Vector database

Once you turn documents into embeddings, you need somewhere to keep them and a fast way to search them by similarity. A vector database does both. You store each embedding with its original text and metadata, then ask for the nearest matches to a query embedding.

Options range from dedicated vector databases to vector extensions for general-purpose databases such as Postgres. For small collections, even a simple in-memory search can be enough.

A vector database is a means, not a goal. Many agent tasks work well with plain keyword search, file search tools, or direct API queries. Reach for vector search when you have a large body of text and users ask questions in their own words.