I built this because I’m tired of the standard RAG stack becoming a distributed system. Usually, if you want GraphRAG to solve multi-hop reasoning, you’re told to add Neo4j or a specialized GraphDB to your stack. Now you have three databases to keep in sync: your relational source of truth, your vector store, and your graph nodes. And also you have to deal with Neo4j and who wants that if you can have Postgres?
postgres-graph-rag is for the Postgres Maximalists. It turns your existing DB into a structured knowledge engine using SQL Recursive CTEs instead of expensive LLM-agent loops.
Features: - No Infra Sprawl: If you have Postgres, you have GraphRAG. - Incremental by Design: Most GraphRAG research is batch-heavy. This handles atomic upserts in real-time. - Forever Schema: Uses a JSONB + Namespace pattern to avoid future migrations.
It’s open source and I’d love to hear your thoughts on the Recursive CTE approach for graph traversal in production environments.