superlocalmemory 2.6.0 → 2.7.0

package/CHANGELOG.md

@@ -16,1884 +16,262 @@ SuperLocalMemory V2 - Intelligent local memory system for AI coding assistants.
 
 ---
 
-## [2.5.1] - 2026-02-13
+## [2.7.0] - 2026-02-16
 
-**Release Type:** Framework Integration Release — "Plugged Into the Ecosystem"
-**Backward Compatible:** Yes (additive packages only, no core changes)
+**Release Type:** Major Feature Release — "Your AI Learns You"
 
-### The Big Picture
-SuperLocalMemory is now a first-class memory backend for LangChain and LlamaIndex — the two largest AI/LLM frameworks. Two pip-installable packages, zero cloud dependencies.
+SuperLocalMemory now learns your patterns, adapts to your workflow, and personalizes recall. All processing happens 100% locally — your behavioral data never leaves your machine. GDPR Article 17 compliant by design.
 
 ### Added
-- **LangChain Integration** (`langchain-superlocalmemory`): Persistent chat message history backed by local SQLite. Works with `RunnableWithMessageHistory` and LCEL chains. `pip install langchain-superlocalmemory`
-- **LlamaIndex Integration** (`llama-index-storage-chat-store-superlocalmemory`): Full `BaseChatStore` implementation. Works with `ChatMemoryBuffer` and `SimpleChatEngine`. `pip install llama-index-storage-chat-store-superlocalmemory`
-- **Example scripts** for both frameworks in `examples/` directory — runnable without API keys
-- **Session isolation**: Framework memories are tagged separately and never appear in normal `slm recall`
-
----
-
-## [2.5.0] - 2026-02-12
-
-**Release Type:** Major Feature Release — "Your AI Memory Has a Heartbeat"
-**Backward Compatible:** Yes (additive tables, columns, and modules only)
-
-### The Big Picture
-SuperLocalMemory transforms from passive storage to **active coordination layer**. Every memory operation now triggers real-time events. Agents are tracked. Trust is scored. The dashboard is live.
-
-### Added
-- **DbConnectionManager** (`src/db_connection_manager.py`): SQLite WAL mode, 5s busy timeout, thread-local read pool, serialized write queue. Fixes "database is locked" errors when multiple agents write simultaneously
-- **Event Bus** (`src/event_bus.py`): Real-time event broadcasting — every memory write/update/delete fires events to SSE, WebSocket, and Webhook subscribers. Tiered retention (hot 48h, warm 14d, cold 30d)
-- **Subscription Manager** (`src/subscription_manager.py`): Durable + ephemeral event subscriptions with filters (event type, importance, protocol)
-- **Webhook Dispatcher** (`src/webhook_dispatcher.py`): Background HTTP POST delivery with exponential backoff retry (3 attempts)
-- **Agent Registry** (`src/agent_registry.py`): Tracks which AI agents connect — protocol, write/recall counters, last seen, metadata
-- **Provenance Tracker** (`src/provenance_tracker.py`): Tracks who created each memory — created_by, source_protocol, trust_score, provenance_chain
-- **Trust Scorer** (`src/trust_scorer.py`): Bayesian trust signal collection — positive (cross-agent recall, high importance), negative (quick delete, burst write). Silent collection in v2.5, enforcement in v2.6
-- **SSE endpoint** (`/events/stream`): Real-time Server-Sent Events with cross-process DB polling, Last-Event-ID replay, event type filtering
-- **Dashboard: Live Events tab**: Real-time event stream with color-coded types, filter dropdown, stats counters
-- **Dashboard: Agents tab**: Connected agents table with protocol badges, trust scores, write/recall counters, trust overview
-- **4 new database tables**: `memory_events`, `subscriptions`, `agent_registry`, `trust_signals`
-- **4 new columns on `memories`**: `created_by`, `source_protocol`, `trust_score`, `provenance_chain`
-- **28 API endpoints** across 8 modular route files
-- **Architecture document**: `docs/ARCHITECTURE-V2.5.md`
+- **Adaptive Learning System** — Three-layer learning architecture that detects tech preferences, project context, and workflow patterns across all your projects
+- **Personalized Recall Ranking** — Search results re-ranked using learned patterns. Three-phase adaptive system: baseline → rule-based → ML (LightGBM LambdaRank)
+- **Synthetic Bootstrap** — ML model works from day 1 by bootstrapping from existing memory patterns. No cold-start degradation.
+- **Multi-Channel Feedback** — Tell the system which memories were useful via MCP (`memory_used`), CLI (`slm useful`), or dashboard clicks
+- **Source Quality Scoring** — Learns which tools produce the most useful memories using Beta-Binomial Bayesian scoring
+- **Workflow Pattern Detection** — Detects your coding workflow sequences (e.g., docs → architecture → code → test) using time-weighted sliding-window mining
+- **Local Engagement Metrics** — Track memory system health locally with zero telemetry (`slm engagement`)
+- **Separate Learning Database** — Behavioral data in `learning.db`, isolated from `memory.db`. One-command erasure: `slm learning reset`
+- **3 New MCP Tools** — `memory_used` (feedback signal), `get_learned_patterns` (transparency), `correct_pattern` (user control)
+- **2 New MCP Resources** — `memory://learning/status`, `memory://engagement`
+- **New CLI Commands** — `slm useful`, `slm learning status/retrain/reset`, `slm engagement`, `slm patterns correct`
+- **New Skill** — `slm-show-patterns` for viewing learned preferences in Claude Code and compatible tools
+- **Auto Python Installation** — `install.sh` now auto-installs Python 3 on macOS (Homebrew/Xcode) and Linux (apt/dnf) for new users
+- **319 Tests** — 229 unit tests + 13 E2E + 14 regression + 19 fresh-install + 42 edge-case tests
+
+### Research Foundations
+- Two-stage BM25 → re-ranker pipeline (eKNOW 2025)
+- LightGBM LambdaRank pairwise ranking (Burges 2010, MO-LightGBM SIGIR 2025)
+- Three-phase cold-start mitigation (LREC 2024)
+- Time-weighted sequence mining (TSW-PrefixSpan, IEEE 2020)
+- Bayesian temporal confidence (MACLA, arXiv:2512.18950)
+- Privacy-preserving zero-communication feedback design
 
 ### Changed
-- **`memory_store_v2.py`**: Replaced 15 direct `sqlite3.connect()` calls with `DbConnectionManager` (read pool + write queue). Added EventBus, Provenance, Trust integration. Fully backward compatible — falls back to direct SQLite if new modules unavailable
-- **`mcp_server.py`**: Replaced 9 per-call `MemoryStoreV2(DB_PATH)` instantiations with shared singleton. Added agent registration and provenance tracking for MCP calls
-- **`ui_server.py`**: Refactored from 2008-line monolith to 194-line app shell + 9 route modules in `routes/` directory using FastAPI APIRouter
-- **`ui/app.js`**: Split from 1588-line monolith to 13 modular files in `ui/js/` directory (largest: 230 lines)
+- **MCP Tools** — Now 12 tools (was 9), 6 resources (was 4), 2 prompts
+- **Skills** — Now 7 universal skills (was 6)
+- **install.sh** — Auto-installs Python if missing, installs learning deps automatically
+- **DMG Installer** — Updated to v2.7.0 with learning modules
 
-### Fixed
-- **"Database is locked" errors**: WAL mode + write serialization eliminates concurrent access failures
-- **Fresh database crashes**: `/api/stats`, `/api/graph`, `/api/clusters` now return graceful empty responses when `graph_nodes`/`graph_edges`/`sessions` tables don't exist yet
-- **Per-call overhead**: MCP tool handlers no longer rebuild TF-IDF vectorizer on every call
+### Dependencies (Optional)
+- `lightgbm>=4.0.0` — ML ranking (auto-installed, graceful fallback if unavailable)
+- `scipy>=1.9.0` — Statistical functions (auto-installed, graceful fallback if unavailable)
 
-### Testing
-- 63 pytest tests (unit, concurrency, regression, security, edge cases, Docker/new-user scenarios)
-- 27 end-to-end tests (all v2.5 components + 50 concurrent writes + reads during writes)
-- 15 fresh-database edge case tests
-- 17 backward compatibility tests against live v2.4.2 database
-- All CLI commands and skill scripts verified
-- Profile isolation verified across 2 profiles
-- Playwright dashboard testing (all 8 tabs, SSE stream, profile switching)
-
----
-
-## [2.4.2] - 2026-02-11
-
-**Release Type:** Bug Fix Release
-**Backward Compatible:** Yes
-
-### Fixed
-- **Profile isolation bug in UI dashboard**: Graph nodes and connections were displaying global counts instead of profile-filtered counts. New/empty profiles incorrectly showed data from other profiles. Fixed by adding `JOIN memories` and `WHERE m.profile = ?` filter to graph stats queries in `ui_server.py` (`/api/stats` endpoint, lines 986-990).
+### Performance
+- Re-ranking adds <15ms latency to recall queries
+- Learning DB typically <1MB for 1,000 memories
+- Bootstrap model trains in <30 seconds for 1,000 memories
+- All BM1-BM6 benchmarks: no regression >10%
 
 ---
 
-## [2.4.1] - 2026-02-11
-
-**Release Type:** Hierarchical Clustering & Documentation Release
-**Backward Compatible:** Yes (additive schema changes only)
+## [2.6.5] - 2026-02-16
 
 ### Added
-- **Hierarchical Leiden clustering** (`graph_engine.py`): Recursive community detection — large clusters (≥10 members) are automatically sub-divided up to 3 levels deep. E.g., "Python" → "FastAPI" → "Authentication patterns". New `parent_cluster_id` and `depth` columns in `graph_clusters` table
-- **Community summaries** (`graph_engine.py`): TF-IDF structured reports for every cluster — key topics, projects, categories, hierarchy context. Stored in `graph_clusters.summary` column, surfaced in `/api/clusters` endpoint and web dashboard
-- **CLI commands**: `python3 graph_engine.py hierarchical` and `python3 graph_engine.py summaries` for manual runs
-- **Schema migration**: Safe `ALTER TABLE` additions for `summary`, `parent_cluster_id`, `depth` columns — backward compatible with existing databases
-
-### Changed
-- `build_graph()` now automatically runs hierarchical sub-clustering and summary generation after flat Leiden
-- `/api/clusters` endpoint returns `summary`, `parent_cluster_id`, `depth` fields
-- `get_stats()` includes `max_depth` and per-cluster summary/hierarchy data
-- `setup_validator.py` schema updated to include new columns
-
-### Documentation
-- **README.md**: v2.4.0→v2.4.1, added Hierarchical Leiden, Community Summaries, MACLA, Auto-Backup sections
-- **Wiki**: Updated Roadmap, Pattern-Learning-Explained, Knowledge-Graph-Guide, Configuration, Visualization-Dashboard, Footer
-- **Website**: Updated features.astro, comparison.astro, index.astro for v2.4.1 features
-- **`.npmignore`**: Recursive `__pycache__` exclusion patterns
+- **Interactive Knowledge Graph** - Fully interactive visualization with zoom, pan, and click-to-explore capabilities
+- **Mobile & Accessibility Support** - Touch gestures, keyboard navigation, and screen reader compatibility
 
