Vector database

A database or database feature optimized for storing vectors and retrieving nearby vectors under a chosen distance function, often with approximate nearest-neighbor indexes.

A vector store associates an embedding with an identifier and often metadata or a source reference. At query time it compares a query vector with indexed vectors and returns candidates under cosine similarity, dot product, Euclidean distance, or another configured metric. Similarity reflects the embedding model and metric, not a universal measure of meaning. A retrieval-augmented generation (RAG) system can use a dedicated vector database, a relational extension, a search engine, or no vector retrieval at all.

Builder example

Search quality depends on the embedding model, representation unit, index settings, filters, query distribution, and reranking, as well as database latency. Store source identity, version, date, access scope, and content location so retrieval can enforce policy and reopen evidence.

You ask about current pricing. The vector database returns last year's pricing document because the language is almost identical.

Add metadata filters like date or version, combine vector search with keyword search, and test retrieval with real questions before trusting it.

Common confusion: "Closest" means most mathematically similar embedding, which is not always the most useful or correct answer. Two document chunks can be semantically close while saying contradictory things.