You Don't Need a Vector Database, Postgres Already Has pgvector

pgvector is a Postgres extension that adds a vector column type and cosine similarity operators, enabling semantic search directly in Postgres without a dedicated vector database like Pinecone or Weaviate. The post walks through building a semantic search demo using Prisma Next and its @prisma-next/extension-pgvector pack, which provides type-safe vector operations, a Vector<N> type tied to declared dimensions, and automatic extension migration. The tutorial covers spawning a temporary Prisma Postgres database, declaring a schema with a vector column, inserting embeddings, and querying by cosine similarity — all with compile-time type checking. A comparison table outlines when pgvector is sufficient (up to a few million vectors, apps already on Postgres) versus when a dedicated vector engine is warranted (billion-scale ANN, hybrid search features).