 ---
 
-## [2.4.0] - 2026-02-11
-
-**Release Type:** Profile System & Intelligence Release
-**Backward Compatible:** Yes (additive schema changes only)
-
-### Added
-- **Column-based memory profiles**: Single `memory.db` with `profile` column — all memories, clusters, patterns, and graph data are profile-scoped. Switch profiles from any IDE/CLI and it takes effect everywhere instantly via shared `profiles.json`
-- **Auto-backup system** (`src/auto_backup.py`): SQLite backup API with configurable interval (daily/weekly), retention policy, and one-click backup from UI
-- **MACLA confidence scorer**: Research-grounded Beta-Binomial Bayesian posterior (arXiv:2512.18950) replaces ad-hoc log2 formula. Pattern-specific priors: preference(1,4), style(1,5), terminology(2,3). Log-scaled competition prevents over-dilution from sparse signals
-- **UI: Profile Management**: Create, switch, and delete profiles from the web dashboard. "+" button in navbar for quick creation, full management table in Settings tab
-- **UI: Settings tab**: Auto-backup status, configuration (interval, max backups, enable toggle), backup history, profile management — all in one place
-- **UI: Column sorting**: Click any column header in the Memories table to sort asc/desc
-- **UI: Enhanced Patterns view**: DOM-based rendering with confidence bars, color coding, type icons
-- **API: Profile isolation on all endpoints**: `/api/graph`, `/api/clusters`, `/api/patterns`, `/api/timeline` now filter by active profile (previously showed all profiles)
-- **API: `get_active_profile()` helper**: Shared function in `ui_server.py` replaces 4 duplicate inline profile-reading blocks
-- **API: Profile CRUD endpoints**: `POST /api/profiles/create`, `DELETE /api/profiles/{name}` with validation and safety (can't delete default or active profile)
-
-### Fixed
-- **Profile switching ValueError**: Rewrote from directory-based to column-based profiles — no more file copy errors on switch
-- **Pattern learner schema validation**: Safe column addition with try/except for `profile` column on `identity_patterns` table
-- **Graph engine schema validation**: Safe column check before profile-filtered queries
-- **Research references**: PageIndex correctly attributed to VectifyAI (not Meta AI), removed fabricated xMemory/Stanford citation, replaced with MemoryBank (AAAI 2024) across wiki and website
-- **Graph tooltip**: Shows project name or Memory ID instead of "Uncategorized" when category is null
-
-### Changed
-- All 4 core layers (storage, tree, graph, patterns) are now profile-aware
-- `memory_store_v2.py`: Every query filters by `WHERE profile = ?` from `_get_active_profile()`
-- `graph_engine.py`: `build_graph()` and `get_stats()` scoped to active profile
-- `pattern_learner.py`: Pattern learning and retrieval scoped to active profile
-- `ui_server.py`: Refactored profile code into shared helper, eliminated 4 duplicate blocks
-
-### Technical Details
-- Schema: `ALTER TABLE memories ADD COLUMN profile TEXT DEFAULT 'default'`
-- Schema: `ALTER TABLE identity_patterns ADD COLUMN profile TEXT DEFAULT 'default'`
-- MACLA formula: `posterior = (alpha + evidence) / (alpha + beta + evidence + log2(total_memories))`
-- Confidence range: 0.0 to 0.95 (capped), with recency and distribution bonuses
-- Backup: Uses SQLite `backup()` API for safe concurrent backup
-- 17 API endpoint tests, 5 core module tests — all passing
+## [2.6.0] - 2026-02-15
 
----
-
-## [2.3.7] - 2026-02-09
+**Release Type:** Security Hardening & Scalability — "Battle-Tested"
 
 ### Added
-- **--full flag**: Show complete memory content without truncation in search/list/recent/cluster commands
-- **Smart truncation**: Memories <5000 chars shown in full, ≥5000 chars truncated to 2000 chars (previously always truncated at 200 chars)
-- **Help text**: Added --full flag documentation to CLI help output
-
-### Fixed
-- **CLI bug**: Fixed `get` command error - `get_memory()` → `get_by_id()` method call
-- **Content display**: Recall now shows full content for short/medium memories instead of always truncating at 200 chars
-- **User experience**: Agents and users can now see complete memory content by default for most memories
+- **Rate Limiting** - 100 writes/min and 300 reads/min per IP (stdlib-only, no dependencies)
+- **API Key Authentication** - Optional API key auth via `~/.claude-memory/api_key` file
+- **CI Workflow** - GitHub Actions test pipeline across Python 3.8, 3.10, and 3.12
+- **Trust Enforcement** - Agents below 0.3 trust score blocked from write/delete
+- **HNSW Vector Index** - O(log n) search with graceful TF-IDF fallback
+- **Hybrid Search** - BM25 + Graph + TF-IDF fusion via `SLM_HYBRID_SEARCH=true`
+- **SSRF Protection** - Webhook URLs validated against private IP ranges
 
 ### Changed
-- **Truncation logic**: 200 char limit → 2000 char preview for memories ≥5000 chars
-- **Node.js wrappers**: memory-recall-skill.js and memory-list-skill.js updated to pass --full flag through
-
-### Technical Details
-- Added `format_content()` helper function in memory_store_v2.py (line 918)
-- Updated search/list/recent/cluster commands to use smart truncation
-- Backward compatible: same output structure, MCP/API calls unaffected
-- All 74+ existing memories tested: short memories show full, long memories truncate intelligently
-
----
-
-## [2.3.5] - 2026-02-09
-
-### Added
-- **ChatGPT Connector Support**: `search(query)` and `fetch(id)` MCP tools per OpenAI spec
-- **Streamable HTTP transport**: `slm serve --transport streamable-http` for ChatGPT 2026+
-- **UI: Memory detail modal**: Click any memory row to see full content, tags, metadata
-- **UI: Dark mode toggle**: Sun/moon icon in navbar, saved to localStorage, respects system preference
-- **UI: Export buttons**: Export All (JSON/JSONL), Export Search Results, Export individual memory as Markdown
-- **UI: Search score bars**: Color-coded relevance bars (red/yellow/green) in search results
-- **UI: Animated stat counters**: Numbers animate up on page load with ease-out cubic
-- **UI: Loading spinners and empty states**: Professional feedback across all tabs
-- npm keywords: chatgpt, chatgpt-connector, openai, deep-research
+- **Graph cap raised to 10,000 memories** (was 5,000) with intelligent random sampling
+- **Profile isolation hardened** - All queries enforce `WHERE profile = ?` filtering
+- **Bounded write queue** - Max 1,000 pending operations (was unbounded)
+- **TF-IDF optimization** - Skip rebuild when memory count changes <5%
+- **Error sanitization** - Internal paths and table names stripped from 15 error sites
+- **Connection pool capped** - Max 50 read connections
 
 ### Fixed
-- **XSS vulnerability**: Replaced inline onclick with JSON injection with safe event delegation
-- **UI: Content preview**: Increased from 80 to 100 characters
+- Replaced bare `except:` clauses with specific exception types in hybrid_search.py and cache_manager.py
 
-### Changed
-- npm package now includes `ui/`, `ui_server.py`, `api_server.py`
+### Performance
+See [wiki Performance Benchmarks](https://superlocalmemory.com/wiki/Performance-Benchmarks) for measured data.
 
 ---
 
-## [2.3.0] - 2026-02-08
-
-**Release Type:** Universal Integration Release
-**Release Date:** February 8, 2026
-**Version Code:** 2.3.0-universal
-**Git Tag:** v2.3.0
-**Backward Compatible:** ✅ Yes (100%)
-
-### 🌐 Universal Integration — MAJOR UPDATE
-
-**SuperLocalMemory now works across 16+ IDEs and CLI tools!**
-
-This release fixes the Claude-first distribution gap by adding proper configs, detection, and integration for the tools where most non-Claude developers live.
-
-**Root Cause:** The architecture was always universal (SQLite + MCP + Skills), but the distribution (configs, installer, docs, npm) was Claude-first with bolted-on support for others. This release fixes that.
-
-### ✨ Added — New Integrations
-
-**New Config Templates:**
-- ✅ `configs/codex-mcp.toml` — OpenAI Codex CLI (TOML format, not JSON)
-- ✅ `configs/vscode-copilot-mcp.json` — VS Code / GitHub Copilot (`"servers"` format)
-- ✅ `configs/gemini-cli-mcp.json` — Google Gemini CLI
-- ✅ `configs/jetbrains-mcp.json` — JetBrains IDEs (IntelliJ, PyCharm, WebStorm)
-
-**New install.sh Detections:**
-- ✅ OpenAI Codex CLI — Auto-configures via `codex mcp add` or TOML fallback
-- ✅ VS Code / GitHub Copilot — Creates `~/.vscode/mcp.json`
-- ✅ Gemini CLI — Merges into `~/.gemini/settings.json`
-- ✅ JetBrains IDEs — Prints manual setup instructions (GUI-based)
-
-**New CLI Command:**
-- ✅ `slm serve [PORT]` — Start MCP HTTP server for ChatGPT/remote access
-  - Default port: 8001
-  - Documents ngrok/cloudflared tunnel workflow
-  - Enables ChatGPT integration (previously broken)
-
-**Universal Symlink:**
-- ✅ `~/.superlocalmemory` → `~/.claude-memory` — Non-Claude users see universal branding
-  - Zero breaking changes (real directory unchanged)
-  - Additive only (removing symlink doesn't break anything)
-
-**MCP Tool Annotations:**
-- ✅ All 6 tools annotated with `readOnlyHint`, `destructiveHint`, `openWorldHint`
-  - Required by ChatGPT and VS Code Copilot for tool classification
-  - Uses `ToolAnnotations` from MCP SDK
-
-**Skills Installer Expansion:**
-- ✅ Added Cursor to `install-skills.sh`
-- ✅ Added VS Code/Copilot to `install-skills.sh`
-- ✅ Added `--auto` flag for non-interactive mode
-- ✅ `install.sh` now calls `install-skills.sh --auto` automatically
-
-### 🔧 Fixed
-
-**ChatGPT Integration (was broken):**
-- Old config used stdio — ChatGPT only supports HTTP transport
-- New: `slm serve` + tunnel workflow documented
-- Config file replaced with setup instructions
-
-### 📝 Documentation Updates
-
-**docs/MCP-MANUAL-SETUP.md:**
-- Added: OpenAI Codex CLI section
-- Added: VS Code / GitHub Copilot section
-- Added: Gemini CLI section
-- Added: JetBrains IDEs section
-- Added: HTTP Transport section
-- Fixed: ChatGPT section (HTTP workflow replaces broken stdio instructions)
-
-**README.md:**
-- Expanded IDE table from 8 to 17 rows
-- Updated "11+ IDEs" → "16+ IDEs" everywhere
-
-### 🔢 Version Bumps
-
-| File | Old | New |
-|------|-----|-----|
-| `package.json` | 2.1.0 | 2.3.0 |
-| `mcp_server.py` | 2.1.0-universal | 2.3.0-universal |
-| `bin/slm` | 2.1.0-universal | 2.3.0-universal |
-| `CLAUDE.md` | 2.1.0-universal | 2.3.0-universal |
-| `postinstall.js` | "11+ AI tools" | "16+ AI tools" |
-
-### 🔒 Backward Compatibility
-
-**100% backward compatible — nothing breaks:**
-- ✅ Existing `~/.claude-memory/` data untouched
-- ✅ Existing MCP configs (Claude, Cursor, etc.) untouched
-- ✅ Existing skills untouched
-- ✅ Existing `slm` commands untouched (`serve` is NEW)
-- ✅ npm reinstall safe (backs up before overwriting)
-- ✅ `git pull && ./install.sh` safe for existing users
+## [2.5.1] - 2026-02-13
 
-### 🎊 Credits
+**Release Type:** Framework Integration — "Plugged Into the Ecosystem"
 
-**Philosophy:** The architecture was already universal. This release makes the distribution universal too.
+SuperLocalMemory is now a first-class memory backend for LangChain and LlamaIndex — the two largest AI/LLM frameworks.
 
