[&str] = &[
paths::IDENTITY,
paths::SOUL,
paths::AGENTS,
paths::USER,
paths::TOOLS,
paths::BOOTSTRAP,
];
/// Check if a path is an identity document that must be isolated to primary scope.
pub fn is_identity_path(path: &str) -> bool {
IDENTITY_PATHS.contains(&path)
}
/// A memory document stored in the database.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct MemoryDocument {
@@ -101,10 +120,7 @@ impl MemoryDocument {
/// Check if this is a well-known identity document.
pub fn is_identity_document(&self) -> bool {
matches!(
self.path.as_str(),
paths::IDENTITY | paths::SOUL | paths::AGENTS | paths::USER
)
is_identity_path(&self.path)
}
}
@@ -128,6 +144,42 @@ impl WorkspaceEntry {
}
}
/// Merge workspace entries from multiple scopes into a deduplicated, sorted list.
///
/// When the same path appears in multiple scopes:
/// - Keeps the most recent `updated_at`
/// - If any scope marks it as a directory, the merged entry is a directory
pub fn merge_workspace_entries(
entries: impl IntoIterator<Item = WorkspaceEntry>,
) -> Vec<WorkspaceEntry> {
let mut seen = std::collections::HashMap::new();
for entry in entries {
seen.entry(entry.path.clone())
.and_modify(|existing: &mut WorkspaceEntry| {
// Keep the most recent updated_at (and its content_preview)
if let (Some(existing_ts), Some(new_ts)) = (&existing.updated_at, &entry.updated_at)
{
if new_ts > existing_ts {
existing.updated_at = Some(*new_ts);
existing.content_preview = entry.content_preview.clone();
}
} else if existing.updated_at.is_none() {
existing.updated_at = entry.updated_at;
existing.content_preview = entry.content_preview.clone();
}
// If either is a directory, mark as directory
if entry.is_directory {
existing.is_directory = true;
existing.content_preview = None;
}
})
.or_insert(entry);
}
let mut result: Vec<WorkspaceEntry> = seen.into_values().collect();
result.sort_by(|a, b| a.path.cmp(&b.path));
result
}
/// A chunk of a memory document for search indexing.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct MemoryChunk {
@@ -226,4 +278,115 @@ mod tests {
};
assert_eq!(entry.name(), "alpha");
}
#[test]
fn test_merge_workspace_entries_empty() {
let result = merge_workspace_entries(vec![]);
assert!(result.is_empty());
}
#[test]
fn test_merge_workspace_entries_keeps_newer_timestamp_and_preview() {
use chrono::TimeZone;
let old_ts = chrono::Utc.with_ymd_and_hms(2025, 1, 1, 0, 0, 0).unwrap();
let new_ts = chrono::Utc.with_ymd_and_hms(2025, 6, 1, 0, 0, 0).unwrap();
let entries = vec![
WorkspaceEntry {
path: "notes.md".to_string(),
is_directory: false,
updated_at: Some(old_ts),
content_preview: Some("old".to_string()),
},
WorkspaceEntry {
path: "notes.md".to_string(),
is_directory: false,
updated_at: Some(new_ts),
content_preview: Some("new".to_string()),
},
];
let result = merge_workspace_entries(entries);
assert_eq!(result.len(), 1);
assert_eq!(result[0].updated_at, Some(new_ts));
assert_eq!(result[0].content_preview, Some("new".to_string()));
}
#[test]
fn test_merge_workspace_entries_directory_wins() {
let entries = vec![
WorkspaceEntry {
path: "projects".to_string(),
is_directory: false,
updated_at: None,
content_preview: Some("file content".to_string()),
},
WorkspaceEntry {
path: "projects".to_string(),
is_directory: true,
updated_at: None,
content_preview: None,
},
];
let result = merge_workspace_entries(entries);
assert_eq!(result.len(), 1);
assert!(result[0].is_directory);
assert!(result[0].content_preview.is_none());
}
#[test]
fn test_merge_workspace_entries_fills_missing_timestamp() {
use chrono::TimeZone;
let ts = chrono::Utc.with_ymd_and_hms(2025, 3, 1, 0, 0, 0).unwrap();
