Vector Database

A Vector Database is a database built to store embedding vectors and retrieve the ones nearest to a query vector, making similarity the primary access pattern instead of exact matching. A conventional index answers "where does this value equal that value"; a vector index answers "what is closest in meaning", which requires different data structures. These systems build approximate nearest-neighbor indexes such as HNSW or IVF, support metadata filters so results can be narrowed by tenant, date, or document type, and handle updates and deletions without rebuilding the whole index. They are the storage layer under RAG, semantic search, recommendation, and deduplication. Options split between dedicated services like Pinecone, Weaviate, Qdrant, and Milvus, and extensions that add vectors to a database a team already runs, notably pgvector for PostgreSQL. The practical pitfall is reaching for a separate system too early. For modest collections, pgvector keeps vectors beside the relational data and removes an entire service to operate and keep in sync. Recall also depends on index parameters, so an untuned index can silently return worse neighbors, and the embedding model must stay fixed because changing it invalidates every stored vector.