-**Author:** Varun Pratap Bhardwaj (Solution Architect)
+### Added
+- **LangChain Integration** - `langchain-superlocalmemory` package for persistent chat history
+- **LlamaIndex Integration** - `llama-index-storage-chat-store-superlocalmemory` for ChatMemoryBuffer
+- **Session isolation** - Framework memories tagged separately, never appear in normal `slm recall`
 
 ---
 
-## [2.2.0] - 2026-02-07
-
-**Release Type:** Feature Release (Optional Search Components)
-**Release Date:** February 7, 2026
-**Version Code:** 2.2.0
-**Git Tag:** v2.2.0
-**Backward Compatible:** ✅ Yes (100%)
-
-### 🚀 Core Search Engine Components (Tasks #17 & #20)
-
-**Production-Grade BM25 and Hybrid Search:**
-- ✅ **BM25 Search Engine** (`src/search_engine_v2.py`) - Industry-standard keyword ranking
-  - Pure Python implementation (no external dependencies for algorithm)
-  - Okapi BM25 with configurable parameters (k1=1.5, b=0.75)
-  - <30ms search for 1K memories (target met)
-  - Inverted index with efficient postings
-  - Full tokenization and stopword filtering
-  - CLI interface for testing and demos
-
-- ✅ **Query Optimizer** (`src/query_optimizer.py`) - Intelligent query enhancement
-  - Spell correction using Levenshtein edit distance (max distance: 2)
-  - Query expansion based on term co-occurrence
-  - Boolean operator parsing (AND, OR, NOT, phrase queries)
-  - Technical term preservation (API, SQL, JWT, etc.)
-  - Vocabulary-based correction with graceful fallback
-
-- ✅ **Cache Manager** (`src/cache_manager.py`) - LRU cache for search results
-  - Least Recently Used (LRU) eviction policy
-  - Time-to-live (TTL) support for cache expiration
-  - Thread-safe operations (optional)
-  - Size-based eviction with configurable max entries
-  - Performance tracking (hit rate, evictions, access counts)
-  - <0.1ms cache hit overhead
-
-- ✅ **Hybrid Search System** (`src/hybrid_search.py`) - Multi-method retrieval fusion
-  - Combines BM25 + TF-IDF + Graph traversal
-  - Weighted score fusion with configurable weights
-  - Reciprocal Rank Fusion (RRF) support
-  - <50ms hybrid search for 1K memories (target met)
-  - Automatic integration with MemoryStoreV2
-  - Backward compatible with existing search API
-
-**Key Features:**
-- 🎯 **3x faster search** - BM25 optimized vs basic FTS
-- 📈 **Better relevance** - 15-20% precision improvement over TF-IDF
-- 🧠 **Query intelligence** - Auto-corrects typos, expands terms
-- 🔄 **Multi-method fusion** - Best of keyword, semantic, and graph
-- ⚡ **Production caching** - 30-50% cache hit rates reduce load
-- 📊 **Complete test suite** - `test_search_engine.py` with 8 test cases
-
-**Performance Benchmarks:**
-| Component | Target | Actual | Status |
-|-----------|--------|--------|--------|
-| BM25 Index 1K | <500ms | 247ms | ✅ |
-| BM25 Search 1K | <30ms | 18ms | ✅ |
-| Query Optimizer | <5ms | 2ms | ✅ |
-| Cache Get/Put | <0.5ms | 0.12ms | ✅ |
-| Hybrid Search | <50ms | 35ms | ✅ |
-
-**Attribution:**
-- Copyright headers on all new files
-- MIT License compliance
-- Created by Varun Pratap Bhardwaj
-- Comprehensive documentation: `docs/SEARCH-ENGINE-V2.2.0.md`
-
-### 🚀 Optional Search Components (Tasks #18 & #19)
-
-**New High-Performance Search Infrastructure:**
-- ✅ **HNSW Index** (`src/hnsw_index.py`) - Fast approximate nearest neighbor search
-  - Sub-10ms search for 10K memories
-  - Sub-50ms search for 100K memories
-  - Incremental updates without full rebuild
-  - Disk persistence for instant startup
-  - Graceful fallback to linear search if hnswlib unavailable
-  - Optional dependency: `pip install hnswlib`
-
-- ✅ **Embedding Engine** (`src/embedding_engine.py`) - Local semantic embedding generation
-  - all-MiniLM-L6-v2 model (384 dimensions, 80MB)
-  - GPU acceleration (CUDA/Apple Silicon MPS) with auto-detection
-  - Batch processing: 100-1000 texts/sec (GPU)
-  - LRU cache for 10K embeddings (<1ms cache hits)
-  - Graceful fallback to TF-IDF if sentence-transformers unavailable
-  - Optional dependency: `pip install sentence-transformers`
-
-**Key Features:**
-- 🔄 **Zero breaking changes** - All dependencies optional with graceful fallback
-- ⚡ **10-20x faster search** with HNSW vs linear search
-- 🧠 **True semantic search** with local embeddings (no API calls)
-- 🔒 **Security limits** - MAX_BATCH_SIZE, MAX_TEXT_LENGTH, input validation
-- 📊 **CLI interfaces** - Test and manage both components
-- 📚 **Complete documentation** - `docs/V2.2.0-OPTIONAL-SEARCH.md`
-
-**Performance Benchmarks:**
-| Component | Without Optional Deps | With Optional Deps | Speedup |
-|-----------|----------------------|-------------------|---------|
-| Search (10K) | ~100ms (TF-IDF) | <10ms (HNSW) | 10x |
-| Embeddings | ~50ms (TF-IDF) | 10-100ms (GPU) | Semantic |
-| Cache hit | N/A | <0.001ms | 100,000x |
-
-**Attribution:**
-- Copyright headers on all new files
-- MIT License compliance
-- Created by Varun Pratap Bhardwaj
-
-### 📦 Installation & Dependencies Overhaul
-
-**Better Dependency Management:**
-
-This release reorganizes optional dependencies into modular requirement files, giving users precise control over what features they install.
-
-**Key Improvements:**
-- ✅ **Modular Requirements:** Separate files for different feature sets
-- ✅ **Interactive Installation:** Clear menu with download sizes and install times
-- ✅ **Installation Verification:** Comprehensive health check script
-- ✅ **Zero Breaking Changes:** Existing installations work unchanged
-- ✅ **Better Documentation:** Clear feature isolation and migration guide
-
-### ✨ New Files
-
-**Requirements Structure:**
-- `requirements.txt` - Core requirements (empty - zero dependencies)
-- `requirements-full.txt` - All optional features (~1.5GB)
-- `requirements-ui.txt` - Web dashboard only (~50MB)
-- `requirements-search.txt` - Advanced search only (~1.5GB)
-
-**Installation Tools:**
-- `verify-install.sh` - Comprehensive installation verification
-  - Checks Python version, core files, CLI wrappers, PATH configuration
-  - Verifies optional features (search, UI)
-  - Performance quick test (init + query timing)
-  - Clear status reporting with ✓/○/✗ indicators
-  - Exit codes for CI/CD integration
-
-**Documentation:**
-- `MIGRATION-V2.2.0.md` - Complete migration guide from v2.1.0
-  - 100% backward compatibility confirmation
-  - Step-by-step upgrade instructions
-  - Dependency comparison tables
-  - Troubleshooting section
-  - FAQ
-
-### 🔧 Enhanced Installation Flow
-
-**Old (v2.1.0):**
-```
-Install optional dependencies now? (y/N)
-```
-
-**New (v2.2.0):**
-```
-Optional Features Available:
-
-  1) Advanced Search (~1.5GB, 5-10 min)
-     • Semantic search with sentence transformers
-     • Vector similarity with HNSWLIB
-     • Better search quality
-
-  2) Web Dashboard (~50MB, 1-2 min)
-     • Graph visualization (D3.js)
-     • API server (FastAPI)
-     • Browser-based interface
-
-  3) Full Package (~1.5GB, 5-10 min)
-     • Everything: Search + Dashboard
-
-  N) Skip (install later)
-
-Choose option [1/2/3/N]:
-```
-
-**Benefits:**
-- Users see exactly what they're installing
-- Clear download sizes and installation times
-- Can choose specific features instead of all-or-nothing
-- Can skip and install later with simple commands
-
-### 📋 Requirements Details
-
-**requirements.txt (Core):**
-```txt
-# SuperLocalMemory V2.2.0 has ZERO core dependencies
-# All functionality works with Python 3.8+ standard library only
-```
-
-**requirements-full.txt (All Features):**
-```txt
-sentence-transformers>=2.2.0  # Advanced semantic search
-hnswlib>=0.7.0                # Vector similarity search
-fastapi>=0.109.0              # Web framework
-uvicorn[standard]>=0.27.0     # ASGI server
-python-multipart>=0.0.6       # File upload support
-diskcache>=5.6.0              # Performance caching
-orjson>=3.9.0                 # Fast JSON serialization
-```
-
-**requirements-ui.txt (Dashboard Only):**
-```txt
-fastapi>=0.109.0              # Web server
-uvicorn[standard]>=0.27.0     # ASGI server
-python-multipart>=0.0.6       # Multipart support
-```
-
-**requirements-search.txt (Search Only):**
-```txt
-sentence-transformers>=2.2.0  # Semantic embeddings
-hnswlib>=0.7.0                # ANN search
-```
-
-### 🔍 Installation Verification
-
-**New verify-install.sh script provides comprehensive checks:**
-
-```bash
-./verify-install.sh
-```
-
-**Verification Steps:**
-
-1. **Core Installation Check:**
-   - Python 3.8+ version verification
-   - Installation directory existence
-   - Core scripts (memory_store_v2.py, graph_engine.py, pattern_learner.py)
-   - CLI wrappers (slm, aider-smart)
-   - PATH configuration (shell config + active session)
-   - Database status (size, existence)
-   - Configuration file
-
-2. **Optional Features Check:**
-   - Advanced Search (sentence-transformers, hnswlib)
-   - Web Dashboard (fastapi, uvicorn)
-   - Clear enabled/disabled status
-
-3. **Performance Quick Test:**
-   - Memory store initialization timing
-   - Database query performance (milliseconds)
-
-4. **Summary Report:**
-   - Overall status (WORKING/FAILED)
-   - Feature availability matrix
-   - Next steps recommendations
-   - Error list (if any)
-
-**Exit Codes:**
-- `0` - Installation verified successfully
-- `1` - Installation verification failed
-
-**CI/CD Integration:**
-```bash
-./verify-install.sh || exit 1
-```
-
-### 🔄 Migration from v2.1.0
-
-**Zero Migration Required:**
-
-This release is 100% backward compatible. Existing installations work unchanged.
-
-**Options for Existing Users:**
-
-1. **Keep Current Setup (Recommended):**
-   - No action needed
-   - Run `./verify-install.sh` to verify health
-
-2. **Update to New Structure:**
-   ```bash
-   git pull origin main
-   ./install.sh
-   ./verify-install.sh
-   ```
-
-3. **Manual Dependency Management:**
-   ```bash
-   pip3 install -r requirements-ui.txt      # Dashboard only
-   pip3 install -r requirements-search.txt  # Search only
-   pip3 install -r requirements-full.txt    # Everything
-   ```
-
-See [MIGRATION-V2.2.0.md](MIGRATION-V2.2.0.md) for complete migration guide.
-
-### 📊 Dependency Comparison
-
-| Component | v2.1.0 | v2.2.0 | Change |
-|-----------|--------|--------|--------|
-| Core | 0 deps | 0 deps | Unchanged |
-| UI requirements | Mixed (UI+core) | UI only | Cleaner |
-| Search requirements | None | Separate file | NEW |
-| Full requirements | None | All features | NEW |
-| Version pinning | Exact (==) | Ranges (>=) | More flexible |
-
-### 🎯 User Experience
-
-**For New Users:**
-- Clear installation options with sizes/times
-- Can choose minimal install and add features later
-- Installation verification confirms success
-- Better documentation for troubleshooting
-
-**For Existing Users:**
-- Zero impact - everything continues working
-- Optional update to new structure
-- Can verify installation health anytime
-- Clear migration path if desired
-
-### 🔒 Backward Compatibility
-
-**100% backward compatible - nothing breaks:**
-- ✅ All CLI commands work unchanged
-- ✅ All skills work unchanged
-- ✅ All MCP tools work unchanged
-- ✅ Database schema unchanged
-- ✅ Configuration format unchanged
-- ✅ Data format unchanged
-- ✅ API unchanged
-- ✅ Profile system unchanged
-
-**Upgrade path:** Simply run `./install.sh` or continue using current installation.
-
-### 📝 Documentation Updates
-
-**New Documentation:**
-- `MIGRATION-V2.2.0.md` - Complete migration guide
-- `verify-install.sh` - Installation verification tool
-- Updated `install.sh` - Interactive installation menu
+## [2.5.0] - 2026-02-12
 
-**Enhanced Documentation:**
-- `requirements.txt` - Clear comment about zero dependencies
-- `requirements-full.txt` - Detailed feature descriptions
-- `requirements-ui.txt` - Cleaner, UI-focused
-- `requirements-search.txt` - Search-specific documentation
+**Release Type:** Major Feature Release — "Your AI Memory Has a Heartbeat"
 
-### 🎊 Credits
+SuperLocalMemory transforms from passive storage to active coordination layer. Every memory operation now triggers real-time events.
 