let entries = vec![
WorkspaceEntry {
path: "a.md".to_string(),
is_directory: false,
updated_at: None,
content_preview: None,
},
WorkspaceEntry {
path: "a.md".to_string(),
is_directory: false,
updated_at: Some(ts),
content_preview: None,
},
];
let result = merge_workspace_entries(entries);
assert_eq!(result.len(), 1);
assert_eq!(result[0].updated_at, Some(ts));
}
#[test]
fn test_merge_workspace_entries_sorted_by_path() {
let entries = vec![
WorkspaceEntry {
path: "z.md".to_string(),
is_directory: false,
updated_at: None,
content_preview: None,
},
WorkspaceEntry {
path: "a.md".to_string(),
is_directory: false,
updated_at: None,
content_preview: None,
},
WorkspaceEntry {
path: "m.md".to_string(),
is_directory: false,
updated_at: None,
content_preview: None,
},
];
let result = merge_workspace_entries(entries);
assert_eq!(result.len(), 3);
assert_eq!(result[0].path, "a.md");
assert_eq!(result[1].path, "m.md");
assert_eq!(result[2].path, "z.md");
}
}
+362 -38
View File
@@ -52,7 +52,10 @@ mod repository;
mod search;
pub use chunker::{ChunkConfig, chunk_document};
pub use document::{MemoryChunk, MemoryDocument, WorkspaceEntry, paths};
pub use document::{
IDENTITY_PATHS, MemoryChunk, MemoryDocument, WorkspaceEntry, is_identity_path,
merge_workspace_entries, paths,
};
pub use embedding_cache::{CachedEmbeddingProvider, EmbeddingCacheConfig};
pub use embeddings::{
EmbeddingProvider, MockEmbeddings, NearAiEmbeddings, OllamaEmbeddings, OpenAiEmbeddings,
@@ -320,6 +323,48 @@ impl WorkspaceStorage {
}
}
}
// ==================== Multi-scope read methods ====================
async fn hybrid_search_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
query: &str,
embedding: Option<&[f32]>,
config: &SearchConfig,
) -> Result<Vec<SearchResult>, WorkspaceError> {
match self {
#[cfg(feature = "postgres")]
Self::Repo(repo) => {
repo.hybrid_search_multi(user_ids, agent_id, query, embedding, config)
.await
}
Self::Db(db) => {
db.hybrid_search_multi(user_ids, agent_id, query, embedding, config)
.await
}
}
}
async fn get_document_by_path_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
path: &str,
) -> Result<MemoryDocument, WorkspaceError> {
match self {
#[cfg(feature = "postgres")]
Self::Repo(repo) => {
repo.get_document_by_path_multi(user_ids, agent_id, path)
.await
}
Self::Db(db) => {
db.get_document_by_path_multi(user_ids, agent_id, path)
.await
}
}
}
}
/// Default template seeded into HEARTBEAT.md on first access.
@@ -340,9 +385,20 @@ const BOOTSTRAP_SEED: &str = include_str!("seeds/BOOTSTRAP.md");
/// Each workspace is scoped to a user (and optionally an agent).
/// Documents are persisted to the database and indexed for search.
/// Supports both PostgreSQL (via Repository) and libSQL (via Database trait).
///
/// ## Multi-scope reads
///
/// By default, a workspace reads from and writes to a single `user_id`.
/// With `with_additional_read_scopes`, read operations (search, read, list)
/// can span multiple user scopes while writes remain isolated to the primary
/// `user_id`. This enables cross-tenant read access (e.g., a user reading
/// from both their own workspace and a "shared" workspace).
pub struct Workspace {
/// User identifier (from channel).
/// User identifier (from channel). All writes go to this scope.
user_id: String,
/// User identifiers for read operations. Includes `user_id` as the first
/// element, plus any additional scopes added via `with_additional_read_scopes`.
read_user_ids: Vec<String>,
/// Optional agent ID for multi-agent isolation.
agent_id: Option<Uuid>,
/// Database storage backend.
