AI and agents
Vector database
A vector database stores embeddings, the numeric representations of text or images, and finds the stored items closest in meaning to a query.
How it works in practice
Ordinary databases find exact matches: this customer ID, that date range. A vector database finds near matches by meaning, so a search for "late delivery refund" can return a policy paragraph titled "Compensation for delayed orders". It does this by comparing vectors and returning the nearest ones, using indexes built for that kind of search.
You do not always need a separate product. Pinecone, Weaviate and Qdrant are dedicated vector databases, while pgvector adds vector search to Postgres, which is what Supabase uses. For a few hundred thousand passages, keeping vectors next to the rest of your data in Postgres is often simpler to run and to secure, because the same access rules apply to both.
A worked example
A recruitment firm stores embeddings of 20,000 CVs in Postgres with pgvector. A recruiter pastes a job description, and the system returns the 30 closest CVs, filtered by location and availability from ordinary columns in the same table.
Examples in this glossary describe typical builds for companies of 1 to 1,000 people. They are not named client results.