-This maintenance release improves the installation experience while preserving 100% backward compatibility.
+### Added
+- **DbConnectionManager** - SQLite WAL mode, write queue, connection pool (fixes "database is locked")
+- **Event Bus** - Real-time SSE/WebSocket/Webhook broadcasting with tiered retention (48h/14d/30d)
+- **Subscription Manager** - Durable + ephemeral subscriptions with event filters
+- **Webhook Dispatcher** - HTTP POST delivery with exponential backoff retry
+- **Agent Registry** - Track connected AI agents (protocol, write/recall counters, last seen)
+- **Provenance Tracker** - Track memory origin (created_by, source_protocol, trust_score, lineage)
+- **Trust Scorer** - Bayesian trust signal collection (silent in v2.5, enforced in v2.6)
+- **Dashboard: Live Events tab** - Real-time event stream with filters and stats
+- **Dashboard: Agents tab** - Connected agents table with trust scores and protocol badges
 
-**Philosophy:** Better tooling and clearer choices make adoption easier without disrupting existing users.
+### Changed
+- **memory_store_v2.py** - Replaced 15 direct `sqlite3.connect()` calls with DbConnectionManager
+- **mcp_server.py** - Replaced 9 per-call instantiations with shared singleton
+- **ui_server.py** - Refactored from 2008-line monolith to 194-line app shell + 9 route modules
+- **ui/app.js** - Split from 1588-line monolith to 13 modular files
 
-**Author:** Varun Pratap Bhardwaj (Solution Architect)
+### Testing
+- 63 pytest tests + 27 e2e tests + 15 fresh-DB edge cases + 17 backward-compat tests
 
 ---
 
-## [2.2.0] - 2026-02-07
-
-**Release Type:** Major Feature Release - Visualization & Search Enhancement
-**Release Date:** February 7, 2026
-**Version Code:** 2.2.0
-**Git Tag:** v2.2.0
-**Commits Since v2.1.0:** TBD commits
-**Lines Changed:** +2,500 lines (est.)
-**New Files:** 2 files (dashboard.py, Visualization-Dashboard.md wiki page)
-**Backward Compatible:** ✅ Yes (100%)
-
-### 🎨 Visualization Dashboard - MAJOR UPDATE
-
-**Interactive web-based dashboard for visual memory exploration!**
-
-This release introduces a professional-grade visualization dashboard built with Dash and Plotly, transforming SuperLocalMemory from a CLI-only tool into a comprehensive visual knowledge management system.
-
-**Key Highlights:**
-- 🎨 **Interactive Web Dashboard** - Timeline, search, graph visualization, statistics
-- 🔍 **Hybrid Search System** - Combines semantic, FTS5, and graph for 89% precision
-- 📈 **Timeline View** - Chronological visualization with importance color-coding
-- 🕸️ **Graph Visualization** - Interactive force-directed layout with zoom/pan
-- 📊 **Statistics Dashboard** - Real-time analytics with memory trends and tag clouds
-- 🌓 **Dark Mode** - Eye-friendly theme for extended use
-- 🎯 **Advanced Filters** - Multi-dimensional filtering across all views
-- ⌨️ **Keyboard Shortcuts** - Quick navigation and actions
-
-### ✨ Added - New Features
-
-**1. Visualization Dashboard (Layer 9)**
-- ✅ **Four Main Views:**
-  - **Timeline View** - All memories chronologically with importance markers
-  - **Search Explorer** - Real-time search with visual score bars (0-100%)
-  - **Graph Visualization** - Interactive knowledge graph with clusters
-  - **Statistics Dashboard** - Memory trends, tag clouds, pattern insights
-- ✅ **Interactive Features:**
-  - Zoom, pan, drag for graph exploration
-  - Click clusters to see members
-  - Hover tooltips for previews
-  - Expand cards for full details
-- ✅ **Visual Elements:**
-  - Color-coded importance (red/orange/yellow/green)
-  - Cluster badges on each memory
-  - Visual score bars for search results
-  - Chart visualizations (line, pie, bar, word cloud)
-- ✅ **Launch Command:**
-  ```bash
-  python ~/.claude-memory/dashboard.py
-  # Opens at http://localhost:8050
-  ```
-- ✅ **Configuration:**
-  - Custom port support (`--port 8080`)
-  - Profile selection (`--profile work`)
-  - Debug mode (`--debug`)
-  - Config file: `~/.claude-memory/dashboard_config.json`
-- ✅ **Dependencies:**
-  - Dash (web framework)
-  - Plotly (interactive charts)
-  - Pandas (data manipulation)
-  - NetworkX (graph layout)
-  - All optional, graceful degradation without them
-
-**2. Hybrid Search System (Layer 8)**
-- ✅ **Three Search Strategies Combined:**
-  1. **Semantic Search (TF-IDF)** - Conceptual similarity (~45ms)
-  2. **Full-Text Search (FTS5)** - Exact phrase matching (~30ms)
-  3. **Graph-Enhanced Search** - Knowledge graph traversal (~60ms)
-- ✅ **Hybrid Mode (Default):**
-  - Runs all three strategies in parallel
-  - Normalizes scores to 0-100%
-  - Merges results with weighted ranking
-  - Removes duplicates
-  - Total time: ~80ms (minimal overhead)
-- ✅ **Performance Metrics:**
-  - **Precision:** 89% (vs 78% semantic-only)
-  - **Recall:** 91% (vs 82% semantic-only)
-  - **F1 Score:** 0.90 (best balance)
-- ✅ **CLI Usage:**
-  ```bash
-  # Hybrid (default)
-  slm recall "authentication"
-
-  # Specific strategy
-  slm recall "auth" --strategy semantic
-  slm recall "JWT tokens" --strategy fts
-  slm recall "security" --strategy graph
-  ```
-- ✅ **API Usage:**
-  ```python
-  store.search("query", strategy="hybrid")  # Default
-  store.search("query", strategy="semantic")
-  store.search("query", strategy="fts")
-  store.search("query", strategy="graph")
-  ```
-
-**3. Timeline View**
-- ✅ Chronological display of all memories
-- ✅ Importance color-coding:
-  - 🔴 Critical (9-10) - Red
-  - 🟠 High (7-8) - Orange
-  - 🟡 Medium (4-6) - Yellow
-  - 🟢 Low (1-3) - Green
-- ✅ Date range filters (last 7/30/90 days, custom)
-- ✅ Cluster badges showing relationships
-- ✅ Hover tooltips with full content preview
-- ✅ Click to expand full memory details
-- ✅ Export as PDF/HTML
-- ✅ Items per page configurable (default: 50)
-
-**4. Search Explorer**
-- ✅ Real-time search (updates as you type)
-- ✅ Visual score bars (0-100% relevance)
-- ✅ Strategy toggle dropdown (semantic/fts/graph/hybrid)
-- ✅ Result highlighting (matched keywords)
-- ✅ Cluster context for each result
-- ✅ Advanced filters:
-  - Minimum score threshold (slider)
-  - Date range (calendar picker)
-  - Tags (multi-select)
-  - Importance level (slider 1-10)
-  - Clusters (multi-select)
-  - Projects (dropdown)
-- ✅ Export results (JSON/CSV)
-- ✅ Keyboard navigation (arrow keys, Enter to select)
-
-**5. Graph Visualization**
-- ✅ Interactive force-directed layout
-- ✅ Zoom (mouse wheel) and pan (drag background)
-- ✅ Drag nodes to rearrange
-- ✅ Click clusters to focus
-- ✅ Click entities to see connected memories
-- ✅ Hover for node/edge details
-- ✅ Cluster coloring (unique color per cluster)
-- ✅ Edge thickness = relationship strength
-- ✅ Node size = connection count
-- ✅ Layout options:
-  - Force-directed (default)
-  - Circular (equal spacing)
-  - Hierarchical (tree-like)
-- ✅ Performance limits:
-  - Max nodes: 500 (configurable)
-  - Min edge weight: 0.3 (hide weak connections)
-- ✅ Export as PNG/SVG
-
-**6. Statistics Dashboard**
-- ✅ **Memory Trends (Line Chart):**
-  - Memories added over time (daily/weekly/monthly)
-  - Toggleable date ranges (7d, 30d, 90d, all)
-  - Growth rate calculation
-- ✅ **Tag Cloud (Word Cloud):**
-  - Most frequent tags sized by usage
-  - Color schemes (configurable)
-  - Minimum frequency filter
-- ✅ **Importance Distribution (Pie Chart):**
-  - Breakdown of importance levels 1-10
-  - Percentages and counts
-- ✅ **Cluster Sizes (Bar Chart):**
-  - Number of memories per cluster
-  - Sortable by size or name
-  - Top N filter
-- ✅ **Pattern Confidence (Table):**
-  - Learned patterns with confidence scores
-  - Filter by threshold (e.g., 60%+)
-  - Sort by confidence or frequency
-- ✅ **Access Heatmap (Calendar Heatmap):**
-  - Memory access frequency over time
-  - Color intensity = access count
-  - Date range selector
-
-**7. Advanced Filtering**
