v2.85.0
3 months ago by Jordan Fanapour
New Features
- LocalEmbeddings Provider: New local embedding provider using Transformers.js for privacy-focused, offline embedding generation without API keys
- Ollama Integration: Complete Ollama provider support for local LLM inference with streaming capabilities
- Memory Vector Service: Lightweight in-memory vector similarity search for small to medium datasets without external databases
- Query Vector Embeddings: Automatic embedding generation for Query entities to enable semantic search and similarity matching
- Qwen Coder 30B Model: Added Alibaba's Qwen 3 Coder 30B model optimized for software development tasks
Improvements
- Enhanced Metadata Sync: Improved dependency ordering with topological sorting for complex entity relationships
- Better Logging: Cleaner metadata sync logging with clear messages for record creation vs updates
- LMStudio Enhancements: Fixed LMStudio implementation with improved model selection and API key handling
- Vector Storage: Added EmbeddingVector and EmbeddingModelID fields to Query entity for vector storage
- Model Management: Automatic selection of highest-ranked embedding models by PowerRank
Bug Fixes
- Metadata Sync Reliability: Fixed confusing error messages and improved handling of missing records during push operations
- Query Field Cleanup: Removed non-existent Sequence field from QueryEntity GraphQL schema
- Dependency Processing: Handles circular dependencies gracefully with proper warnings