Vector Database

A vector database is a storage and retrieval system optimized for high-dimensional numeric vectors – the embeddings produced by models like OpenAI's text-embedding-3, Cohere's embed-v3, or open-source alternatives like nomic-embed and BGE. Where a relational database indexes structured columns, a vector database indexes vectors so you can find the most semantically similar entries to a query vector in milliseconds, even across millions of records. The math is straightforward (cosine similarity, dot product, or Euclidean distance) but the engineering – clustering, sharding, approximate nearest neighbor algorithms – is what makes purpose-built systems orders of magnitude faster than naive search. Pinecone, Weaviate, Qdrant, Milvus, and Chroma are the dedicated vector databases most often deployed in production AI systems. Postgres with pgvector and SQLite with sqlite-vec offer pragmatic alternatives that piggyback on existing operational tooling – and for systems below tens of millions of vectors, those embedded options often win on simplicity. For AI builders, the vector database is the retrieval engine of a RAG pipeline. The quality of your retrieval – and therefore the quality of your model's grounded answers – depends on the embeddings you store, the chunking strategy you use to break documents up, and the metadata filters you apply at query time. Web-connected RAG systems often skip the storage step entirely and call an AI search API at query time; for static corpora, the vector database is the durable layer.