-- ✅ Multi-dimensional filters:
-  - Date range (preset + custom)
-  - Tags (multi-select, AND/OR logic)
-  - Importance (range slider 1-10)
-  - Clusters (multi-select)
-  - Projects (dropdown)
-  - Score threshold (search only)
-- ✅ Filter combinations (multiple filters simultaneously)
-- ✅ Filter persistence (saved across sessions)
-- ✅ Export/import filter presets
-- ✅ Reset filters button
-
-**8. User Experience Enhancements**
-- ✅ **Dark Mode:**
-  - Toggle switch in top-right corner
-  - Automatic OS theme detection (optional)
-  - High contrast colors for readability
-  - Preference saved across sessions
-- ✅ **Keyboard Shortcuts:**
-  - `Ctrl+1/2/3/4` - Switch views
-  - `Ctrl+F` - Focus search box
-  - `Ctrl+D` - Toggle dark mode
-  - `Ctrl+R` - Refresh current view
-  - `Esc` - Close modal/overlay
-- ✅ **Responsive Design:**
-  - Works on desktop, tablet, mobile
-  - Touch gestures for graph (zoom/pan)
-  - Optimized layouts for small screens
-- ✅ **Real-time Updates:**
-  - CLI changes appear immediately in dashboard
-  - Auto-refresh on database changes
-  - No manual reload needed
-
-**9. Documentation**
-- ✅ **New Wiki Page:** `Visualization-Dashboard.md` (2,000+ words)
-  - Complete dashboard guide
-  - Getting started tutorial
-  - Feature tour with examples
-  - Configuration options
-  - Performance tips
-  - Troubleshooting section
-  - Screenshot placeholders
-  - Use cases for developers and teams
-- ✅ **Updated Wiki Pages:**
-  - `Universal-Architecture.md` - Added Layer 8 and Layer 9
-  - `Installation.md` - Added "Start Visualization Dashboard" section
-  - `Quick-Start-Tutorial.md` - Added "Step 5: Explore Dashboard"
-  - `FAQ.md` - Added 5 questions about UI and search
-  - `_Sidebar.md` - Added Visualization-Dashboard link
-- ✅ **Updated README.md:**
-  - Added "Visualization Dashboard" section
-  - Added "Advanced Search" section with strategies table
-  - Added "Performance" section with benchmark tables
-  - Updated architecture diagram (9 layers)
-  - SEO keywords integrated naturally
-
-### 🔧 Enhanced
-
-**Architecture:**
-- ✅ Expanded from 7-layer to **9-layer architecture**:
-  - Layer 9: Visualization (NEW)
-  - Layer 8: Hybrid Search (NEW)
-  - Layers 1-7: Unchanged (backward compatible)
-- ✅ All layers share the same SQLite database (no duplication)
-- ✅ Dashboard reads from existing database (zero migration)
-
-**Search System:**
-- ✅ **Hybrid search (default):**
-  - Combines semantic + FTS5 + graph
-  - 89% precision (vs 78% semantic-only)
-  - 91% recall (vs 82% semantic-only)
-  - F1 score: 0.90
-- ✅ **Strategy selection:**
-  - CLI flag: `--strategy semantic|fts|graph|hybrid`
-  - API parameter: `strategy="hybrid"`
-  - Dashboard dropdown: visual toggle
-- ✅ **Score normalization:**
-  - All strategies output 0-100% scores
-  - Consistent across CLI, API, dashboard
-  - Visual bars in dashboard
-
-**Performance:**
-- ✅ **Dashboard load times:**
-  - 100 memories: < 100ms
-  - 500 memories: < 300ms
-  - 1,000 memories: < 500ms
-  - 5,000 memories: < 2s
-- ✅ **Search speeds (hybrid):**
-  - 100 memories: 55ms
-  - 500 memories: 65ms
-  - 1,000 memories: 80ms
-  - 5,000 memories: 150ms
-- ✅ **Graph rendering:**
-  - 100 nodes: < 200ms
-  - 500 nodes: < 500ms
-  - 1,000 nodes: < 1s (with limits)
-- ✅ **Timeline rendering:**
-  - 1,000 memories: < 300ms
-  - 5,000 memories: < 1s
-
-**Configuration:**
-- ✅ New config file: `~/.claude-memory/dashboard_config.json`
-  - Port and host settings
-  - Default view preference
-  - Timeline pagination
-  - Search defaults
-  - Graph layout options
-  - Statistics refresh interval
-  - Cache settings
-
-### 📊 Performance Benchmarks
-
-**Hybrid Search vs Single Strategy (500 memories):**
-
-| Strategy | Time | Precision | Recall | F1 Score |
-|----------|------|-----------|--------|----------|
-| Semantic | 45ms | 78% | 82% | 0.80 |
-| FTS5 | 30ms | 92% | 65% | 0.76 |
-| Graph | 60ms | 71% | 88% | 0.79 |
-| **Hybrid** | **80ms** | **89%** | **91%** | **0.90** |
-
-**Dashboard Performance (Load Times):**
-
-| Dataset Size | Timeline | Search | Graph | Stats |
-|--------------|----------|--------|-------|-------|
-| 100 memories | 100ms | 35ms | 200ms | 150ms |
-| 500 memories | 200ms | 45ms | 500ms | 300ms |
-| 1,000 memories | 300ms | 55ms | 1s | 500ms |
-| 5,000 memories | 1s | 85ms | 3s | 2s |
-
-**Scalability:**
-
-| Memories | Hybrid Search | Dashboard Load | Graph Build | RAM Usage |
-|----------|---------------|----------------|-------------|-----------|
-| 100 | 55ms | < 100ms | 0.5s | < 30MB |
-| 500 | 65ms | < 300ms | 2s | < 50MB |
-| 1,000 | 80ms | < 500ms | 5s | < 80MB |
-| 5,000 | 150ms | < 2s | 30s | < 150MB |
-| 10,000 | 300ms | < 5s | 90s | < 250MB |
-
-### 🔒 Backward Compatibility
-
-**100% backward compatible - nothing breaks:**
-- ✅ All v2.1 and v2.0 commands work unchanged
-- ✅ Database schema unchanged (only additions, no modifications)
-- ✅ Configuration format unchanged (new optional fields only)
-- ✅ API unchanged (only additions, no breaking changes)
-- ✅ Performance unchanged or improved (no regressions)
-- ✅ Profile system unchanged
-- ✅ MCP integration unchanged
-- ✅ Skills unchanged
-- ✅ CLI unchanged
-
-**Optional features:**
-- Dashboard requires optional dependencies (dash, plotly, pandas, networkx)
-- Graceful degradation if dependencies not installed
-- CLI and API work without dashboard dependencies
-- Installation prompts for optional dependencies (no forced install)
-
-**Upgrade path:**
-- Simply run `git pull && ./install.sh`
-- Existing memories preserved
-- No migration required
-- Dashboard auto-configures on first launch
-- Optional: `pip install dash plotly pandas networkx` for dashboard
-
-### 🐛 Fixed
-
-**No bugs introduced (pure feature addition release)**
-
-### 🔐 Security
-
-**No security changes (maintained v2.1.0 security posture):**
-- Dashboard binds to localhost only (127.0.0.1)
-- No external network access
-- 100% local processing
-- No telemetry or analytics
-- Optional dependencies validated
-
-### 💡 Breaking Changes
-
-**NONE - 100% backward compatible**
-
-### 📝 Documentation Updates
-
-**README.md:**
-- ✅ Added "Visualization Dashboard" section after Quick Start
-- ✅ Added "Advanced Search" section with strategies table
-- ✅ Added "Performance" section with benchmark tables
-- ✅ Updated architecture diagram to 9 layers
-- ✅ SEO keywords: visualization, dashboard, semantic search, timeline view, hybrid search
-
-**Wiki Pages (New):**
-- ✅ `Visualization-Dashboard.md` (2,000+ words)
-  - Complete dashboard guide
-  - Feature tour with examples
-  - Configuration and troubleshooting
-  - Use cases for developers and teams
-  - Screenshot placeholders
-  - SEO optimized
-
-**Wiki Pages (Updated):**
-- ✅ `Universal-Architecture.md` - Added Layer 8 (Hybrid Search) and Layer 9 (Visualization), updated to 9-layer
-- ✅ `Installation.md` - Added "Start Visualization Dashboard" section
-- ✅ `Quick-Start-Tutorial.md` - Added "Step 5: Explore Dashboard"
-- ✅ `FAQ.md` - Added 5 questions about UI and search
-- ✅ `_Sidebar.md` - Added link to Visualization-Dashboard
-
-**Attribution:**
-- ✅ Varun Pratap Bhardwaj attribution on all new pages
-- ✅ Maintained existing attribution throughout
-
-### 🎊 Credits
-
-This release transforms SuperLocalMemory from a CLI-only tool into a **comprehensive visual knowledge management system** while maintaining 100% backward compatibility and the core principle of local-first, privacy-preserving operation.
-
-**Philosophy:** Advanced features should enhance, not replace. The CLI remains powerful and fast, while the dashboard adds visual exploration for users who need it.
-
-**Acknowledgments:**
-- Built on Dash (Plotly) for interactive visualizations
-- TF-IDF and FTS5 for hybrid search
-- Co-authored with Claude Sonnet 4.5
-- Solution Architect: Varun Pratap Bhardwaj
+## [2.4.2] - 2026-02-11
 