@@ -371,6 +427,7 @@ impl Workspace {
let user_id_str = user_id.into();
let memory_layers = crate::workspace::layer::MemoryLayer::default_for_user(&user_id_str);
Self {
read_user_ids: vec![user_id_str.clone()],
user_id: user_id_str,
agent_id: None,
storage: WorkspaceStorage::Repo(Repository::new(pool)),
@@ -390,6 +447,7 @@ impl Workspace {
let user_id_str = user_id.into();
let memory_layers = crate::workspace::layer::MemoryLayer::default_for_user(&user_id_str);
Self {
read_user_ids: vec![user_id_str.clone()],
user_id: user_id_str,
agent_id: None,
storage: WorkspaceStorage::Db(db),
@@ -474,6 +532,12 @@ impl Workspace {
///
/// Also updates read_user_ids to include all layer scopes.
pub fn with_memory_layers(mut self, layers: Vec<crate::workspace::layer::MemoryLayer>) -> Self {
// Add layer scopes to read_user_ids (same dedup logic as with_additional_read_scopes)
for layer in &layers {
if !self.read_user_ids.contains(&layer.scope) {
self.read_user_ids.push(layer.scope.clone());
}
}
self.memory_layers = layers;
self
}
@@ -496,11 +560,37 @@ impl Workspace {
&self.memory_layers
}
/// Get the user ID.
/// Add additional user scopes for read operations.
///
/// The primary `user_id` is always included. Additional scopes allow
/// read operations (search, read, list) to span multiple tenants while
/// writes remain isolated to the primary scope.
///
/// Duplicate scopes are ignored.
pub fn with_additional_read_scopes(mut self, scopes: Vec<String>) -> Self {
for scope in scopes {
if !self.read_user_ids.contains(&scope) {
self.read_user_ids.push(scope);
}
}
self
}
/// Get the user ID (primary scope for writes).
pub fn user_id(&self) -> &str {
&self.user_id
}
/// Get the user IDs used for read operations.
pub fn read_user_ids(&self) -> &[String] {
&self.read_user_ids
}
/// Whether this workspace has multiple read scopes.
fn is_multi_scope(&self) -> bool {
self.read_user_ids.len() > 1
}
/// Get the agent ID.
pub fn agent_id(&self) -> Option<Uuid> {
self.agent_id
@@ -518,6 +608,33 @@ impl Workspace {
/// println!("{}", doc.content);
/// ```
pub async fn read(&self, path: &str) -> Result<MemoryDocument, WorkspaceError> {
let path = normalize_path(path);
if self.is_multi_scope() && is_identity_path(&path) {
// Identity files must only come from the primary scope.
self.storage
.get_document_by_path(&self.user_id, self.agent_id, &path)
.await
} else if self.is_multi_scope() {
self.storage
.get_document_by_path_multi(&self.read_user_ids, self.agent_id, &path)
.await
} else {
self.storage
.get_document_by_path(&self.user_id, self.agent_id, &path)
.await
}
}
/// Read a file from the **primary scope only**, ignoring additional read scopes.
///
/// Use this for identity and configuration files (AGENTS.md, SOUL.md, USER.md,
/// IDENTITY.md, TOOLS.md, BOOTSTRAP.md) where inheriting content from another
/// scope would be a correctness/security issue — the agent must never silently
/// present itself as the wrong user.
///
/// For memory files that should span scopes (MEMORY.md, daily logs), use
/// [`read`] instead.
pub async fn read_primary(&self, path: &str) -> Result<MemoryDocument, WorkspaceError> {
let path = normalize_path(path);
self.storage
.get_document_by_path(&self.user_id, self.agent_id, &path)
@@ -556,6 +673,9 @@ impl Workspace {
/// Uses a single `\n` separator (suitable for log-style entries).
/// For semantic separation (e.g., memory entries), use `append_memory()`
/// which uses `\n\n`.
