qmd

What This Skill Does

qmd is a high-performance, local search engine specifically optimized for Markdown-based knowledge bases and documentation collections. Unlike general-purpose grep tools, qmd provides a hybrid search approach, enabling users to index vast amounts of Markdown content and retrieve relevant information via BM25 keyword matching or vector-based semantic searching. By prioritizing local execution, it ensures that your sensitive notes and documents never leave your local environment, making it a privacy-first solution for managing personal or professional knowledge graphs.

Installation

To get started with qmd, ensure you have Bun version 1.0.0 or higher installed. On macOS, you must also have SQLite extensions installed via brew install sqlite. Once prerequisites are met, install the tool globally using: bun install -g https://github.com/tobi/qmd. After installation, run qmd collection add to map your document directories, then execute qmd embed to generate the necessary vectors for intelligent semantic searching. Ensure that $HOME/.bun/bin is present in your system PATH to allow the agent to execute commands successfully.

Use Cases

This skill is ideal for professionals who rely on large Obsidian vaults, Zettelkasten systems, or technical documentation repositories. It is most effective when searching for conceptual links between documents that don't share identical keywords or when performing rapid full-text retrieval across heterogeneous Markdown files. Use it to consolidate information from scattered local files into a unified, queryable database.

Example Prompts

1. "Search my notes for all references to my project architecture and provide a summary of the database schema."
2. "I need to find related notes regarding my research on distributed systems; please search my markdown collection for relevant insights."
3. "Retrieve the documentation file about API authentication from my local docs folder using a keyword search."

Tips & Limitations

The default search mode, qmd search, should be your primary tool as it is essentially instant. Reserve qmd vsearch for instances where traditional keyword matches fail, as it invokes a local LLM and can be significantly slower. Avoid qmd query for standard interactions due to its high computational cost and risk of timeouts. Note that qmd is not a replacement for code-specific search tools; its chunking algorithm is optimized for prose, not syntax-heavy source code repositories.