bear-app-rag
Local-first semantic search for Bear notes. The installed package has two direct dependencies, uses local ONNX embeddings, and exposes retrieval through MCP.
Documentation
- Architecture – The current pipeline, module by module
- Building a RAG Pipeline Without LangChain – The engineering choices and corrections behind the project
- Evaluation Methodology – What the benchmark measures and where its conclusions stop
- Benchmark Visualization – Charts generated from the committed eval artifact
- Architecture Decision Records – Eight decisions, including the later embedding-model comparison