///
/// Uses a read-modify-write pattern that is not concurrency-safe:
/// concurrent appends to the same path may lose writes.
pub async fn append(&self, path: &str, content: &str) -> Result<(), WorkspaceError> {
let path = normalize_path(path);
// Scan system-prompt-injected files for prompt injection.
@@ -676,6 +796,20 @@ impl Workspace {
}
/// Write to a layer, with append semantics.
///
/// Note: privacy classification only examines the new `content`, not the
/// full document after concatenation. See [`PatternPrivacyClassifier`]
/// limitations for details.
///
/// When a privacy redirect occurs, the append targets a **separate
/// document** in the private scope at the same path — the shared-scope
/// document is left unmodified. Subsequent multi-scope reads will return
/// the private copy (primary scope wins), effectively shadowing the
/// shared document at that path. The `WriteResult::redirected` flag
/// indicates when this has happened.
///
/// Uses a read-modify-write pattern that is not concurrency-safe:
/// concurrent appends to the same path may lose writes.
pub async fn append_to_layer(
&self,
layer_name: &str,
@@ -706,13 +840,25 @@ impl Workspace {
}
/// Check if a file exists.
///
/// When multi-scope reads are configured, checks across all read scopes.
pub async fn exists(&self, path: &str) -> Result<bool, WorkspaceError> {
let path = normalize_path(path);
match self
.storage
.get_document_by_path(&self.user_id, self.agent_id, &path)
.await
{
let result = if self.is_multi_scope() && is_identity_path(&path) {
// Identity files only checked in primary scope.
self.storage
.get_document_by_path(&self.user_id, self.agent_id, &path)
.await
} else if self.is_multi_scope() {
self.storage
.get_document_by_path_multi(&self.read_user_ids, self.agent_id, &path)
.await
} else {
self.storage
.get_document_by_path(&self.user_id, self.agent_id, &path)
.await
};
match result {
Ok(_) => Ok(true),
Err(WorkspaceError::DocumentNotFound { .. }) => Ok(false),
Err(e) => Err(e),
@@ -747,16 +893,55 @@ impl Workspace {
/// ```
pub async fn list(&self, directory: &str) -> Result<Vec<WorkspaceEntry>, WorkspaceError> {
let directory = normalize_directory(directory);
self.storage
.list_directory(&self.user_id, self.agent_id, &directory)
.await
if self.is_multi_scope() {
// Iterate per-scope rather than using list_directory_multi because
// we need to filter identity paths from secondary scopes only — the
// merged _multi result loses scope attribution.
let primary = self
.storage
.list_directory(&self.user_id, self.agent_id, &directory)
.await?;
let mut all_entries = primary;
for scope in &self.read_user_ids[1..] {
let entries = self
.storage
.list_directory(scope, self.agent_id, &directory)
.await?;
all_entries.extend(entries.into_iter().filter(|e| !is_identity_path(&e.path)));
}
Ok(merge_workspace_entries(all_entries))
} else {
self.storage
.list_directory(&self.user_id, self.agent_id, &directory)
.await
}
}
/// List all files recursively (flat list of all paths).
///
/// When multi-scope reads are configured, lists across all read scopes.
pub async fn list_all(&self) -> Result<Vec<String>, WorkspaceError> {
self.storage
.list_all_paths(&self.user_id, self.agent_id)
.await
if self.is_multi_scope() {
// Iterate per-scope rather than using list_all_paths_multi because
// we need to filter identity paths from secondary scopes only.