----
+**Release Type:** Bug Fix
 
-## [2.1.0-universal] - 2026-02-07
-
-**Release Type:** Major Feature Release
-**Release Date:** February 7, 2026
-**Version Code:** 2.1.0-universal
-**Git Tag:** v2.1.0-universal
-**Commits Since v2.0.0:** 18 commits
-**Lines Changed:** +3,375 lines, -320 lines (net: +3,055)
-**New Files:** 22 files
-**Backward Compatible:** ✅ Yes (100%)
-
-### 🌐 Universal Integration - MAJOR UPDATE
-
-**SuperLocalMemory now works across ALL IDEs and CLI tools!**
-
-This release transforms SuperLocalMemory from Claude-Code-only to a universal memory system that integrates with 11+ tools while maintaining 100% backward compatibility.
-
-**Key Highlights:**
-- 🌐 **11+ IDE Support:** Cursor, Windsurf, Claude Desktop, Continue.dev, Cody, Aider, ChatGPT, Perplexity, Zed, OpenCode, Antigravity
-- 🔧 **Three-Tier Access:** MCP + Skills + CLI (all use same database)
-- 🤖 **6 Universal Skills:** remember, recall, list-recent, status, build-graph, switch-profile
-- 🛠️ **MCP Server:** 6 tools, 4 resources, 2 prompts
-- 🔒 **Attribution Protection:** 6-layer protection system with legal compliance
-- 📊 **Knowledge Graph:** Leiden clustering with TF-IDF entity extraction
-- 🧠 **Pattern Learning:** Multi-dimensional identity extraction with confidence scoring
-- 🚀 **Zero Config:** Auto-detection and configuration during installation
-- 📝 **Comprehensive Docs:** 1,400+ lines of new documentation
-
-### 🔧 Post-Release Enhancements (Same Day)
-
-**Documentation Additions:**
-- ✅ `docs/MCP-MANUAL-SETUP.md` - Comprehensive manual setup guide for 8+ additional tools
-  - ChatGPT Desktop App integration
-  - Perplexity AI integration
-  - Zed Editor configuration
-  - OpenCode setup instructions
-  - Antigravity IDE configuration
-  - Custom MCP client examples (Python/HTTP)
-- ✅ `docs/UNIVERSAL-INTEGRATION.md` - Complete universal strategy documentation (15,000+ words)
-- ✅ `docs/MCP-TROUBLESHOOTING.md` - Debugging guide with 20+ common issues and solutions
-
-**Enhanced Documentation:**
-- ✅ `ARCHITECTURE.md` - Added universal integration architecture section
-- ✅ `QUICKSTART.md` - Improved three-tier access method documentation
-- ✅ `docs/CLI-COMMANDS-REFERENCE.md` - Enhanced with new `slm` wrapper commands
-- ✅ `README.md` - Added V3 cross-reference and version comparison
-
-**Critical Bug Fixes:**
-- ✅ Fixed MCP server method calls to match actual API:
-  - `store.list_memories()` → `store.list_all()`
-  - `engine.get_clusters()` → `engine.get_stats()`
-  - `learner.get_context()` → `learner.get_identity_context()`
-- ✅ Enhanced MCP server startup banner with version info
-- ✅ Improved config file formatting for better readability
-
-**Total IDE Support:** 11+ tools (Cursor, Windsurf, Claude Desktop, Continue.dev, Cody, Aider, ChatGPT, Perplexity, Zed, OpenCode, Antigravity, plus any terminal)
-
-### ✨ Added - New Integrations
-
-**MCP (Model Context Protocol) Integration:**
-- ✅ Cursor IDE - Native MCP support with auto-configuration
-- ✅ Windsurf IDE - Full MCP integration
-- ✅ Claude Desktop - Built-in MCP server support
-- ✅ VS Code Continue - MCP tools accessible to AI
-- Auto-detection during installation
-- Zero manual configuration required
-
-**Enhanced Skills Support:**
-- ✅ Continue.dev - Slash commands (`/slm-remember`, `/slm-recall`, `/slm-list-recent`, `/slm-status`, `/slm-build-graph`, `/slm-switch-profile`)
-- ✅ Cody - Custom commands integrated (all 6 skills)
-- ✅ Claude Code - Native skills (unchanged, backward compatible)
-- Auto-configuration for detected tools
-- Backward compatible with existing Claude Code skills
-
-### 🎯 Universal Skills System
-
-**6 Production-Ready Skills:**
-
-1. **slm-remember** - Save content with intelligent indexing
-   - Automatic entity extraction for knowledge graph
-   - Pattern learning from saved content
-   - Tags, project, and importance metadata
-   - Full documentation in `skills/slm-remember/SKILL.md`
-
-2. **slm-recall** - Search memories with multi-method retrieval
-   - Semantic search via TF-IDF vectors
-   - Full-text search via SQLite FTS5
-   - Knowledge graph context enhancement
-   - Confidence-scored results
-   - Full documentation in `skills/slm-recall/SKILL.md`
-
-3. **slm-list-recent** - Display recent memories
-   - Configurable limit (default 10)
-   - Formatted output with metadata
-   - Quick context retrieval
-   - Full documentation in `skills/slm-list-recent/SKILL.md`
-
-4. **slm-status** - System health and statistics
-   - Memory count and database size
-   - Knowledge graph statistics (clusters, entities)
-   - Pattern learning statistics
-   - Current profile info
-   - Full documentation in `skills/slm-status/SKILL.md`
-
-5. **slm-build-graph** - Build/rebuild knowledge graph
-   - Leiden clustering algorithm
-   - TF-IDF entity extraction
-   - Auto-cluster naming
-   - Relationship discovery
-   - Full documentation in `skills/slm-build-graph/SKILL.md`
-
-6. **slm-switch-profile** - Change active profile
-   - Isolated memory contexts
-   - Use cases: work/personal/client separation
-   - Profile-specific graphs and patterns
-   - Full documentation in `skills/slm-switch-profile/SKILL.md`
-
-**Skills Architecture:**
-- Metadata-first design (SKILL.md in each skill directory)
-- Version tracked (2.1.0)
-- MIT licensed with attribution preserved
-- Compatible with Claude Code, Continue.dev, Cody
-- Progressive disclosure (simple → advanced usage)
-- Comprehensive documentation (100+ lines per skill)
-
-**Universal CLI Wrapper:**
-- ✅ New `slm` command - Simple syntax for any terminal
-- ✅ `aider-smart` wrapper - Auto-context injection for Aider CLI
-- Works with any scripting environment
-- Bash and Zsh completion support
-
-### 📦 New Files
-
-**Core:**
-- `mcp_server.py` - Complete MCP server implementation (6 tools, 4 resources, 2 prompts)
-- `bin/slm` - Universal CLI wrapper
-- `bin/aider-smart` - Aider integration with auto-context
-
-**Configurations:**
-- `configs/claude-desktop-mcp.json` - Claude Desktop MCP config
-- `configs/cursor-mcp.json` - Cursor IDE MCP config
-- `configs/windsurf-mcp.json` - Windsurf IDE MCP config
-- `configs/continue-mcp.yaml` - Continue.dev MCP config
-- `configs/continue-skills.yaml` - Continue.dev slash commands
-- `configs/cody-commands.json` - Cody custom commands
-
-**Completions:**
-- `completions/slm.bash` - Bash autocomplete
-- `completions/slm.zsh` - Zsh autocomplete
-
-### 🔧 Enhanced
-
-**install.sh:**
-- Auto-detects installed IDEs (Cursor, Windsurf, Claude Desktop, Continue, Cody)
-- Auto-configures MCP server for detected tools
-- Installs MCP SDK if not present
-- Installs universal CLI wrapper
-- Configures shell completions
-- Zero breaking changes to existing installation
-
-**install-skills.sh:**
-- Detects Continue.dev and configures slash commands
-- Detects Cody and configures custom commands
-- Backs up existing configurations
-- Smart merging for existing configs
-
-**README.md:**
-- Added "Works Everywhere" section
-- Updated comparison table with universal integration
-- New CLI commands section (simple + original)
-- Auto-detection documentation
-
-### 🎯 User Experience
-
-**For Existing Users:**
-- ✅ Zero breaking changes - all existing commands work unchanged
-- ✅ Automatic upgrade path - just run `./install.sh`
-- ✅ New tools auto-configured during installation
-- ✅ Original skills preserved and functional
-
-**For New Users:**
-- ✅ One installation works everywhere
-- ✅ Auto-detects and configures all tools
-- ✅ Simple CLI commands (`slm remember`)
-- ✅ Zero manual configuration
-
-### 🏗️ Architecture
-
-**Three-Tier Access Model:**
-1. **MCP** (Modern) - Native IDE integration via Model Context Protocol
-2. **Skills** (Enhanced) - Slash commands in Claude, Continue, Cody
-3. **CLI** (Universal) - Simple commands that work anywhere
-
-**All tiers use the SAME local SQLite database** - no data duplication, no conflicts.
-
-### 📊 Compatibility Matrix
-
-| Tool | Integration Method | Status |
-|------|-------------------|--------|
-| Claude Code | Skills (unchanged) | ✅ |
-| Cursor | MCP Auto-configured | ✅ |
-| Windsurf | MCP Auto-configured | ✅ |
-| Claude Desktop | MCP Auto-configured | ✅ |
-| Continue.dev | MCP + Skills | ✅ |
-| Cody | Custom Commands | ✅ |
-| Aider | Smart Wrapper | ✅ |
-| Any Terminal | Universal CLI | ✅ |
-
-### 🐛 Fixed
-
-**Critical Fixes (Pre-Release):**
-- Fixed MCP server method calls to match actual API:
-  - `store.list_memories()` → `store.list_all()` (correct method name)
-  - `engine.get_clusters()` → `engine.get_stats()` (correct method name)
-  - `learner.get_context()` → `learner.get_identity_context()` (correct method name)
-- Fixed Python script references in CLI hooks (memory_store_v2.py path)
-- Fixed shell detection for PATH configuration (bash vs zsh)
-- Fixed auto-configure PATH for truly global CLI access
-
-**Installation Fixes:**
-- Interactive optional dependencies installation (no forced installs)
-- Proper error handling for missing Python packages
-- Better dependency detection (scikit-learn, leidenalg)
-- Fixed database auto-initialization with full V2 schema
-
-**MCP Server Fixes:**
-- Fixed non-existent method calls causing startup failures
-- Enhanced error messages with specific method names
-- Proper JSON formatting in config files
-- Added version info to startup banner
-
-**CLI Fixes:**
-- Fixed slm wrapper command not found issues
-- Corrected aider-smart script permissions
-- Fixed bash completion path detection
-- Proper symlink handling for bin directory
-
-**Documentation Fixes:**
-- Corrected installation paths in all documentation
-- Fixed broken internal links
-- Updated version numbers consistently
-- Improved troubleshooting steps
-
-### 🔒 Backward Compatibility
-
-**100% backward compatible - nothing breaks:**
-- ✅ All existing skills work unchanged
-- ✅ All bash commands work unchanged
-- ✅ Database schema unchanged (only additions, no modifications)
-- ✅ Configuration format unchanged (only new optional fields)
-- ✅ Performance unchanged (no regressions)
-- ✅ Profile system unchanged
-- ✅ API unchanged (only additions, no breaking changes)
-
-**Upgrade path:** Simply run `./install.sh` - new features auto-configure while preserving existing functionality.
-
-**Migration notes:** None required - v2.0.0 users can upgrade seamlessly.
-
-### 📝 Documentation
-
-**New Documentation:**
-- Universal integration implementation plan (15,000+ words)
-- Testing checklist (150+ test cases)
-- Progress tracking system
-- Per-tool quick-start guides
-- `docs/MCP-MANUAL-SETUP.md` - Manual configuration guide for 8+ additional IDEs
-- `docs/MCP-TROUBLESHOOTING.md` - Comprehensive troubleshooting guide
-- `docs/UNIVERSAL-INTEGRATION.md` - Complete universal strategy documentation
-
-**Updated Documentation:**
-- README.md - Universal positioning and V3 cross-reference
-- INSTALL.md - Auto-detection details
-- ARCHITECTURE.md - Universal integration architecture
-- QUICKSTART.md - Three-tier access methods
-- CLI-COMMANDS-REFERENCE.md - New slm commands
-
-### 🔐 Attribution Protection System
-
-**Multi-Layer Attribution Protection:**
-- ✅ **Layer 1: Source Code Headers** - Copyright headers in all Python files (legally required)
-- ✅ **Layer 2: Documentation Attribution** - Footer attribution in all markdown files
-- ✅ **Layer 3: Database-Level Attribution** - Creator metadata embedded in SQLite database
-  - `creator_metadata` table with cryptographic signature
-  - Includes: creator name, role, GitHub, project URL, license, version
-  - Verification hash: `sha256:c9f3d1a8b5e2f4c6d8a9b3e7f1c4d6a8b9c3e7f2d5a8c1b4e6f9d2a7c5b8e1`
-- ✅ **Layer 4: Runtime Attribution** - Startup banners display attribution
-- ✅ **Layer 5: License-Based Protection** - MIT License with explicit attribution requirements
-- ✅ **Layer 6: Digital Signature** - Cryptographic signature in ATTRIBUTION.md
-
-**New Attribution Files:**
-- `ATTRIBUTION.md` - Comprehensive attribution requirements and enforcement
-- `docs/ATTRIBUTION-PROTECTION-SUMMARY.md` - Multi-layer protection documentation
-- `ATTRIBUTION-IMPLEMENTATION-REPORT.md` - Technical implementation details
-
-**API Enhancements:**
-- `MemoryStoreV2.get_attribution()` - Retrieve creator metadata from database
-- Attribution display in MCP server startup banner
-- Attribution preserved in all skills metadata
-
-**Legal Compliance:**
-- MIT License with attribution requirements clearly documented
-- Prohibited uses explicitly stated (credit removal, impersonation, rebranding)
-- Enforcement procedures documented
-- Digital signature for authenticity verification
-
-### 🔒 Security
-
-**Security Hardening (v2.0.0 foundation):**
-- ✅ **API Server:** Binds to localhost only (127.0.0.1) instead of 0.0.0.0
-  - Prevents external network access
-  - Only local processes can connect
-  - No exposure to public internet
-
-- ✅ **Path Traversal Protection:** Profile management validates paths
-  - Prevents directory traversal attacks (../)
-  - Sanitizes user input for file paths
-  - Restricts operations to designated directories
-
-- ✅ **Input Validation:** Size limits on all user inputs
-  - Content: 1MB maximum
-  - Summary: 10KB maximum
-  - Tags: 50 characters each, 20 tags maximum
-  - Prevents memory exhaustion attacks
-
-- ✅ **Resource Limits:** Graph build limits
-  - Maximum 5000 memories per graph build
-  - Prevents CPU/memory exhaustion
-  - Graceful degradation for large datasets
-
-- ✅ **No External Dependencies:** Zero external API calls
-  - No telemetry or tracking
-  - No auto-updates
-  - No cloud sync
-  - Complete air-gap capability
-
-- ✅ **Data Integrity:** SQLite ACID transactions
-  - Atomic operations
-  - Consistent state even on crashes
-  - Automatic backups before destructive operations
-
-**Privacy Guarantees:**
-- 100% local storage (no cloud sync)
-- No telemetry or analytics
-- No external network calls
-- User owns all data
-- Standard filesystem permissions
-
-### 🎊 Credits
-
-This release was completed in a single day with parallel implementation streams, comprehensive testing, and zero breaking changes to existing functionality.
-
-**Philosophy:** Universal integration should be additive, not disruptive. Every existing user's workflow remains unchanged while gaining new capabilities automatically.
-
-**Acknowledgments:**
-- Built on research from GraphRAG (Microsoft), PageIndex (Meta AI), xMemory (Stanford)
-- Co-authored with Claude Sonnet 4.5
-- Solution Architect: Varun Pratap Bhardwaj
+### Fixed
+- Profile isolation bug in UI dashboard - Graph stats now filter by active profile
 
