Vector database
A vector database is a database optimised for storing and querying embeddings, so that semantic similarity becomes a query you can run. It is the storage layer underneath retrieval-augmented generation and semantic search, where finding the closest meaning matters more than matching the exact words.
A vector database indexes embeddings — the numerical fingerprints that capture the meaning of a piece of text, image or audio — and lets you ask "what in my data is closest to this?" in milliseconds. That single capability is what makes retrieval-augmented generation work: when a user asks a large language model a question, the vector database finds the most relevant snippets from your own documents and the model answers using that context instead of guessing.
Typical options are dedicated stores like Pinecone, Weaviate, Qdrant and Milvus, or extensions on top of existing databases such as pgvector on Postgres. For most teams that are already on Postgres, pgvector is the boring and correct choice until the scale demands otherwise.
The honest take is that a vector database is another moving part to run, monitor and pay for, and the quality of results depends heavily on the embedding model and chunking strategy you choose around it. It is only worth it when you genuinely need semantic recall over your own data — a regular full-text search index is often perfectly fine, and frequently faster to ship and easier to debug.
