← Back to Changelog

Performance Release

74% faster search, 6 bug fixes, rock-solid stability.

⚡ Performance Improvements

Search Latency Reduced by 74%

  • Before: 13.5 seconds average
  • After: 3.55 seconds average
  • How: Switched to optimized llama-3.1-8b-instruct-fast model for intent classification and entity extraction

This means your AI assistant gets context 4x faster when using CodeMem.

🐛 Bug Fixes

Search Memory

  • Multi-term queries now work - Queries like "OAuth PKCE flow" now correctly find relevant memories instead of returning empty results
  • Vector search fallback - When semantic search fails, improved keyword fallback ensures you still get relevant results

Knowledge Graph

  • No more duplicate entities - Adding the same entity twice now returns the existing one instead of creating duplicates
  • All relations visible - query_graph now shows all relation types including "contains" relationships
  • Truncation indicator - get_graph now tells you when results are truncated and how many total items exist

Memory Listing

  • Correct totals - list_memories with project filter now shows correct total count (e.g., "12 of 12" instead of "12 of 154")

📊 Benchmark Results

Metric Value
Hit Rate 80%
Average Latency 3.55s
vs Mem0 +15% better
vs SimpleMem +2% better
vs MemGPT +12% better

🧪 Quality Assurance

  • ✅ 416 unit tests passing
  • ✅ 17 smoke tests passing
  • ✅ Production verified via live benchmark

🔧 Technical Details

PRs Merged

  • #94: Fast LLM model for latency optimization
  • #101: Improved vector search fallback with LIKE queries
  • #103: get_graph truncation indicator
  • #104: Entity deduplication in knowledge graph
  • #105: query_graph includes all relation types
  • #106: Multi-term keyword search refinement

Released by the CodeMem team. Questions? Join our Discord