Qmd

Qmd is a local search and indexing CLI that brings powerful document retrieval to your filesystem. It combines BM25 keyword search, vector embeddings, and hybrid queries with an optional MCP (Model Context Protocol) mode for seamless integration into AI agent workflows.

Instead of grepping through files or relying on memory, Qmd indexes your documents locally and lets you search them using the best algorithm for each query type — keyword, semantic, or both.

Key Features

• Multiple search algorithms including BM25 for keyword matching, vector search via embeddings, and hybrid queries that combine both approaches
• Local-first indexing with all data stored at ~/.cache/qmd, keeping your content on your machine
• Collection management with pattern-based file selection for organizing different document sets
• Document retrieval with line-number specifications for precise content extraction
• MCP mode for integrating search capabilities directly into AI agent workflows through the Model Context Protocol
• Ollama integration for local embedding generation and result reranking without sending data to external services

Use Cases

• Index and search project documentation, READMEs, and markdown files across large codebases
• Perform semantic searches to find conceptually related content even when exact keywords differ
• Integrate document search into AI coding workflows via MCP mode for context-aware assistance
• Build searchable collections of reference materials organized by project or topic

How It Works

Users create collections by specifying file paths and patterns (e.g., all Markdown files in a directory). Qmd indexes the content locally using BM25 and optionally generates vector embeddings via a local Ollama instance. Searches can use keyword matching, semantic similarity, or a hybrid of both approaches, with results optionally reranked for relevance.

Getting Started

Install globally with npm. For basic BM25 keyword search, no additional setup is needed. For vector search and reranking, install and run Ollama locally at localhost:11434. Create a collection from your documents and start searching.