qmd

Offline search of local notes and documents is achieved by indexing your knowledge base on-device, enabling fast retrieval without internet dependency. The system applies hybrid BM25 keyword matching, vector search, and LLM reranking to deliver high-quality results.

Hybrid retrieval combines BM25 keyword matching with vector search and LLM reranking to improve local knowledge base search. This multi-layered approach ensures both precise keyword hits and semantic context are captured, yielding higher quality on-device results.

Yes, you can use MCP integration to connect agent workflows with local search. The tooling supports both CLI usage and MCP connections, automatically downloading required models and offering daemon options to maintain warm performance during retrieval.

Yes, on-device search supports indexing for project transcripts, notes, and various documents. It processes these local files to enable fast retrieval, applying hybrid search techniques to extract relevant content across offline environments.

To maintain fast search performance for local knowledge bases, you can utilize daemon options that keep the service warm. Combined with automatic model downloads and on-device indexing, this ensures rapid hybrid retrieval during your queries.

No, you do not need to manually download models to run vector search and LLM reranking. The system features automatic model downloads, seamlessly setting up the necessary components for on-device hybrid retrieval.