// Primary scope: all paths. Secondary scopes: filter identity paths.
let mut all_paths = self
.storage
.list_all_paths(&self.user_id, self.agent_id)
.await?;
for scope in &self.read_user_ids[1..] {
let paths = self.storage.list_all_paths(scope, self.agent_id).await?;
all_paths.extend(paths.into_iter().filter(|p| !is_identity_path(p)));
}
// Deduplicate and sort
all_paths.sort();
all_paths.dedup();
Ok(all_paths)
} else {
self.storage
.list_all_paths(&self.user_id, self.agent_id)
.await
}
}
// ==================== Convenience Methods ====================
@@ -791,7 +976,7 @@ impl Workspace {
/// comments, which the heartbeat runner treats as "effectively empty"
/// and skips the LLM call.
pub async fn heartbeat_checklist(&self) -> Result<Option<String>, WorkspaceError> {
match self.read(paths::HEARTBEAT).await {
match self.read_primary(paths::HEARTBEAT).await {
Ok(doc) => Ok(Some(doc.content)),
Err(WorkspaceError::DocumentNotFound { .. }) => Ok(Some(HEARTBEAT_SEED.to_string())),
Err(e) => Err(e),
@@ -799,7 +984,29 @@ impl Workspace {
}
/// Helper to read or create a file.
///
/// When multi-scope reads are configured, checks all read scopes before
/// creating. If the file exists in any scope, returns it. If not found in
/// any scope, creates it in the primary (write) scope.
///
/// **Important:** In multi-scope mode, the returned document may belong to
/// a secondary scope. Callers that intend to **write** to the document
/// (via `update_document(doc.id, ...)`) must NOT use this method — use
/// `storage.get_or_create_document_by_path(&self.user_id, ...)` instead
/// to guarantee writes target the primary scope. See `append_memory` for
/// the correct pattern.
async fn read_or_create(&self, path: &str) -> Result<MemoryDocument, WorkspaceError> {
if self.is_multi_scope() {
match self
.storage
.get_document_by_path_multi(&self.read_user_ids, self.agent_id, path)
.await
{
Ok(doc) => return Ok(doc),
Err(WorkspaceError::DocumentNotFound { .. }) => {}
Err(e) => return Err(e),
}
}
self.storage
.get_or_create_document_by_path(&self.user_id, self.agent_id, path)
.await
@@ -811,9 +1018,18 @@ impl Workspace {
///
/// This is for important facts, decisions, and preferences worth
/// remembering long-term.
///
/// Uses `get_or_create_document_by_path` with the primary `user_id`
/// instead of `self.memory()` to guarantee writes always target the
/// primary (write) scope. `self.memory()` delegates to `read_or_create`,
/// which in multi-scope mode may return a document owned by a secondary
/// scope; writing to that document by UUID would violate write isolation.
pub async fn append_memory(&self, entry: &str) -> Result<(), WorkspaceError> {
// Use double newline for memory entries (semantic separation)
let doc = self.memory().await?;
// Always get/create in the primary scope to preserve write isolation.
let doc = self
.storage
.get_or_create_document_by_path(&self.user_id, self.agent_id, paths::MEMORY)
.await?;
let new_content = if doc.content.is_empty() {
entry.to_string()
} else {
@@ -905,9 +1121,16 @@ impl Workspace {
// Safety net: if `profile_onboarding_completed` was already set (the
// LLM completed onboarding but forgot to delete BOOTSTRAP.md), skip
// injection to avoid repeating the first-run ritual.
//
// Identity and config files use read_primary() to prevent cross-scope
// bleed in multi-scope workspaces. Without this, a user with read access
// to other scopes could silently inherit another user's identity if their
// own copy is missing — the agent would present as the wrong person.
// Memory files (MEMORY.md, daily logs) intentionally use multi-scope
// read() since sharing memory across scopes is a feature.
let bootstrap_injected = if self.is_bootstrap_completed() {
if self
.read(paths::BOOTSTRAP)
.read_primary(paths::BOOTSTRAP)
.await
.is_ok_and(|d| !d.content.is_empty())
{
@@ -917,7 +1140,7 @@ impl Workspace {
);
}
false
} else if let Ok(doc) = self.read(paths::BOOTSTRAP).await
} else if let Ok(doc) = self.read_primary(paths::BOOTSTRAP).await
&& !doc.content.is_empty()
{
parts.push(format!("## First-Run Bootstrap\n\n{}", doc.content));
@@ -926,7 +1149,8 @@ impl Workspace {
false
};
// Load identity files in order of importance
// Load identity files in order of importance.