 ---
 
-## [2.0.0] - 2026-02-05
-
-### Initial Release - Complete Rewrite
-
-SuperLocalMemory V2 represents a complete architectural rewrite with intelligent knowledge graphs, pattern learning, and enhanced organization capabilities.
-
----
+## [2.4.1] - 2026-02-11
 
-## Added - New Features
-
-### 4-Layer Architecture
-
-**Layer 1: Enhanced Storage**
-- SQLite database with FTS5 full-text search
-- Tag management system
-- Metadata support for extensibility
-- Parent-child memory relationships
-- Compression tiers (1-3) for space optimization
-
-**Layer 2: Hierarchical Index (PageIndex-inspired)**
-- Tree structure for memory organization
-- Parent-child relationship management
-- Breadcrumb navigation paths
-- Contextual grouping capabilities
-- Fast ancestor/descendant queries
-
-**Layer 3: Knowledge Graph (GraphRAG)**
-- TF-IDF entity extraction from memories
-- Leiden clustering algorithm for relationship discovery
-- Auto-naming of thematic clusters
-- Similarity-based memory connections
-- Graph statistics and visualization data
-- Related memory suggestions
-
-**Layer 4: Pattern Learning (xMemory-inspired)**
-- Frequency analysis across memories
-- Context extraction for user preferences
-- Multi-category pattern recognition:
-  - Framework preferences (React, Vue, Angular, etc.)
-  - Language preferences (Python, JavaScript, etc.)
-  - Architecture patterns (microservices, monolith, etc.)
-  - Security approaches (JWT, OAuth, etc.)
-  - Coding style priorities
-- Confidence scoring (0.0-1.0 scale)
-- Identity profile generation for AI context
-
-### 🕸️ Knowledge Graph Features (GraphRAG)
-
-**Leiden Clustering Algorithm:**
-- **Community Detection:** Finds thematic groups automatically without manual tagging
-- **Resolution Parameter:** Adjustable granularity (default: 1.0)
-- **Deterministic:** Same memories always produce same clusters
-- **Scalable:** Handles 100-500 memories efficiently
-- **Quality Metrics:** Modularity scoring for cluster quality
-
-**TF-IDF Entity Extraction:**
-- **Automatic Entity Discovery:** Extracts important terms from memories
-- **Frequency-based Weighting:** More important = higher weight
-- **Stop Word Filtering:** Removes common words (the, and, etc.)
-- **Case Insensitive:** "React" and "react" treated as same entity
-- **Minimum Threshold:** Only entities with TF-IDF score > 0.1
-
-**Cluster Auto-Naming:**
-- **Smart Name Generation:** Uses top entities to create descriptive cluster names
-- **Multiple Strategies:**
-  - Single dominant entity: "React Development"
-  - Multiple related entities: "JWT & OAuth Security"
-  - Topic grouping: "Performance Optimization"
-- **Fallback:** "Topic 1", "Topic 2" if auto-naming fails
-
-**Relationship Discovery:**
-- **Similarity-Based Connections:** Cosine similarity between memory vectors
-- **Related Memory Suggestions:** Find memories related to a specific memory
-- **Cross-Cluster Relationships:** Discovers connections across thematic groups
-- **Strength Scoring:** 0.0-1.0 similarity scores for relationships
-
-**Graph Statistics:**
-- Total clusters count
-- Cluster size distribution (min/max/average)
-- Total entities extracted
-- Memory distribution across clusters
-- Isolated memories (not in any cluster)
-
-**MCP Integration:**
-- `build_graph()` tool - Rebuild entire graph
-- `memory://graph/clusters` resource - View all clusters
-- Graph statistics in `get_status()` tool
-- Cluster information in search results
-
-**Example Clusters Discovered:**
-- "Authentication & Security" (JWT, tokens, OAuth, sessions)
-- "Frontend Development" (React, components, hooks, state)
-- "Performance Optimization" (caching, indexes, queries, speed)
-- "Database Design" (SQL, schema, migrations, relationships)
-- "API Development" (REST, GraphQL, endpoints, versioning)
-
-### 🧠 Pattern Learning System (xMemory)
-
-**Multi-dimensional Analysis:**
-
-1. **Framework Preferences:**
-   - Detects: React, Vue, Angular, Svelte, Next.js, etc.
-   - Confidence scoring based on frequency
-   - Example: "React (73% confidence)" means 73% of frontend mentions use React
-
-2. **Language Preferences:**
-   - Detects: Python, JavaScript, TypeScript, Go, Rust, etc.
-   - Context-aware (API vs frontend vs backend)
-   - Example: "Python for APIs, TypeScript for frontend"
-
-3. **Architecture Patterns:**
-   - Detects: Microservices, monolith, serverless, event-driven
-   - Style preferences (REST vs GraphQL, SQL vs NoSQL)
-   - Example: "Microservices (58% confidence)"
-
-4. **Security Approaches:**
-   - Detects: JWT, OAuth, API keys, certificates
-   - Session management patterns
-   - Example: "JWT tokens (81% confidence)"
-
-5. **Coding Style Priorities:**
-   - Detects: Performance vs readability, TDD vs pragmatic
-   - Testing preferences (Jest, Pytest, etc.)
-   - Example: "Performance over readability (58% confidence)"
-
-6. **Domain Terminology:**
-   - Learns project-specific terms
-   - Industry vocabulary (fintech, healthcare, etc.)
-   - Team conventions
-
-**Confidence Scoring Algorithm:**
-- **Frequency-based:** More mentions = higher confidence
-- **Recency weighting:** Recent patterns weighted more
-- **Threshold:** Only patterns with >30% confidence reported
-- **Statistical:** Uses standard deviation for significance
-
-**Adaptive Learning:**
-- Patterns evolve with new memories
-- Automatic recomputation on pattern update
-- Incremental learning (no full rebuild required)
-- Context decay for old patterns
-
-**Identity Context Generation:**
-- Creates AI assistant context from learned patterns
-- Configurable confidence threshold (default: 0.5)
-- Formatted for Claude/GPT prompt injection
-- Example output:
-  ```
-  Your Coding Identity:
-  - Framework preference: React (73% confidence)
-  - Language: Python for backends (65% confidence)
-  - Style: Performance-focused (58% confidence)
-  - Testing: Jest + React Testing Library (65% confidence)
-  - API style: REST over GraphQL (81% confidence)
-  ```
-
-**MCP Integration:**
-- `memory://patterns/identity` resource - View learned patterns
-- Pattern statistics in `get_status()` tool
-- Automatic pattern learning on `remember()` calls
-- Identity context in AI tool prompts
-
-**Storage:**
-- `learned_patterns` table in SQLite
-- Includes: category, pattern, confidence, frequency, last_seen
-- Queryable via SQL for custom analysis
-- Preserved across profile switches
-
-### Compression System
-
-- **Progressive Summarization:**
-  - Tier 1: Original full content (recent memories)
-  - Tier 2: 60% compression via intelligent summarization
-  - Tier 3: 96% compression via cold storage archival
-
-- **Age-based Tiering:** Automatic promotion based on memory age
-- **Lossless Archive:** Tier 3 memories stored in JSON format
-- **Space Savings:** 60-96% reduction for older memories
-
-### Profile Management
-
-- **Multi-Profile Support:** Separate memory contexts
-- **Isolated Databases:** Each profile has independent storage
-- **Profile Switching:** Easy context changes via CLI
-- **Use Cases:**
-  - Separate work/personal memories
-  - Client-specific knowledge bases
-  - Project-specific contexts
-  - Team collaboration spaces
-
-**CLI Commands:**
-```bash
-memory-profile create <name>
-memory-profile switch <name>
-memory-profile list
-memory-profile delete <name>
-```
-
-### Reset System
-
-- **Soft Reset:** Clear memories, preserve schema and configuration
-- **Hard Reset:** Complete database deletion with confirmation
-- **Layer-Selective Reset:** Reset specific layers (graph, patterns, etc.)
-- **Automatic Backups:** Created before all destructive operations
-- **Safety Confirmations:** Required for hard resets
-
-**Reset options:**
-```bash
-memory-reset soft              # Clear data, keep structure
-memory-reset hard --confirm    # Nuclear option
-memory-reset layer --layers graph patterns  # Selective reset
-```
-
-### CLI Enhancements
-
-**New Commands:**
-- `memory-status` - System overview and statistics
-- `memory-profile` - Profile management
-- `memory-reset` - Safe reset operations
-
-**Improved Output:**
-- Color-coded status indicators
-- Progress bars for long operations
-- Detailed error messages
-- Safety warnings for destructive actions
+**Release Type:** Hierarchical Clustering & Documentation
 
+### Added
+- **Hierarchical Leiden clustering** - Large clusters auto-subdivided up to 3 levels deep
+- **Community summaries** - TF-IDF structured reports for every cluster
+- Schema migration for `summary`, `parent_cluster_id`, `depth` columns
 
 ---
 
-## Changed - Improvements Over V1
-
-### Performance Enhancements
-
-**Search Speed:**
-- V1: ~150ms average
-- V2: ~45ms average
-- **Improvement: 3.3x faster**
-
-**Graph Building:**
-- 20 memories: <0.03 seconds
-- 100 memories: ~2 seconds
-- 500 memories: ~15 seconds
-
-**Database Efficiency:**
-- With compression: 60% smaller for aged memories
-- With archival: 96% reduction for old memories
-
-### Architecture Improvements
-
-**V1 Limitations:**
-- Flat memory storage
-- No relationship discovery
-- Manual organization only
-- No pattern learning
-- Single profile
-
-**V2 Enhancements:**
-- 4-layer intelligent architecture
-- Auto-discovered relationships
-- Hierarchical organization
-- Pattern learning with confidence scores
-- Multi-profile support
-
-### Search Improvements
-
-**V1:**
-- Basic keyword search
-- Tag filtering
-- No relationship context
-
-**V2:**
-- FTS5 full-text search (faster)
-- Graph-enhanced results
-- Related memory suggestions
-- Cluster-based discovery
-- Pattern-informed context
-
-### User Experience
+## [2.4.0] - 2026-02-11
 
-**Better Feedback:**
-- Progress indicators for long operations
-- Detailed statistics (graph, patterns, compression)
-- Safety confirmations for destructive actions
-- Clear error messages with suggestions
+**Release Type:** Profile System & Intelligence
 
-**Easier Management:**
-- Profile switching via simple commands
-- Visual status dashboard
-- Automated maintenance tasks
-- Comprehensive CLI help
+### Added
+- **Column-based memory profiles** - Single DB with `profile` column, instant switching via `profiles.json`
+- **Auto-backup system** - Configurable interval, retention policy, one-click backup from UI
+- **MACLA confidence scorer** - Beta-Binomial Bayesian posterior (arXiv:2512.18950)
+- **UI: Profile Management** - Create, switch, delete profiles from dashboard
+- **UI: Settings tab** - Auto-backup config, history, profile management
+- **UI: Column sorting** - Click headers in Memories table
 
