Vector database

A vector database stores embeddings and answers the question "which stored items are closest in meaning to this one" quickly, across millions of entries. A conventional database can technically do this, but it has to compare against every row, which stops being viable early.

It is the storage layer under most retrieval-augmented generation systems and semantic search features. Several established databases now offer vector search as an extension, which is usually the better starting point than adding a separate specialised system to your infrastructure.

Treat it as a derived store, never a source of truth. The authoritative documents live in your normal database or document store; the vectors are an index that can be rebuilt from them, and at some point will need to be.