GitHub - tobi/qmd: mini cli search engine for your docs, knowledge bases, meeting notes, whatever. Tracking current sota approaches while being all local
4 hours ago
- #knowledge management
- #local AI
- #on-device search
- QMD is an on-device search engine designed for indexing and searching markdown notes, meeting transcripts, documentation, and knowledge bases locally.
- It combines BM25 full-text search, vector semantic search, and LLM re-ranking, all running locally via node-llama-cpp with GGUF models, ensuring privacy and offline functionality.
- The tool supports installation via npm or Bun, and can be run globally or directly with npx/bunx, offering commands for creating collections, adding context, generating embeddings, and performing searches.
- QMD features a hybrid search pipeline that includes query expansion, parallel retrieval from FTS and vector indexes, RRF fusion with position-aware blending, and LLM re-ranking to optimize relevance and preserve exact matches.
- It integrates with agentic workflows through structured JSON and file outputs, and exposes an MCP server for seamless integration with tools like Claude Desktop, enabling tools for querying, retrieving documents, and batch operations.
- As a library, QMD provides an SDK for Node.js and Bun applications, allowing developers to create stores, perform searches, manage collections and contexts, and customize embedding models, with support for multilingual corpora via environment variables.
- Key features include context addition for improved search relevance, AST-aware chunking for code files, editor URI configuration for clickable results, and a detailed scoring system that indicates relevance from highly to low relevant.
- The underlying architecture uses SQLite for indexing, with smart chunking algorithms that respect markdown and code structure, and models are cached locally, supporting customization via HuggingFace URIs for different use cases.