 ---
 
-## Technical Details
-
-### Dependencies
-
-**Core System:**
-- Python 3.8+ (required)
-- SQLite 3.35+ (usually pre-installed)
-- Python standard library only (no external packages)
-
-**Optional Enhancements:**
-- scikit-learn (for advanced TF-IDF)
-- leidenalg (for advanced clustering)
-
-**Fallback implementations provided** for systems without optional dependencies.
-
-### Database Schema
-
-**New Tables:**
-```sql
--- Graph storage
-CREATE TABLE graph_clusters (...)
-CREATE TABLE graph_cluster_members (...)
-CREATE TABLE graph_entities (...)
-
--- Pattern learning
-CREATE TABLE learned_patterns (...)
-
--- Compression
-CREATE TABLE compression_archives (...)
-```
-
-**Enhanced Tables:**
-```sql
--- Memory enhancements
-ALTER TABLE memories ADD COLUMN tier INTEGER DEFAULT 1;
-ALTER TABLE memories ADD COLUMN parent_id INTEGER;
-```
+## [2.3.7] - 2026-02-09
 
-### API Changes
+### Added
+- `--full` flag to show complete memory content without truncation
+- Smart truncation: <5000 chars shown in full, ≥5000 chars truncated to 2000 chars
 
-**Initial Configuration:**
-- Database automatically initialized on first run
-- Default config.json provided
-- CLI commands available immediately after installation
+### Fixed
+- CLI bug: `get` command now uses correct `get_by_id()` method
 
 ---
 
-## Research Foundation
-
-SuperLocalMemory V2 is built on cutting-edge 2026 research:
-
-**GraphRAG (Microsoft Research):**
-- Knowledge graph construction from unstructured text
-- Community detection for clustering
-- Entity extraction and relationship mapping
-
-**PageIndex (Meta AI):**
-- Hierarchical indexing for fast navigation
-- Tree-based memory organization
-- Contextual grouping strategies
+## [2.3.5] - 2026-02-09
 
-**xMemory (Stanford):**
-- Identity pattern learning from interactions
-- Preference extraction with confidence scoring
-- Adaptive context generation
+### Added
+- **ChatGPT Connector Support** - `search(query)` and `fetch(id)` MCP tools
+- **Streamable HTTP transport** - `slm serve --transport streamable-http`
+- **UI enhancements** - Memory detail modal, dark mode, export buttons, search score bars
 
-**A-RAG (Multi-level Retrieval):**
-- Layer-based retrieval architecture
-- Progressive information density
-- Context-aware search
+### Fixed
+- XSS vulnerability via inline onclick replaced with safe event delegation
 
 ---
 
-## Performance Benchmarks
-
-### Search Performance
-
-| Memories | V1 Search | V2 Search | Improvement |
-|----------|-----------|-----------|-------------|
-| 20       | 120ms     | 30ms      | 4.0x        |
-| 100      | 150ms     | 45ms      | 3.3x        |
-| 500      | 200ms     | 60ms      | 3.3x        |
-
-### Graph Building
+## [2.3.0] - 2026-02-08
 
-| Memories | Build Time | Clusters | Entities |
-|----------|-----------|----------|----------|
-| 20       | 0.03s     | 3-5      | 10-15    |
-| 100      | 2.0s      | 10-15    | 40-60    |
-| 500      | 15s       | 30-50    | 150-250  |
+**Release Type:** Universal Integration
 
-### Storage Efficiency
+SuperLocalMemory now works across 16+ IDEs and CLI tools.
 
-| Tier | Description | Compression | Use Case |
-|------|-------------|-------------|----------|
-| 1    | Full content | 0%         | Recent memories |
-| 2    | Summarized   | 60%        | 30-90 days old |
-| 3    | Archived     | 96%        | 90+ days old |
+### Added
+- **MCP configs** - Auto-detection for Cursor, Windsurf, Claude Desktop, Continue.dev, Codex, Copilot, Gemini, JetBrains
+- **Universal CLI wrapper** - `slm` command works in any terminal
+- **Skills installer expansion** - Cursor, VS Code/Copilot auto-configured
+- **Tool annotations** - `readOnlyHint`, `destructiveHint`, `openWorldHint` for all MCP tools
 
 ---
 
-## Getting Started
-
-### First-Time Setup
-
-```bash
-# 1. Install system
-./install.sh
-
-# 2. Verify installation
-memory-status
-
-# 3. Build initial graph (after adding memories)
-python3 ~/.claude-memory/graph_engine.py build
+## [2.2.0] - 2026-02-07
 
-# 4. Learn initial patterns (after adding memories)
-python3 ~/.claude-memory/pattern_learner.py update
-```
+**Release Type:** Feature Release (Optional Search Components)
 
-**System Features:**
-- Automatic database initialization
-- Default profile created on first run
-- Graph and pattern infrastructure ready to use
-- Profile management available from the start
+### Added
+- **BM25 Search Engine** - Okapi BM25 with <30ms search for 1K memories
+- **Query Optimizer** - Spell correction, query expansion, technical term preservation
+- **Cache Manager** - LRU cache with TTL, <0.1ms cache hit overhead
+- **Hybrid Search System** - BM25 + TF-IDF + Graph fusion, <50ms for 1K memories
+- **HNSW Index** - Sub-10ms search for 10K memories (optional: `pip install hnswlib`)
+- **Embedding Engine** - Local semantic embeddings with GPU acceleration (optional: `pip install sentence-transformers`)
+- **Modular Requirements** - Separate files: `requirements-ui.txt`, `requirements-search.txt`, `requirements-full.txt`
+- **Installation Verification** - `verify-install.sh` script for comprehensive health checks
 
 ---
 
-## Known Limitations
-
-### Current Limitations
-
-**1. Scalability:**
-- Optimal for < 500 memories
-- Graph builds take longer with 1000+ memories
-- Recommendation: Use profile splitting for large datasets
-
-**2. Language Support:**
-- Entity extraction optimized for English
-- Other languages may have reduced clustering quality
-
-**3. Compression:**
-- Manual trigger required (no auto-compression yet)
-- Tier promotion based on age only (not access patterns)
+## [2.1.0-universal] - 2026-02-07
 
-**4. Graph:**
-- Full rebuild required for updates (no incremental)
-- Clustering deterministic but may vary with algorithm parameters
+**Release Type:** Major Feature Release - Universal Integration
 
-### Future Improvements
+### Added
+- **6 Universal Skills** - remember, recall, list-recent, status, build-graph, switch-profile
+- **MCP Server** - 6 tools, 4 resources, 2 prompts for native IDE integration
+- **6-layer Attribution Protection** - Source headers, docs, database metadata, runtime banners, license, digital signature
+- **11+ IDE Support** - Cursor, Windsurf, Claude Desktop, Continue.dev, Cody, Aider, ChatGPT, Perplexity, Zed, OpenCode, Antigravity
 
-Planned for future releases:
-- Incremental graph updates
-- Auto-compression based on access patterns
-- Multi-language entity extraction
-- Graph visualization UI
-- Real-time pattern updates
+### Documentation
+- `docs/MCP-MANUAL-SETUP.md` - Manual setup guide for 8+ additional IDEs
+- `docs/MCP-TROUBLESHOOTING.md` - Debugging guide with 20+ common issues
+- `docs/UNIVERSAL-INTEGRATION.md` - Complete universal strategy (15,000+ words)
 
 ---
 
-## Security
-
-### Privacy-First Design
-
-**No External Communication:**
-- Zero API calls
-- No telemetry
-- No auto-updates
-- No cloud sync
-
-**Local-Only Storage:**
-- All data on your machine
-- Standard filesystem permissions
-- Full user control
-
-**Data Integrity:**
-- SQLite ACID transactions
-- Automatic backups
-- Schema validation
+## [2.0.0] - 2026-02-05
 
-See [SECURITY.md](SECURITY.md) for complete security policy.
+### Initial Release - Complete Rewrite
 
----
+SuperLocalMemory V2 represents a complete architectural rewrite with intelligent knowledge graphs, pattern learning, and enhanced organization.
 
-## Acknowledgments
+### Added
+- **4-Layer Architecture** - Enhanced Storage, Hierarchical Index, Knowledge Graph, Pattern Learning
+- **TF-IDF Entity Extraction** - Automatic entity discovery with frequency weighting
+- **Leiden Clustering** - Community detection for automatic thematic grouping
+- **Pattern Learning** - Multi-dimensional analysis (frameworks, languages, architecture, security, coding style)
+- **Compression System** - Progressive summarization (Tier 1: 0%, Tier 2: 60%, Tier 3: 96%)
+- **Profile Management** - Multi-profile support with isolated databases
 
-Built on research from:
-- **GraphRAG** (Microsoft Research) - Knowledge graph construction
-- **PageIndex** (Meta AI) - Hierarchical indexing
-- **xMemory** (Stanford) - Identity pattern learning
-- **A-RAG** - Multi-level retrieval architecture
+### Performance
+- 3.3x faster search (45ms vs 150ms in V1)
+- 60-96% storage reduction with compression
 
-Special thanks to the AI research community for advancing local-first, privacy-preserving systems.
+### Research Foundation
+Built on GraphRAG (Microsoft), PageIndex (VectifyAI), MemoryBank (AAAI 2024), A-RAG
 
 ---
 
-## Links
+## Archive
 
-- **Homepage:** [GitHub Repository](https://github.com/varun369/SuperLocalMemoryV2)
-- **Documentation:** [docs/](docs/)
-- **Installation:** [INSTALL.md](INSTALL.md)
-- **Quick Start:** [QUICKSTART.md](QUICKSTART.md)
-- **Architecture:** [ARCHITECTURE.md](ARCHITECTURE.md)
-- **Contributing:** [CONTRIBUTING.md](CONTRIBUTING.md)
-- **Security:** [SECURITY.md](SECURITY.md)
+See [CHANGELOG-ARCHIVE.md](CHANGELOG-ARCHIVE.md) for detailed release notes from v2.0.0 through v2.4.x.
 
 ---
 
@@ -1904,10 +282,9 @@ We use [Semantic Versioning](https://semver.org/):
 - **MINOR:** New features (backward compatible, e.g., 2.0.0 → 2.1.0)
 - **PATCH:** Bug fixes (backward compatible, e.g., 2.1.0 → 2.1.1)
 
-**Current Version:** v2.3.0-universal
-**Previous Version:** v2.2.0
-**Next Planned:** v2.4.0 (incremental graph updates, auto-compression)
-**npm:** `npm install -g superlocalmemory` (available since v2.1.0)
+**Current Version:** v2.6.5
+**Website:** [superlocalmemory.com](https://superlocalmemory.com)
+**npm:** `npm install -g superlocalmemory`
 
 ---
 
@@ -1915,19 +292,6 @@ We use [Semantic Versioning](https://semver.org/):
 
 SuperLocalMemory V2 is released under the [MIT License](LICENSE).
 
-**TL;DR:** Free to use, modify, and distribute for any purpose.
-
----
-
-**Ready to get started?**
-
-See [INSTALL.md](INSTALL.md) for installation instructions and [QUICKSTART.md](QUICKSTART.md) for your first 5 minutes.
-
 ---
 
-**Questions or feedback?**
-
-- Open an issue: [GitHub Issues](https://github.com/varun369/SuperLocalMemoryV2/issues)
-- Start a discussion: [GitHub Discussions](https://github.com/varun369/SuperLocalMemoryV2/discussions)
-
 **100% local. 100% private. 100% yours.**