@openez-graph/cli 0.9.1 → 0.10.0

package/README.md

@@ -69,19 +69,24 @@ openez setup codex              # wire up Codex
 openez setup opencode           # wire up OpenCode
 openez setup windsurf           # wire up Windsurf / Devin Desktop
 openez setup devin              # wire up Devin CLI
+openez config get [key]         # show embedding config (all if no key)
+openez config set <key> <value> # set embedding config value
+openez config list              # list all DB-stored config overrides
 ```
 
+Valid config keys: `embedding.provider`, `embedding.openai_api_key`, `embedding.openai_base_url`, `embedding.openai_model`, `embedding.ollama_base_url`, `embedding.ollama_model`. API keys are encrypted at rest with AES-256-GCM.
+
 ## MCP Tools
 
-| Tool | Description |
-|------|-------------|
-| `list_workspaces` | List all registered workspaces |
-| `code_query` | Hybrid FTS/vector search + graph expansion over indexed code and docs |
-| `code_context` | Get budgeted symbol context with callers, callees, and related files |
-| `graph_neighbors` | Traverse graph edges from a node or label |
-| `memory_recall` | Recall active memory entries and technical decisions |
-| `memory_write` | Write a memory entry (notes, decisions, patterns) |
-| `index_workspace` | Trigger indexing for a workspace |
+| Tool              | Description                                                           |
+| ----------------- | --------------------------------------------------------------------- |
+| `list_workspaces` | List all registered workspaces                                        |
+| `code_query`      | Hybrid FTS/vector search + graph expansion over indexed code and docs |
+| `code_context`    | Get budgeted symbol context with callers, callees, and related files  |
+| `graph_neighbors` | Traverse graph edges from a node or label                             |
+| `memory_recall`   | Recall active memory entries and technical decisions                  |
+| `memory_write`    | Write a memory entry (notes, decisions, patterns)                     |
+| `index_workspace` | Trigger indexing for a workspace                                      |
 
 `memory_query` is accepted as a deprecated compatibility alias for `code_query`, but is not advertised to new clients.
 
@@ -96,29 +101,36 @@ openez setup devin              # wire up Devin CLI
 
 ## Supported languages
 
-| Language | Indexing depth |
-|----------|---------------|
+| Language                | Indexing depth                                         |
+| ----------------------- | ------------------------------------------------------ |
 | TypeScript / JavaScript | Richest — `ts-morph` symbol extraction, imports, calls |
-| Python | Basic top-level symbol extraction |
-| Go | Basic top-level symbol extraction |
-| Rust | Basic top-level symbol extraction |
-| YAML / JSON / TOML | Structure-aware chunking |
-| Markdown | Section-oriented chunking |
+| Python                  | Basic top-level symbol extraction                      |
+| Go                      | Basic top-level symbol extraction                      |
+| Rust                    | Basic top-level symbol extraction                      |
+| YAML / JSON / TOML      | Structure-aware chunking                               |
+| Markdown                | Section-oriented chunking                              |
 
 ## Retrieval quality
 
-Benchmarked on 18 real queries against the openez codebase itself (118 files, 640 chunks):
+Benchmarked on 23 queries (17 keyword + 6 semantic) against the openez codebase (128 files, 810 chunks):
+
+| Metric           | FTS only | FTS + Embedding (bge-m3) |
+| ---------------- | -------: | -----------------------: |
+| Recall@5         |   91.30% |                   95.65% |
+| Keyword queries  |  100.00% |                  100.00% |
+| Semantic queries |   66.67% |                   83.33% |
+| Avg latency      |     5 ms |                   249 ms |
+
+**FTS-only is the default** — 100% recall on keyword queries, 50x faster. **Embedding adds semantic search with +16.67% semantic recall and no keyword regression** via full RRF fusion (FTS weight 2x, vector weight 1x).
 
-| Metric | Value |
-|--------|-------|
-| Recall@5 | 94.44% |
-| MRR | 0.6565 |
-| Avg latency | 38.68 ms |
-| p50 latency | 18.65 ms |
-| Duplicate path rate | 0% |
-| Quality gate | PASS |
+```bash
+# Enable Ollama embeddings (bge-m3 recommended for code search)
+openez config set embedding.provider ollama
+openez config set embedding.ollama_model bge-m3
+openez reindex .
+```
 
-FTS5 with BM25 ranking handles 94% of queries in under 40ms — no embeddings needed for the default path. Embeddings (OpenAI/Ollama) are optional semantic fallback for queries without direct keyword overlap.
+See [BENCHMARK.md](https://github.com/asta-nguyen/openez-graph/blob/main/BENCHMARK.md) for full analysis.
 
 ## Web dashboard
 
@@ -127,6 +139,7 @@ openez serve --web
 ```
 
