qmd

What This Skill Does

The qmd skill acts as a high-performance local intelligence engine for your filesystem. Unlike standard command-line tools like 'find' or 'grep' which rely on literal string matching, qmd integrates three distinct search methodologies: BM25 keyword matching, vector-based semantic retrieval, and local LLM-powered reranking. By indexing your directories, the tool creates a searchable knowledge graph on your local machine, allowing the OpenClaw agent to retrieve relevant information from markdown notes, documentation, or codebase snippets without ever sending data to an external server. It effectively turns your unstructured local files into a queryable database.

Installation

To integrate this skill, use the ClawHub package manager. Run the following command in your terminal: clawhub install openclaw/skills/skills/bheemreddy181/qmd-search Ensure you have sufficient disk space to accommodate the local embedding and reranking models (embeddinggemma-300M, qwen3-reranker-0.6b, and Qwen3-0.6B) which will be downloaded upon the initial initialization of the skill.

Use Cases

• Documentation Retrieval: Quickly locate specific API definitions or setup instructions across large documentation folders using semantic queries rather than guessing file names.
• Code Exploration: Identify implementation patterns or configurations within large repositories where standard 'find' commands return too many irrelevant noise results.
• Knowledge Management: Search personal notes, journals, or research files by concept. For example, asking about 'machine learning best practices' will surface relevant snippets even if the exact keywords are not present in every file.
• Rapid Discovery: Replace slow, recursive system find commands with indexed lookup for instantaneous project file identification.

Example Prompts

1. "Use qmd to search the 'projects' collection for any notes related to setting up Docker for local development."
2. "I need to find the implementation logic for the user authentication middleware. Search the codebase and extract the relevant code snippets."
3. "Search my notes for information on 'project architecture' and provide a summary of the findings using the retrieved context."

Tips & Limitations

• Indexing is Mandatory: Always run qmd update whenever you add significant amounts of new content to your indexed folders to ensure the vector store remains accurate.
• Scoping: Leverage the -c flag for collections to prevent searching through irrelevant directories, which improves both speed and result relevance.
• Resource Usage: Because this tool runs heavy local models, initial embedding creation can be resource-intensive. Perform this during idle hours.
• Limitation: The current version is optimized for text-heavy files (Markdown/Python/Text). Binary files and highly compressed formats are not supported.