Hello community,
I built Rivet-KAG, an open-source implementation of KAG-RAG (knowledge AG) for Q&A from vector and graph DBs. This repo is MVP to collect feedback. How would you measure whether graph retrieval adds value over vector retrieval alone?
It queries Milvus vector search and Neo4j graph retrieval in parallel, merges the evidence, and produces source-cited answers. I also included ingestion, evaluation gates, guardrails, cost controls, audit data, and a Docker path, because retrieval quality alone is not enough for an internal knowledge assistant.
I’d welcome feedback, issues, and suggestions. If you find it useful, please mark a 
.
Much appreciated,
Egor