// These MUST use read_primary() — see comment above.
let identity_files = [
(paths::AGENTS, "## Agent Instructions"),
(paths::SOUL, "## Core Values"),
@@ -935,7 +1159,7 @@ impl Workspace {
];
for (path, header) in identity_files {
if let Ok(doc) = self.read(path).await
if let Ok(doc) = self.read_primary(path).await
&& !doc.content.is_empty()
{
parts.push(format!("{}\n\n{}", header, doc.content));
@@ -944,7 +1168,8 @@ impl Workspace {
// Tool notes: environment-specific guidance the agent or user has written.
// TOOLS.md does not control tool availability; it is guidance only.
if let Ok(doc) = self.read(paths::TOOLS).await
// Uses read_primary() — tool config is per-user, not inherited.
if let Ok(doc) = self.read_primary(paths::TOOLS).await
&& !doc.content.is_empty()
{
parts.push(format!("## Tool Notes\n\n{}", doc.content));
@@ -1235,6 +1460,8 @@ impl Workspace {
}
/// Search with custom configuration.
///
/// When multi-scope reads are configured, searches across all read scopes.
pub async fn search_with_config(
&self,
query: &str,
@@ -1254,15 +1481,46 @@ impl Workspace {
None
};
self.storage
.hybrid_search(
&self.user_id,
self.agent_id,
query,
embedding.as_deref(),
&config,
)
.await
if self.is_multi_scope() {
let results = self
.storage
.hybrid_search_multi(
&self.read_user_ids,
self.agent_id,
query,
embedding.as_deref(),
&config,
)
.await?;
// Post-filter: exclude identity documents from secondary scopes.
// Collect document IDs that are identity paths in secondary scopes.
let mut excluded_doc_ids = std::collections::HashSet::new();
for result in &results {
if is_identity_path(&result.document_path) {
// Check if this document belongs to a secondary scope
match self.storage.get_document_by_id(result.document_id).await {
Ok(doc) if doc.user_id != self.user_id => {
excluded_doc_ids.insert(result.document_id);
}
_ => {}
}
}
}
Ok(results
.into_iter()
.filter(|r| !excluded_doc_ids.contains(&r.document_id))
.collect())
} else {
self.storage
.hybrid_search(
&self.user_id,
self.agent_id,
query,
embedding.as_deref(),
&config,
)
.await
}
}
// ==================== Indexing ====================
@@ -1323,13 +1581,13 @@ impl Workspace {
// Check freshness BEFORE seeding identity files, otherwise the
// seeded files make the workspace look non-fresh and BOOTSTRAP.md
// never gets created.
let is_fresh_workspace = if self.read(paths::BOOTSTRAP).await.is_ok() {
let is_fresh_workspace = if self.read_primary(paths::BOOTSTRAP).await.is_ok() {
false // BOOTSTRAP already exists
} else {
let (agents_res, soul_res, user_res) = tokio::join!(
self.read(paths::AGENTS),
self.read(paths::SOUL),
self.read(paths::USER),
self.read_primary(paths::AGENTS),
self.read_primary(paths::SOUL),
self.read_primary(paths::USER),
);
matches!(agents_res, Err(WorkspaceError::DocumentNotFound { .. }))
&& matches!(soul_res, Err(WorkspaceError::DocumentNotFound { .. }))
@@ -1338,8 +1596,10 @@ impl Workspace {
let mut count = 0;
for (path, content) in seed_files {
// Skip files that already exist (never overwrite user edits)
match self.read(path).await {
// Skip files that already exist in the primary scope (never overwrite user edits).
