Qmd

qmd is a local semantic search engine and RAG knowledge base indexer that functions as a Model Context Protocol server. It converts local documents, markdown files, and codebases into a searchable database to provide retrieval augmented generation capabilities for AI agents.

The system exposes its search and retrieval tools via stdio or HTTP. It utilizes local model files for embeddings and reranking, supporting query expansion across multiple languages.

The project employs abstract syntax tree based chunking to split source code at function and class boundaries. It implements hybrid vector-keyword indexing and metadata-driven context assignment to improve retrieval accuracy, while operating as a background daemon to maintain model residency in memory.

Features

Semantic Search Engines - Functions as a local semantic search engine indexing code and markdown for meaning-based retrieval.

Local Knowledge Base Indexers - Converts local markdown files and codebases into a searchable database for rapid semantic retrieval.

Local Document Indexing - Indexes local documents into vector stores using semantic chunking to support RAG pipelines.

Model Context Protocol - Uses the Model Context Protocol to connect local document search tools to AI agents.

RAG Document Retrieval - Provides retrieval of specific document snippets and line ranges to serve as grounded context for LLM responses.

MCP Servers - Provides a Model Context Protocol server implementation to expose search and retrieval tools to AI agents.

Model Context Protocol Servers - Acts as an MCP server that exposes local search and retrieval tools to large language models.

Code-Aware Chunking - Implements AST-based chunking to ensure source code is split at logical function and class boundaries.

Semantic Search - Implements semantic search that understands query intent across local files using hybrid retrieval and reranking.

Hybrid Search - Combines vector similarity and keyword search with reranking to find highly relevant content in local documents.

Hybrid Vector-Keyword Indexing - Implements a hybrid indexing system combining dense vector embeddings with inverted keyword indices.

Local Knowledge Bases - Converts local markdown files and codebases into a searchable database for private information retrieval.

Search & Information Retrieval - Implements capabilities to query indexed local documents and retrieve relevant information based on user requests.

Syntax-Aware Chunking - Uses abstract syntax tree analysis to split source code at logical boundaries for precise AI context.

AI-Workflow Code Index Builders - Builds structured, searchable indexes of local codebases split by function and class for AI agents.

RAG Pipelines - Provides a RAG pipeline using local embeddings and query expansion to augment AI model outputs.

Local Embedding Generators - Generates vector embeddings and performs reranking locally using on-device model files.

Local Model Integrations - Integrates local model files to perform multi-language embeddings and reranking without external APIs.

In-Memory Model Sessions - Keeps embedding and reranking models loaded in memory as a background process to reduce inference latency.

Retrieval Context Metadata - Attaches descriptive metadata to folders and collections to improve the relevance of retrieved content for language models.

Markdown Indexers - Scans local directories for markdown documents to build a searchable index for fast content retrieval.

Background Daemons - Operates as a long-lived background daemon to maintain model residency in memory.

tobiqmd

Features

Star history