pgGraph

and run graph traversal, semantic search, text search, hybrid retrieval, and ranking in one place. If Postgres is already your system of record, we want to make it possible to avoid maintaining separate source-of-truth, vector, and graph databases unless you really need to. pgvector covers the vector side, but we wanted the missing graph component too, so we built pgGraph as an open-source Postgres extension: https://github.com/Evokoa/pgGraph For this HN launch, we built a public Simple English Wikipedia demo as a retrieval test surface: https://polygres.com/wikipedia The demo does not generate answers with an LLM; it is meant to show the retrieval layer itself. I’d love feedback on whether keeping graph and vector retrieval inside Postgres is useful for your workflows, what APIs or query patterns you would expect from something like this, and whether the Wikipedia demo makes the retrieval behavior clear. I’m also curious how people here feel about pgvector in production. We’ve heard mixed feedback, especially around scale, and are trying to understand whether the right path is to build around it, contribute to it, or eventually replace parts of the vector layer ourselves. Happy to answer technical questions, and very open to criticism.