qmd

You can index local files for searching by adding directories to a custom collection and updating the index. The system uses BM25 and vector embeddings, storing the generated indexes locally in the ~/.cache/qmd directory.

Hybrid search for local document retrieval works by combining traditional BM25 keyword matching with vector similarity. This approach uses local Ollama models to generate embeddings and rerank results, yielding highly relevant content from your indexed collections.

Yes, you can use Ollama for local embeddings and reranking. The tool integrates directly with Ollama to generate vector embeddings and rerank search results, ensuring the entire local file indexing and retrieval process remains completely private.

The best way to search across a large collection of Markdown notes is to add them to a local collection and update the index. You can then perform targeted searches using BM25, vector similarity, or a hybrid approach to quickly retrieve specific topics.

When adding specific file types to collections, you must apply custom masks to filter directories. Indexes are stored locally in ~/.cache/qmd, so limitations depend on available local storage and the performance of your Ollama instance for embeddings.

Yes, this local search CLI supports updating existing indexes. After initially adding directories to a collection, you can run update commands to refresh the index, ensuring your BM25 and vector search results reflect the most recent file changes.