Vector database

A vector database stores data as high-dimensional numeric vectors, typically embeddings produced by a machine learning model, and is built around one core operation: given a query vector, quickly find the stored vectors closest to it by some distance measure, most often cosine similarity or Euclidean distance.

This differs from a conventional database, which is built for exact matches and structured filters. Under the hood, vector databases use approximate nearest-neighbor indexes, such as HNSW graphs, that trade a small amount of accuracy for large gains in search speed at scale, since checking every vector exactly would be too slow for large collections. Many products also support hybrid search, combining vector similarity with traditional keyword or metadata filters.

Vector databases are the retrieval component behind most retrieval-augmented generation systems and other semantic search applications, storing document or product chunks so a large language model or search interface can pull back the most relevant matches for a query. A common pitfall is treating similarity as correctness: a nearest vector is semantically close, not necessarily the right or most authoritative answer, and index quality depends heavily on the embedding model and chunking strategy used upstream.