A vector database stores embeddings — the numeric fingerprints of meaning — and is optimised for one question: "what's most similar to this?" Give it a query, and it returns the closest matches by meaning rather than by keyword. It's the storage-and-search engine that makes retrieval-augmented generation practical at scale.
When a tool "searches your documents" and pulls the relevant passage even though your words don't appear verbatim, a vector database is usually doing the lookup behind the scenes.
Why it matters at your desk. For an engineer building an AI feature over a company's own content, the vector database is the missing piece between "we have documents" and "the AI can answer from them": you embed the documents once, store the vectors, and query them at question time. For a researcher, it's why tools like Consensus and Perplexity can surface conceptually related sources fast across huge libraries.
What to watch for: a vector database only finds what you put in it, and "similar" is the embedding model's judgement, not ground truth — so retrieval quality caps answer quality. If the right passage isn't retrieved, the model can't use it, no matter how capable it is. Garbage in, confidently-worded garbage out.