QMD - Query Markup Documents

What is QMD - Query Markup Documents?

QMD is a mini command-line search engine designed for local document retrieval, indexing markdown notes, meeting transcripts, documentation, and knowledge bases. It operates entirely on-device, ensuring privacy and offline capability. The tool combines three search methodologies: BM25 full-text search for keyword matching, vector semantic search for understanding context and meaning, and LLM re-ranking to refine results using local language models via node-llama-cpp with GGUF models. This hybrid approach delivers highly relevant results from natural language queries or traditional keywords.

Key features include multiple output formats (JSON, files) optimized for agentic workflows, and an MCP (Model Context Protocol) server for tight integration with AI agents like Claude. The MCP server supports both stdio and HTTP transports, allowing shared, long-lived server instances that avoid repeated model loading. The HTTP server binds to localhost by default and exposes endpoints for querying and searching, returning URIs in the format qmd://collection/path. LLM models stay loaded in VRAM across requests, with embedding and reranking contexts disposed after 5 minutes of idle time and transparently recreated on the next request (incurring a ~1 second penalty).

Use cases include personal knowledge management, developer documentation search, meeting note retrieval, and integration into AI agent pipelines. QMD can be used as a standalone CLI tool or as a library in Node.js or Bun applications via createStore() with three modes. Technical details: it relies on local GGUF models for LLM capabilities, supports MCP for agent integration, and provides configurable host and port settings. The tool is ideal for users who want a private, fast, and accurate search engine for their local documents without relying on cloud services.

Key Features

Who is it for?

developers, knowledge workers, AI agents, researchers, technical writers, privacy, conscious users