 Opens a full web dashboard at `http://localhost:17881` with:
+
 - Workspace overview (documents, chunks, nodes, edges)
 - Graph explorer with force-directed layout
 - Query interface for memory retrieval

package/dist/CHANGELOG.md

@@ -5,19 +5,50 @@ All notable changes to OpenEZ Graph are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.10.0] - 2026-08-05
+
+### Fixed
+
+- MCP stdio corruption: retrieval diagnostics now write to `stderr` instead of `stdout`, preserving JSON-RPC framing for `code_query` calls
+- `lint-staged` downgraded from v17 to v16.4.0 for Node.js 20 compatibility (v17 requires Node >=22.22.1)
+- Master key race condition: atomic exclusive file creation (`O_CREAT|O_EXCL`) prevents concurrent processes from generating different keys
+- Master key malformed-file handling: invalid key content now throws an explicit error instead of silently overwriting the file
+- Master key file permissions: existing key files are now `chmod`-ed to `0o600` on read if permissions are too open
+- Embedding dedup: chunks skipped by `input_hash` match now receive a copied embedding row so they appear in vector search results
+- Embedding dedup legacy cleanup: `ROW_NUMBER()` ordering (newest, non-null `input_hash` first) replaces `MIN(id)` to preserve the most useful embedding
+- Composite index `(provider, model, input_hash)` added to speed up embedding dedup lookups on large workspaces
+- `vectorSearch` now wraps `getEmbeddingProvider()` in the same `try` block so initialization failures fall back to FTS instead of aborting `codeQuery`
+- Web API `PUT /api/settings/embedding` validates `embedding.provider` against `none`, `openai`, `ollama` (returns 400 for invalid values)
+- Web API clearing a setting (empty value) now deletes the DB override instead of skipping it, restoring env/default fallback
+- CLI `config get <key>` resolves the effective DB-plus-environment value and prints `not set` when absent
+- CLI sensitive-key masking uses `isSensitiveKey()` consistently across all `config` commands
+- Settings UI syncs external config to form state via `useEffect` instead of setState-during-render; form updates after save refetch
+- Settings UI resets the "Saved!" indicator when any field is edited after a successful save
+- RRF `identity()` function computed once per entry instead of twice (get + set)
+- Benchmark test isolated from the user's real registry DB via temp `AI_MEMORY_REGISTRY_DB_PATH`
+- Benchmark stats workspace path resolved from registry instead of `cwd` to match `codeQuery`
+- `withRetry` dead code replaced with a type-safe unreachable return
+
+### Changed
+
+- README storage section updated from three to four local artifacts (added `master.key`)
+
 ## [0.9.0] - 2026-08-02
 
 ### Added
+
 - MCP server version and Git build identity in the protocol handshake
 - Token budgets for `code_query`, `code_context`, `graph_neighbors`, and `memory_recall`, with compact graph/context responses
 - Live `code_query` token telemetry on the dashboard and benchmark page
 - MCP contract tests for response budgets, multi-workspace attribution, graph context, and startup indexing
 
 ### Changed
+
 - Multi-workspace retrieval now applies one global serialized-response budget and attributes delivered tokens exactly across workspaces
 - CLI npm package now exposes the bundled entry point and installs only `better-sqlite3` at runtime
 
