qmd transforms Claude's ability to navigate local directories by replacing slow, traditional search operations with advanced, local-first search technology. It integrates BM25 full-text search, vector semantic search, and LLM-powered reranking to find specific files, code snippets, or conceptual patterns without requiring external API keys. This skill is essential for developers and researchers managing large collections of notes, technical documentation, or complex codebases where standard file system searches often fail to provide relevant context efficiently.
主な機能
01Local LLM Reranking: Uses Qwen and Gemma models locally to prioritize the most relevant search results.
02Collection Management: Scopes searches to specific projects or folders to reduce noise and improve speed.
03Precision Retrieval: Extract specific line ranges or full documents with markdown formatting and line numbers.
0416 GitHub stars
05Performance Optimization: Designed to handle large directories without the performance lag typical of standard find commands.
06Hybrid Search: Combines BM25 keyword matching with vector-based semantic search for maximum accuracy.
ユースケース
01Performing semantic searches across personal knowledge bases and technical documentation to find conceptual answers.
02Gathering relevant context and snippets from local files before prompting Claude for code generation or editing.
03Locating specific code implementations or patterns across large, complex local repositories.