pgvector is an open-source extension that adds vector storage and similarity search to PostgreSQL, the popular relational database. With it installed, a Postgres database gains a vector data type and the ability to perform nearest-neighbour queries, so vectors can live in ordinary tables alongside regular columns.
Its great appeal is that it brings vector search to a database many teams already run. Instead of adopting and operating a separate specialised system, you can store embeddings next to your existing relational data, join across them, filter with full SQL, and rely on Postgres’s mature transactions, backups, and tooling. For teams with an existing Postgres stack and moderate vector workloads, this is often the simplest path to adding semantic search.
The trade-off comes at scale. A general-purpose relational database with a vector extension may not match a purpose-built vector database in raw search speed, recall under heavy filtering, or billion-scale performance. Companion projects and tuning extend how far it can go, but pgvector is generally best when vector search is one feature among many rather than the core, performance-critical workload.