Retrieval
Vector database
A store built to find the nearest [embeddings](/dictionary/embedding) quickly, so semantic search stays fast as your data grows.
Once your documents are embedded you need somewhere to search them. A vector database indexes those numbers so "find the twenty most similar passages" stays fast across millions of records.
You often do not need a separate product. PostgreSQL with the pgvector extension handles this well, which means one database instead of two — less to run, less to keep in sync, and your vectors sit next to the business data they relate to.
Reach for a dedicated vector database when scale genuinely demands it, not by default.