// project

VelociRAG

· github · Python FAISS ONNX MCP

4-layer retrieval engine. Vector + BM25 + knowledge graph + metadata. No GPU, no API keys.

Open-source 4-layer retrieval engine fusing vector search, BM25 keywords, knowledge graph traversal, and metadata filtering with RRF + cross-encoder reranking.

ONNX-only. No PyTorch, no GPU, no API keys. Sub-200ms warm search. Ships with MCP server, CLI, and Unix socket daemon.

Published on PyPI, AUR, and the official MCP Registry. 550+ tests, ~24K LOC.

Related posts

  1. 001
    11 min

    Searching My Second Brain in Half a Second

    I built a four-layer search engine so I could actually find my own notes. No PyTorch, no GPU, runs anywhere. Here's how it came together.

    ai rag search deep-dive velocirag
← back to projects