pgvector is an open-source extension for PostgreSQL that adds vector storage and similarity search directly to the database. Its tagline is “open-source vector similarity search for Postgres,” and its central appeal is letting teams “store your vectors with the rest of your data” rather than running a separate, dedicated vector database alongside their existing relational one.
The extension introduces a vector data type (plus compressed and binary variants) and the distance operators needed for similarity search, including L2 distance, inner product, and cosine distance. For speed at scale it supports approximate nearest-neighbor indexes: HNSW, which gives better query performance at the cost of slower builds and more memory, and IVFFlat, which builds faster and uses less memory. Crucially, because the vectors live inside PostgreSQL, they inherit the database’s transactions, backups, and access controls, and queries can mix vector similarity with ordinary SQL filters on the same rows.
Why business readers should care: pgvector represents a recurring pattern in the AI tooling economy, where capabilities first sold as standalone products get absorbed into infrastructure organizations already run. For many teams, adding vector search to the PostgreSQL they already operate is simpler and cheaper than adopting a new managed service, and it keeps embeddings under the same governance as the rest of their data.