v0.10.4

Background Auto-Updater

Memorix now silently self-updates when a new version is available on npm. No peer MCP server has this feature.

How it works

1. On \memorix serve\ startup, async checks npm registry (non-blocking, doesn't slow MCP connection)
2. If newer version found, runs
   pm install -g memorix@latest\ in background
3. Current session uninterrupted update takes effect on next restart
4. Rate-limited: checks at most once per 24 hours (cached in ~/.memorix/update-check.json)

Why

Users previously had to manually run
pm update -g memorix\ after every release. Now it just works like Chrome or VS Code auto-updates.

Test Results

• 606/606 tests passed (45 test files, 13 new update-checker tests)

v0.10.3

Changes

Bug Fixes

• Fix agent rules fragmentation risk: RULE 1 now calls \memorix_session_start\ first (direct read of previous session summary + key memories), then \memorix_search\ for additional context. Previously only used search, which could miss important context.
• Fixed Kiro hook prompt and OpenCode compaction prompt similarly.

Documentation

• Removed competitor comparison table from README / README.zh-CN
• Added Star History chart (star-history.com, supports dark/light mode)
• Updated outdated numbers across all docs (tests 534593, tools 1625, adapters 69)
• Added \memorix configure\ to CLI section
• ARCHITECTURE.md: added LLM/API embedding layers, fixed adapter counts
• API_REFERENCE.md: updated tool overview to 25 tools in 6 categories

Full Test Suite

• 593/593 tests passed (44 test files, 0 failures)

v0.9.29 Fix Gemini/Antigravity Hooks & Improve Rules

v0.9.29 — Fix Gemini/Antigravity Hooks & Improve Rules Compliance

Bug Fixes

Gemini CLI hooks not firing — Two root causes fixed:

1. Missing enableHooks: true + hooks.enabled: true flags (Gemini CLI hooks are experimental and require explicit opt-in)
2. Merge logic only merged hooks key but not tools key in settings.json

Antigravity rules installed to wrong path — Changed from .gemini/rules/memorix.md to GEMINI.md (Gemini's default context file, similar to Codex's AGENTS.md)

Improvements

Rules template rewritten for better AI compliance — Stronger imperative language (MUST/CRITICAL), explicit trigger checklist, and clear examples of what to record vs skip. This is especially important for Antigravity IDE which has no native hooks support (confirmed limitation).

Important Note

Antigravity IDE does NOT support lifecycle hooks (confirmed by InfoWorld review, Feb 2026). For Antigravity, only rules-based memory recording via GEMINI.md is available. Gemini CLI (terminal) does support hooks with the new enableHooks flag.

Full Changelog

v0.9.28...v0.9.29

v0.9.0 Consolidation + Temporal Queries + Explainable Recall + Export/Import

🚀 Memorix v0.9.0 — Consolidation + Temporal Queries + Explainable Recall + Export/Import

✨ New Features

🧹 Memory Consolidation (memorix_consolidate)

Find and merge similar observations to reduce memory bloat. Uses Jaccard text similarity to cluster observations by entity+type, then merges them preserving all facts, files, and concepts.

• action: "preview" — Dry run showing merge candidates
• action: "execute" — Actually merge clusters
• Configurable similarity threshold

⏰ Temporal Queries

memorix_search now supports since and until parameters:

"What auth decisions did we make last week?"
→ memorix_search(query: "auth", type: "decision", since: "2026-02-17")

🔍 Explainable Recall

Search results now show why each result was found with a Matched column:

ID	Time	T	Title	Tokens	Matched
#42	2:14 PM	🟤	JWT auth decision	~155	title, concept

📦 Export/Import

• memorix_export — JSON (full fidelity, importable) or Markdown (human-readable for PRs/docs)
• memorix_import — Restore from JSON, re-assigns IDs, deduplicates by topicKey

📋 Dashboard Sessions Panel

New "Sessions" tab in the web dashboard with timeline view, active/completed counts, agent info, and session summaries. Bilingual (EN/中文).

🔗 Auto sessionId

memorix_store now automatically associates the current active session's ID with stored observations.

📊 Stats

• MCP Tools: 20 → 24
• Tests: 484/484 passing (+16 new)
• Package size: 838.2 kB

v0.8.0 Session Lifecycle + Topic Key Upsert

🚀 Memorix v0.8.0 — Session Lifecycle + Topic Key Upsert

This is a milestone release that brings cross-session context continuity and memory deduplication to Memorix.

✨ New MCP Tools (16 → 20)

Tool	Description
memorix_session_start	Start a coding session, auto-inject context from previous sessions (summaries + key observations)
memorix_session_end	End a session with structured summary (Goal/Discoveries/Accomplished/Files format)
memorix_session_context	Manually retrieve session history and context
memorix_suggest_topic_key	Suggest stable topic keys for memory upserts

🔄 Topic Key Upsert

memorix_store now accepts an optional topicKey parameter. When an observation with the same topicKey + projectId already exists, it is updated in-place instead of creating a duplicate. revisionCount increments on each upsert.

Before: Testing 10 versions = 10 records
After: Testing 10 versions = 1 record + revision history

📋 Session Lifecycle

Agents can now track coding sessions. When a new session starts, previous session summaries and key observations are automatically injected — no manual recall needed.

📊 Stats

• Tests: 468/468 passing (+30 new)
• Package size: 809.1 kB
• New files: src/memory/session.ts, 2 test files
• Modified: types.ts, observations.ts, persistence.ts, server.ts

🙏 Inspired by

• Engram — Session lifecycle design, topic_key upsert pattern

v0.7.0 Memory-Driven Skills Engine + Transformers.js Embedding

What's New in v0.7.0

🧠 Memory-Driven Skills Engine

New memorix_skills MCP tool that turns your project memory into reusable agent skills:

• list — Discover all SKILL.md files across 7 agent directories
• generate — Auto-generate project-specific skills from observation patterns (gotchas, decisions, how-it-works)
• inject — Return full skill content directly into agent context
• Intelligent scoring: requires skill-worthy observation types, not just volume
• Write to any target agent with write: true, target: "<agent>"

🔌 Transformers.js Embedding Provider

• Pure JavaScript fallback via @huggingface/transformers — no native deps required
• Provider chain: fastembed → transformers.js → fulltext-only graceful degradation
• Quantized model (q8) for small footprint

📊 Dashboard Enhancements

• Canvas donut chart for observation type distribution
• Embedding provider status card (enabled/provider/dimensions)
• Search result highlighting with <mark> tags
• Graph legend with light/dark theme compatibility

✅ Quality

• 422 tests passing across 34 test files (up from 405)
• Zero regressions

---

Quick Start

{
  "mcpServers": {
    "memorix": {
      "command": "npx",
      "args": ["-y", "memorix@latest", "serve"]
    }
  }
}

Add to your agent's MCP config and restart. Supports Windsurf · Cursor · Claude Code · Codex · VS Code Copilot · Antigravity · Kiro.

Full config paths: Agent Configuration