What is a vector database?

A vector database stores embeddings and answers nearest-neighbor queries: given a query vector, return the most similar stored vectors in milliseconds, even across millions of items. It is the retrieval layer under semantic search and RAG, built around indexes that trade a little recall for large speed gains.

The workflow is embed once, query forever: each document becomes a vector on intake, and every search is a vector comparison rather than a keyword match. Embedding is cheap at catalog prices ($0.008 per million tokens for the entry embedding model), so indexing a large archive costs single-digit dollars.

Approximate-nearest-neighbor indexes (HNSW and relatives) are what make scale work; exact search over millions of vectors would be too slow. For most product features, a vector index inside your existing database is enough before a dedicated system is justified.

Related terms