pg_pinecone: Under the hood of vector search in Postgres

How do you scale vector search in the Postgres you already run? Jörg Schad, VP of Engineering at Pinecone, walks through how pg_pinecone works: three engines behind one SQL interface, why filtered search breaks HNSW, and how a smaller index keeps memory in check. pg_pinecone is an open-source (Apache 2.0) Postgres extension for vector and full-text search. 📖 Launch post: https://www.pinecone.io/blog/introducing-pg-pinecone/ 💻 GitHub: https://github.com/pinecone-io/pg_pinecone In this video: → Three engines, same SQL: wrap your existing pgvector index, run the native engine, or scale out to Pinecone → Why HNSW post-filters, and how filtering during the search keeps recall high → How RaBitQ quantization and IVF partitioning shrink the index → Why memory matters when your index shares a single node with Postgres Chapters: 0:00 What is pg_pinecone? 0:16 How do you use pg_pinecone? 1:30 What does remote mode add? 2:17 How is it different from HNSW? 3:25 What are the performance gains? 4:55 How does it beat HNSW on latency? 5:42 Why does HNSW need so much memory? 7:00 Why open source it? #pgvector #HNSW #Postgres #VectorSearch #OpenSource