// Uses read_primary to avoid false positives from secondary scopes —
// a file in another scope should not suppress seeding in this scope.
match self.read_primary(path).await {
Ok(_) => continue,
Err(WorkspaceError::DocumentNotFound { .. }) => {}
Err(e) => {
@@ -1360,7 +1620,8 @@ impl Workspace {
// may already have a profile from a previous install and doesn't need
// onboarding). This prevents existing users from getting a spurious
// first-run ritual after upgrading.
let has_profile = self.read(paths::PROFILE).await.is_ok_and(|d| {
// Uses read_primary() to avoid false positives from secondary scopes.
let has_profile = self.read_primary(paths::PROFILE).await.is_ok_and(|d| {
!d.content.trim().is_empty()
&& serde_json::from_str::<crate::profile::PsychographicProfile>(&d.content).is_ok()
});
@@ -1791,4 +2052,67 @@ mod seed_tests {
"BOOTSTRAP.md should NOT have been seeded with existing profile"
);
}
#[test]
fn test_default_single_scope() {
// Verify backward compatibility: default workspace has single read scope
// matching user_id.
let user_id = "alice";
let read_user_ids = [user_id.to_string()];
assert_eq!(read_user_ids.len(), 1);
assert_eq!(read_user_ids[0], user_id);
}
#[test]
fn test_additional_read_scopes() {
// Verify that additional read scopes are added correctly.
let user_id = "alice".to_string();
let mut read_user_ids = Vec::from([user_id.clone()]);
// Simulate with_additional_read_scopes logic
let scopes = ["shared", "team"];
for scope in scopes {
let s = scope.to_string();
if !read_user_ids.contains(&s) {
read_user_ids.push(s);
}
}
assert_eq!(read_user_ids.len(), 3);
assert_eq!(read_user_ids[0], "alice");
assert_eq!(read_user_ids[1], "shared");
assert_eq!(read_user_ids[2], "team");
}
#[test]
fn test_additional_read_scopes_dedup() {
// Verify that duplicate scopes are ignored.
let user_id = "alice".to_string();
let mut read_user_ids = Vec::from([user_id.clone()]);
let scopes = ["shared", "alice", "shared"];
for scope in scopes {
let s = scope.to_string();
if !read_user_ids.contains(&s) {
read_user_ids.push(s);
}
}
assert_eq!(read_user_ids.len(), 2);
assert_eq!(read_user_ids[0], "alice");
assert_eq!(read_user_ids[1], "shared");
}
#[test]
fn test_is_multi_scope_logic() {
// Test the multi-scope detection logic: > 1 means multi-scope
let single_count = 1_usize;
let multi_count = 2_usize;
// Single scope: not multi
assert!(single_count <= 1);
// Multi scope: is multi
assert!(multi_count > 1);
}
}
+199
View File
@@ -502,4 +502,203 @@ impl Repository {
})
.collect())
}
// ==================== Multi-scope search (optimized SQL) ====================
/// Hybrid search across multiple user scopes with efficient SQL.
///
/// Uses `user_id = ANY($1::text[])` instead of N separate queries.
pub async fn hybrid_search_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
query: &str,
embedding: Option<&[f32]>,
config: &SearchConfig,
) -> Result<Vec<SearchResult>, WorkspaceError> {
let fts_results = if config.use_fts {
self.fts_search_multi(user_ids, agent_id, query, config.pre_fusion_limit)
.await?
} else {
Vec::new()
};
let vector_results = if config.use_vector {
if let Some(embedding) = embedding {
self.vector_search_multi(user_ids, agent_id, embedding, config.pre_fusion_limit)
.await?
} else {
Vec::new()
}
} else {
Vec::new()
};
Ok(fuse_results(fts_results, vector_results, config))
}
/// FTS search across multiple user scopes.
async fn fts_search_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
query: &str,
limit: usize,
) -> Result<Vec<RankedResult>, WorkspaceError> {
let conn = self.conn().await?;
let rows = conn
.query(
r#"
SELECT c.id as chunk_id, c.document_id, d.path as document_path,
c.content,
ts_rank_cd(c.content_tsv, plainto_tsquery('english', $3)) as rank
FROM memory_chunks c
JOIN memory_documents d ON d.id = c.document_id
WHERE d.user_id = ANY($1::text[]) AND d.agent_id IS NOT DISTINCT FROM $2
AND c.content_tsv @@ plainto_tsquery('english', $3)
ORDER BY rank DESC
LIMIT $4
"#,
&[&user_ids, &agent_id, &query, &(limit as i64)],
)
.await
.map_err(|e| WorkspaceError::SearchFailed {
reason: format!("FTS multi-scope query failed: {}", e),
})?;
Ok(rows
.iter()
.enumerate()
.map(|(i, row)| RankedResult {
chunk_id: row.get("chunk_id"),
document_id: row.get("document_id"),
document_path: row.get("document_path"),
content: row.get("content"),
rank: (i + 1) as u32,
})
.collect())
}
/// Vector search across multiple user scopes.
