qmd

What This Skill Does

The qmd skill is a powerful command-line interface tool integrated into the OpenClaw agent ecosystem, specifically designed for high-efficiency information retrieval from markdown-based knowledge bases, such as Obsidian vaults. Unlike naive search tools that might index entire file contents and consume excessive tokens, qmd utilizes a sophisticated hybrid approach. By combining BM25 keyword matching with vector-based semantic embeddings, it provides targeted, context-aware snippets rather than raw file dumps. This methodology results in a massive 96% reduction in token usage, ensuring that the AI agent operates with both speed and cost-effectiveness. The tool handles indexing, collection management, and complex search queries locally, maintaining data privacy while ensuring high-performance recall for large knowledge sets.

Installation

To integrate this skill, use the ClawHub manager: clawhub install openclaw/skills/skills/anshumanbh/anshumanbh-qmd. Once installed, ensure you have the necessary runtime dependencies. If not pre-installed, you can set it up via bun install -g https://github.com/tobi/qmd. After installation, initialize your knowledge source by running qmd collection add ~/path/to/your/notes --name my-vault, followed by qmd embed --collection my-vault to build the initial semantic index for vector searches.

Use Cases

This skill is indispensable for knowledge workers, developers, and researchers who maintain personal knowledge management (PKM) systems in markdown. It is ideal for retrieving specific technical documentation, synthesizing cross-referenced notes during complex projects, or finding obscure conceptual connections that a standard 'find' command would miss. Whether you are searching for a specific function name in a codebase or trying to recall a philosophical concept from your diary, qmd adapts to your retrieval needs.

Example Prompts

1. "/qmd search 'error handling strategy' --collection obsidian-notes"
2. "/qmd 'how to refactor legacy python modules' --semantic"
3. "/qmd --setup"

Tips & Limitations

To maximize effectiveness, always prioritize BM25 search for exact strings or code snippets, as it is faster and more precise for literal matches. Reserve semantic (vector) searches for fuzzy queries where synonyms or intent-based logic are required. Note that performance is directly tied to the quality of your embeddings; re-run qmd embed whenever you make major structural changes to your vault. Avoid using hybrid search for every query, as the LLM-based reranking step adds significant latency; use it only when standard searches fail to retrieve the desired context.