Velocirag

VelociRAG is a fast, lightweight, multi-layer retrieval-augmented generation (RAG) engine for AI agents and developer tools. It combines vector similarity, BM25 keyword matching, knowledge graph traversal, and metadata filtering, all fused through reciprocal rank fusion and reranked by ONNX-based TinyBERT. Designed to run locally without GPU or PyTorch, VelociRAG is easy to integrate with Model Context Protocol (MCP) agent infrastructure, offering a Python API, command-line interface, and daemon service for real-time, sub-200ms search over structured and unstructured knowledge bases. Ideal for AI developers and agent system builders who need high-performance, transparent retrieval on commodity hardware.