qmd

What This Skill Does

qmd is a high-performance, local-first search engine specifically optimized for Markdown-based knowledge bases, notes, and documentation. By leveraging a hybrid approach, it allows users to navigate large collections of personal or professional Markdown files without needing to rely on cloud-based indexing or external third-party services. The skill implements BM25 keyword search as its primary engine for instant retrieval, while providing advanced semantic search capabilities for situations where keyword matching is insufficient to find conceptually related content.

Installation

To get started with qmd, ensure you have Bun version 1.0.0 or higher installed on your system. For macOS users, ensure the SQLite extensions are available by running brew install sqlite. Once prerequisites are met, install the tool globally via: bun install -g https://github.com/tobi/qmd. After installation, add your directories via qmd collection add /path/to/notes and run qmd embed to generate the vector data necessary for semantic searches. Ensure your PATH includes $HOME/.bun/bin to allow OpenClaw to invoke the binary directly.

Use Cases

This skill is ideal for knowledge workers, developers, and researchers who maintain personal wikis, Zettelkasten note systems, or local project documentation. It is designed to handle 'messy' markdown files by indexing content-based chunks rather than relying on strict header hierarchies. Use it when you need to quickly locate information across hundreds of files, correlate distinct topics that may not share identical keywords, or synthesize information from a fragmented collection of local markdown documents.

Example Prompts

1. "Search my notes for the meeting minutes regarding the Q3 budget review."
2. "I need to find related notes about my recent research into decentralized architecture; please perform a semantic search in my knowledge base."
3. "Retrieve the markdown document about the API authentication flow from my project collection."

Tips & Limitations

Performance is a key consideration for this skill. qmd search (BM25) is the most efficient method and should be the default for 90% of requests because it is typically instant. Reserve qmd vsearch for when exact keyword matches fail, as it invokes a local model that may take up to a minute to initialize on some machines. Avoid qmd query unless you absolutely require LLM-reranked results, as the latency is significantly higher. Note that this is not a general-purpose code search tool; for source code repositories, standard grep or AST-based tools are preferred. Keep your index updated by re-running qmd embed whenever you make significant changes to your note collection to ensure retrieval accuracy.