async fn vector_search_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
embedding: &[f32],
limit: usize,
) -> Result<Vec<RankedResult>, WorkspaceError> {
let conn = self.conn().await?;
let embedding_vec = Vector::from(embedding.to_vec());
let rows = conn
.query(
r#"
SELECT c.id as chunk_id, c.document_id, d.path as document_path,
c.content, 1 - (c.embedding <=> $3) as similarity
FROM memory_chunks c
JOIN memory_documents d ON d.id = c.document_id
WHERE d.user_id = ANY($1::text[]) AND d.agent_id IS NOT DISTINCT FROM $2
AND c.embedding IS NOT NULL
ORDER BY c.embedding <=> $3
LIMIT $4
"#,
&[&user_ids, &agent_id, &embedding_vec, &(limit as i64)],
)
.await
.map_err(|e| WorkspaceError::SearchFailed {
reason: format!("Vector multi-scope query failed: {}", e),
})?;
Ok(rows
.iter()
.enumerate()
.map(|(i, row)| RankedResult {
chunk_id: row.get("chunk_id"),
document_id: row.get("document_id"),
document_path: row.get("document_path"),
content: row.get("content"),
rank: (i + 1) as u32,
})
.collect())
}
/// List all file paths across multiple user scopes with a single query.
pub async fn list_all_paths_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
) -> Result<Vec<String>, WorkspaceError> {
let conn = self.conn().await?;
let rows = conn
.query(
r#"
SELECT DISTINCT path FROM memory_documents
WHERE user_id = ANY($1::text[]) AND agent_id IS NOT DISTINCT FROM $2
ORDER BY path
"#,
&[&user_ids, &agent_id],
)
.await
.map_err(|e| WorkspaceError::SearchFailed {
reason: format!("List paths multi-scope failed: {}", e),
})?;
Ok(rows.iter().map(|row| row.get("path")).collect())
}
/// Get a document by path across multiple user scopes.
///
/// Returns the first match (ordered by the input user_ids priority).
pub async fn get_document_by_path_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
path: &str,
) -> Result<MemoryDocument, WorkspaceError> {
let conn = self.conn().await?;
let row = conn
.query_opt(
r#"
SELECT id, user_id, agent_id, path, content,
created_at, updated_at, metadata
FROM memory_documents
WHERE user_id = ANY($1::text[]) AND agent_id IS NOT DISTINCT FROM $2 AND path = $3
ORDER BY array_position($1::text[], user_id)
LIMIT 1
"#,
&[&user_ids, &agent_id, &path],
)
.await
.map_err(|e| WorkspaceError::SearchFailed {
reason: format!("get_document_by_path_multi failed: {}", e),
})?;
match row {
Some(row) => Ok(self.row_to_document(&row)),
None => Err(WorkspaceError::DocumentNotFound {
doc_type: path.to_string(),
user_id: format!("[{}]", user_ids.join(", ")),
}),
}
}
/// List directory contents across multiple user scopes.
///
/// Iterates per scope and merges results. A future migration could add an
/// optimised SQL function, at which point this method can call it directly.
pub async fn list_directory_multi(
&self,
user_ids: &[String],
agent_id: Option<Uuid>,
directory: &str,
) -> Result<Vec<WorkspaceEntry>, WorkspaceError> {
let mut all_entries = Vec::new();
for uid in user_ids {
all_entries.extend(self.list_directory(uid, agent_id, directory).await?);
}
Ok(crate::workspace::merge_workspace_entries(all_entries))
}
}