Skip to content

CHANGELOG.md

Latest commit

 

History

History
127 lines (91 loc) · 8.92 KB

CHANGELOG.md

File metadata and controls

127 lines (91 loc) · 8.92 KB
Raw

v0.10.7 — 2026-06-09

Features

  • Native Hermes Agent memory provider — Cortex can now be used as a first-class Hermes memory.provider (a self-hosted alternative to Mem0) with automatic pre-turn recall and post-turn ingestion, instead of only the manual MCP path. A zero-dependency reference implementation targeting Hermes' agent/memory_provider.py ABC ships in integrations/hermes/cortex, exposing cortex_search / cortex_remember / cortex_forget tools and hermes memory setup config. (requested by @Yat-mo, #23)

Bug Fixes

  • sqlite-vec dimension change now actually rebuilds the vec0 table — Switching embedding models to a different dimension (e.g. default 1536 → bge-m3 1024) silently left the old FLOAT[1536] virtual table in place, so every subsequent upsert/search failed with a dimension mismatch and semantic recall fell back to plain text. The dimension check read a non-existent dimensions key from vec0's _info shadow table and never matched, so the rebuild never ran — this is why a manual database wipe was the only workaround. The check now reads the declared dimension straight from the table's CREATE statement in sqlite_master. Completes the partial v0.10.4 fix. (reported by @Miantiao1231, #21)
  • OpenClaw plugin install no longer blocked by the security scanner — @cortexmem/openclaw tripped OpenClaw's "environment variable access combined with network send — possible credential harvesting" install scanner because the bundled plugin read process.env.* (including CORTEX_AUTH_TOKEN) in the same file as its fetch calls. The plugin now sources all configuration from OpenClaw's sanctioned pluginConfig channel and reads no environment variables, so the heuristic no longer fires. New authToken and pairingCode config keys replace the former CORTEX_AUTH_TOKEN / CORTEX_PAIRING_CODE env vars. (reported by @southxs, #22)

Packages

  • @cortexmem/openclaw@0.6.7 published (security-scanner fix; config-only, no env var reads)

v0.10.4 — 2026-03-28

Bug Fixes

  • OpenClaw plugin ID mismatch fixed — @cortexmem/openclaw now correctly declares plugin id "openclaw", resolving the plugin id mismatch (manifest uses "cortex-bridge", entry hints "openclaw") error on install (reported by @rollingshmily, #17)
  • Reindex now rebuilds vec0 table on dimension change — Clicking "重建向量索引" after switching embedding models (e.g. voyage-3.5 → text-embedding-3-large) no longer fails with Expected 1536 dimensions but received 1024. The reindex endpoint now reinitializes the vector backend before re-embedding (reported by @rollingshmily, #17)
  • Embedding hot-reload reinitializes vector backend — Saving a new embedding config via Dashboard or PATCH /api/v1/config now immediately rebuilds the vec0 table if dimensions changed, no restart needed
  • Neo4j relation deduplication — Repeated extraction of the same (subject, predicate, object, agent_id) tuple now updates a single edge instead of creating duplicate edges (contributed by @xieyuanqing, #16)
  • Correction cascade for state memories — correction memories now correctly supersede related todo, decision, goal, project_state memories; search coverage expanded to include todo, constraint, policy categories (contributed by @xieyuanqing, #16)

Packages

  • @cortexmem/openclaw@0.6.5 published (plugin id fix)

v0.10.3 — 2026-03-25

Features

  • pairing_code: Agent namespace isolation — Solves multi-instance memory mixing. When multiple OpenClaw instances, MCP clients, or API callers share the same agent_id (e.g. "main"), memories would previously bleed into each other. With pairing_code, each client gets its own isolated namespace.

    How it works: Optional pairing_code field in all recall/ingest requests. Memories stored/retrieved by composite key (agent_id + pairing_code). No pairing_code = backward-compatible unchanged behavior.

    Configuration:

    • OpenClaw plugin: set CORTEX_PAIRING_CODE=<unique-code> env var
    • MCP / API: pass "pairing_code": "<code>" in request body

    Database: migration 014 adds pairing_code column + composite index.

