Introducing pg_pinecone: Open-Source Vector Search Inside Postgres

If your data lives in Postgres, vector search gets harder to run well as it grows. Today we're releasing pg_pinecone, an open-source (Apache 2.0) Postgres extension based on the search engine Pinecone uses at scale. It works on the vector columns you already have and runs inside your database, so your data never leaves it. 📖 Launch post: https://www.pinecone.io/blog/introducing-pg-pinecone/ 💻 GitHub: https://github.com/pinecone-io/pg_pinecone 🐳 Docker Hub: https://hub.docker.com/r/pineconeteam/pg_pinecone What you get: → A much smaller index: 136 MB for 1M 768-dimension vectors, vs 2.0 GB for halfvec and 4.1 GB for full-size vectors on HNSW → Accurate filtered search: filters are applied first, so a narrow filter, like one tenant out of a hundred, still returns a full result list → Faster searches and builds: 1.27 ms at 0.95 recall, and a 45s index build → The same SQL you already write: no Pinecone account, API key, or network needed for the native engine → Room to grow: when you outgrow one server, an optional API key moves search to Pinecone and adds embeddings, full-text search, and reranking It's our first release, and we want to hear what works and what's missing. Try it on a table you already have and let us know in the comments. Chapters: 0:00 Vector search in Postgres today 0:24 Introducing pg_pinecone 0:43 Install and use 1:20 Room to grow 1:31 Try it today #pgvector #HNSW #Postgres #VectorSearch #OpenSource