Rag retrieval evaluation tuning — Atom

A server exposing intelligent tools for enhancing RAG applications with entity extraction, query refinement, and relevance checking capabilities.

A minimal RAG service that exposes a vector index for document retrieval via REST and MCP, allowing querying for relevant document chunks and returning a suggested LLM prompt.

Automated RAG pipeline optimization and serving. It interviews users, builds and evaluates candidate configurations on their data, and registers the best ones as a fleet queryable via MCP.

Enables agents to optimize context-window attention by reordering retrieved documents so critical evidence lands at prompt boundaries, mitigating lost-in-the-middle degradation. It combines BM25 lexical scoring, dense vector similarity, reciprocal rank fusion, knowledge-graph expansion, and recency-decay memory caching, exposed as a zero-dependency MCP stdio server.