Changelog

v0.10.2 — 2026-03-21

Features

  • Recall timeout: Configurable total recall timeout (recallTimeoutMs, default 5000ms) — if expansion + search + rerank exceeds budget, returns partial results instead of blocking
  • Dashboard recall timeout setting: New field in Gate section with variant-count recommendations (1→3s, 3→5s, 6→8s, 10+→12s)

Performance

  • Reranker instant failover: Auth errors (401/403) skip immediately instead of waiting for timeout
  • Parallel variant embedding: Query expansion variants embedded in a single batch API call instead of sequential
  • Budget-aware timeouts: Reranker and expansion respect remaining recall budget

Fixes

  • Timer leak: Recall timeout timer properly cleaned up in try/finally
  • Reranker/expansion honor recallTimeoutMs: No longer use hardcoded timeouts that can overshoot the total budget

v0.10.1 — 2026-03-21

Fixes

  • Reindex auto-rebuilds vec table on dimension change: When embedding dimensions change (e.g. switching from text-embedding-3-small 1536d to text-embedding-3-large 3072d), the vec0 virtual table is now automatically dropped and recreated with the correct dimensions on startup
  • Reindex reports upsert errors: Vector upsert failures during reindex are now caught and counted in the errors field instead of being silently swallowed

v0.10.0 — 2026-03-21

Features

  • Memory Self-Improvement System: Feedback-driven importance auto-adjustment
    • Feedback API: POST /api/v1/memories/:id/feedback (+1/0/-1 scoring)
    • Implicit feedback via recall usage tracking (POST /api/v1/recall/:recallId/usage)
    • Lifecycle Phase 6c: automatic importance adjustment with stability safeguards
    • Configurable: window size, min feedbacks, max delta, implicit/explicit weights
    • Full audit trail in importance_adjustments table for rollback
  • Dashboard UX/UI Redesign: Modern Linear/Vercel-inspired aesthetic
    • New CSS design system with layered backgrounds, 4-level text hierarchy
    • Redesigned sidebar with section dividers and refined navigation
    • Stats page with improved charts and trend indicators
    • Memory browser with category-colored cards and inline feedback buttons
    • Settings page with self-improvement configuration panel
    • All pages polished: Agents, ExtractionLogs, SystemLogs, LifecycleMonitor, SearchDebug

Security

  • Rate limiting on feedback API endpoints
  • Input validation: maxLength on comments, maxItems on memory_ids arrays
  • Recall session map size cap (10K) with TTL cleanup

Improvements

  • Self-improvement settings section in Dashboard
  • i18n: 24 new strings in both English and Chinese
  • Focus-visible CSS styles for keyboard accessibility

v0.9.5 — 2026-03-21

Fixes

  • Timing leak in safeCompare: Pad both strings to equal length before constant-time comparison, preventing length-based timing attacks.
  • Deep merge for env overrides in loadConfig: Use deepMerge instead of shallow spread so env var overrides (e.g. CORTEX_DB_PATH) don't clobber nested file config.
  • PLUGIN_VERSION: Corrected cortex-bridge version from 0.5.1 to 0.6.1.
  • README accuracy: Fixed Node.js requirement (≥ 20), removed speculative model names, added cortex_relations to MCP tools table, fixed pnpm start → pnpm dev.
  • CHANGELOG: Added missing v0.9.3 entry and dates.
  • LICENSE: Added MIT license file.

v0.9.4 — 2026-03-15

Features

  • Multi-agent tool isolation: Bridge tools (cortex_recall, cortex_remember, cortex_ingest) now accept an optional agent_id parameter, enabling per-request agent switching in multi-agent orchestrators.
  • Dynamic agent ID in MCP client: CORTEX_AGENT_ID env var is re-read per request, supporting dynamic agent switching without restart.
  • Per-session agent resolution in bridge: ctx.agentId is resolved per session and cached, preventing race conditions in concurrent multi-agent scenarios.

Improvements

  • Selective engine rebuild on config change: reloadProviders now only reconstructs engines whose configuration actually changed, instead of rebuilding all engines on any config update.
  • Settings key preservation: Switching providers in the dashboard Settings no longer loses previously configured API keys on the first switch; existing keys are tracked from initial config load.

Fixes

  • Fixed version display endpoint to return correct version string.

v0.9.3 — 2026-03-10

  • Bump cortex-bridge to v0.6.1 for npm publish.

v0.9.2 — 2026-03-08

  • Use config.storage.dbPath instead of non-existent config.dataDir in backupDb().

v0.9.1 — 2026-03-05

  • Bump @cortexmem/mcp to 0.1.2 with auth-token support.

v0.9.0 — 2026-03-01

  • Broaden fast channel coverage for diverse users.
  • Comprehensive fast channel and prompt audit.