🚀 MCP Memory Service v3.2.0

Major Improvements

This release resolves critical semantic search functionality and significantly improves the reliability of the SQLite-vec backend.

✅ Issues Resolved

• #64: SQLite-vec semantic search returning 0 results - FIXED
• #67: HTTP API search endpoints showing 0.0 similarity scores - COMPLETELY RESOLVED

🔧 Critical Fixes

Semantic Search Restoration

• Before: Semantic search returned results but all similarity scores were 0.0
• After: Meaningful similarity scores (0.2+ instead of 0.0) with proper ranking

Zero Vector Embeddings Fixed

• Problem: All 28+ embeddings in databases were invalid zero vectors
• Solution: Enhanced repair tools detect and regenerate proper embeddings
• Result: All HTTP API search endpoints now work correctly

Dependencies Moved to Core

• sentence-transformers and torch moved from optional to core dependencies
• Prevents silent failures during embedding generation
• Ensures reliable semantic search functionality out of the box

🛠️ New Tools Added

Diagnostic & Repair Suite

• repair_zero_embeddings.py - Enhanced repair for invalid zero vector embeddings
• analyze_sqlite_vec_db.py - Database analysis without dependencies
• check_sqlite_vec_status.py - Health checker for SQLite-vec databases
• migrate_sqlite_vec_embeddings.py - Full migration with timestamped backups
• test_sqlite_vec_embeddings.py - Comprehensive diagnostic test suite

Zero Data Loss Migration

• All repair tools preserve existing memory content
• Automatic backup creation before any changes
• Multiple repair strategies (quick repair → full migration)

🌐 HTTP API Endpoints Working

All search endpoints now function correctly with meaningful results:

• ✅ POST /api/search - Semantic similarity search
• ✅ POST /api/search/by-tag - Tag-based search
• ✅ POST /api/search/by-time - Time-based recall
• ✅ GET /api/search/similar/{hash} - Find similar memories

📊 Performance & Reliability

• Semantic search: ~400-500ms (includes embedding generation)
• Tag search: ~20-30ms (database query only)
• Memory listing: ~10-20ms (simple query)
• Database repair: ~2-3 minutes for 30+ memories

🔄 Migration Guide

For existing installations with broken semantic search:

Option 1: Quick Repair (Try First)

python3 scripts/repair_zero_embeddings.py /path/to/sqlite_vec.db

Option 2: Full Migration (If Needed)

python3 scripts/migrate_sqlite_vec_embeddings.py /path/to/sqlite_vec.db

Update Dependencies

# Reinstall with new core dependencies
uv pip install -e .
# or
pip install -e .

🧪 Testing & Verification

Comprehensive Test Suite

# Test all functionality
python3 scripts/test_sqlite_vec_embeddings.py

# Check database health
python3 scripts/check_sqlite_vec_status.py /path/to/db

# Analyze database structure
python3 scripts/analyze_sqlite_vec_db.py /path/to/db

HTTP API Testing

# Semantic search
curl -X POST "http://localhost:8000/api/search" \
  -H "Content-Type: application/json" \
  -d '{"query": "embedding model", "n_results": 3}'

# Tag search  
curl -X POST "http://localhost:8000/api/search/by-tag" \
  -H "Content-Type: application/json" \
  -d '{"tags": ["test"]}'

🔗 Related Work

This release brings Issue #64 (Remote Memory Bridge) significantly closer to completion by resolving the critical semantic search backend problems.

⚠️ Breaking Changes

None - All changes are backward compatible. Migration tools preserve existing data.

💝 Acknowledgments

Special thanks to the community for reporting these critical issues and helping test the fixes.

Full Changelog: v3.1.0...v3.2.0

🤖 Generated with Claude Code

Co-Authored-By: Claude noreply@anthropic.com