 ### Fixed
+
 - Empty workspaces no longer re-index on every MCP server restart
 - Very small token budgets are rejected instead of returning an oversized response
 - Token-savings telemetry now uses selected full-file tokens minus the actual serialized response size
@@ -25,6 +56,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.8.0] - 2026-08-02
 
 ### Added
+
 - Memory management UI at `/memories` with list, search, detail view, and create dialog
 - Memory API routes: `GET /api/memories` (list + search), `GET /api/memories/:id`, `POST /api/memories`, `DELETE /api/memories/:id`
 - Recent memories section on the dashboard now populated from the workspace database
@@ -33,12 +65,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Integration tests for hybrid retrieval ranking and registry operations
 
 ### Fixed
+
 - Changelog page stuck loading — `findChangelogPath` now walks up from `serverDir` and `cwd` to locate `CHANGELOG.md` in the monorepo root
 - Memory endpoints no longer fail when the first registered workspace has a stale root path
 
 ## [0.7.0] - 2026-08-01
 
 ### Added
+
 - Web dashboard changelog page (`/changelog`) with structured rendering of release notes from `CHANGELOG.md`
 - API endpoint `GET /api/changelog` serving the repo changelog
 - Changelog link in CLI README for npm package page
@@ -47,16 +81,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `memory_recall` MCP tool for retrieving active technical decisions and learned notes written by `memory_write`
 
 ### Changed
+
 - Renamed the public `memory_query` MCP tool to `code_query`; the old name remains a deprecated, hidden compatibility alias
 - Code retrieval now fuses both FTS and vector results, and `code_context` supports result/token budgets
 
 ## [0.6.1] - 2026-08-01
 
 ### Added
+
 - Syntax-highlighted context blocks in the web Query page (`prism-react-renderer`), with per-source file metadata (path, line range, score) and line numbers
 - Markdown context blocks now render fenced code blocks with their own language highlighting (e.g. ` ```bash ` gets real bash highlighting)
 
 ### Fixed
+
 - Dark boxes obscuring markdown inline code in context blocks (stripped token `backgroundColor` from the nightOwl theme)
 
 ## [0.6.0] - 2026-08-01
@@ -64,9 +101,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 Remediation release — index/graph correctness, data protection, and web flow fixes.
 
 ### Security
+
 - Web API and CLI dashboard now bind to loopback (`127.0.0.1`) by default (FIX-01)
 
 ### Fixed
+
 - Full reindex no longer wipes memories, query logs, and run history — only rebuildable index artifacts are reset (FIX-02)
 - Incremental indexing preserves inbound graph edges (calls/imports) to symbols in changed files; symbol and file node identities are now stable across re-parses (FIX-03)
 - Graph edges are deduplicated by `(from, to, type)` with a unique SQLite index and `ON CONFLICT DO NOTHING` inserts (FIX-04)
@@ -75,15 +114,19 @@ Remediation release — index/graph correctness, data protection, and web flow f
 - Regenerated stale route tree referencing the deleted `/jobs` route; removed the CI lint no-op (`turbo run lint` with zero lint tasks) (FIX-08)
 
 ### Removed
+
 - Jobs page and related API endpoints
 
 ## [0.5.1] - 2026-07-31
 
 ### Fixed
+
 - Blazing-fast indexing for large codebases (batch transactions, optimized write mode)
 - Error handling and validation for import path extraction
 - CLI npm packaging
 
+[0.10.0]: https://github.com/asta-nguyen/openez-graph/compare/fbcad4f...HEAD
+[0.9.1]: https://github.com/asta-nguyen/openez-graph/compare/fbcad4f...HEAD
 [0.9.0]: https://github.com/asta-nguyen/openez-graph/compare/fbcad4f...HEAD
 [0.8.0]: https://github.com/asta-nguyen/openez-graph/compare/a7ce4df...849060b
 [0.7.0]: https://github.com/asta-nguyen/openez-graph/compare/9b7cc78...a7ce4df