VelociRAG
Lightning-fast RAG engine with four-layer retrieval fusion powered by ONNX Runtime for sub-200ms search without PyTorch.
High-performance retrieval-augmented generation engine that combines vector similarity (FAISS), BM25 keyword matching, knowledge graph traversal, and metadata filtering through reciprocal rank fusion with cross-encoder reranking. Achieves sub-200ms cold-start search and ~3ms warm queries using ONNX Runtime, requiring no PyTorch or GPU dependencies. Includes a Unix socket search daemon for persistent warm model and index state, smart header-aware document chunking, and query expansion with acronym and variant handling.
Source
Repository: https://github.com/haseebkhalid1507/velocirag
Maintain VelociRAG?
Let people know it's listed here — add the badge (live metrics, light/dark aware) or a plain link to your README or docs.
[](https://getagentictools.com/mcp/haseebkhalid1507-velocirag?ref=badge) Related mcp
UniFuncs Provides a bridge to the UniFuncs API for web search and web reading capabilities through TypeScript implementation with Express a
UK Legal Research UK legal research covering case law, legislation, Hansard, bills, parliamentary votes, and HMRC guidance.
Jira Integrates with Jira Cloud REST API v3 to enable issue creation, updates, searching with JQL, comment management, workflow transit
Hive Vault Connect AI assistants to Obsidian vaults with on-demand context, search, and git integration.
SeekLink Semantic search for local Markdown vaults via natural-language queries.
Agent Comms Enables cross-harness communication between LLM agents with rooms, direct messages, and presence tracking over TCP.