Quick answer
A vector database stores information as embeddings — long lists of numbers that represent meaning — so an AI app can search by concept instead of exact keywords. Ask "how do I get my money back" and it can find a document that says "refund policy" even though no words match, because the two phrases land near each other in the database's numerical space. Vector databases are the piece of infrastructure that makes retrieval-augmented generation (RAG) — an AI answering questions using your own documents — actually work at scale.
A normal database finds exact or close text matches. A vector database finds things that mean something similar, even if the words are completely different — which is exactly what you need when a user's question doesn't use the same phrasing as the document that answers it.
How it actually works
Every piece of text (a paragraph, a support article, a product description) gets converted into an embedding — a list of a few hundred to a few thousand numbers produced by an AI model that represents that text's meaning as a point in space. Texts about similar topics end up as nearby points; unrelated texts end up far apart. A vector database is built specifically to store millions of these points and quickly find the ones nearest to a new query.
- You ask a question; it gets converted into an embedding the same way
- The database finds the stored embeddings closest to your question's embedding
- The matching original text gets pulled back and handed to an AI model as context
- The model writes an answer grounded in that retrieved text, instead of guessing from memory
Why this matters for RAG
Retrieval-augmented generation is the standard technique for making an AI answer questions using a specific company's documents, a knowledge base, or a codebase, instead of only what the model learned during training. None of that works without a fast way to find the RIGHT few documents out of possibly millions — that retrieval step is exactly what a vector database is built for.
Popular options in 2026 include Pinecone and Weaviate (dedicated vector databases), and pgvector (a vector-search extension bolted onto ordinary Postgres) — the latter is popular specifically because teams can add vector search without adopting a whole new database.
Related reading
Bottom line
Vector databases are unglamorous infrastructure doing essential work: turning "search by meaning" from a research idea into something that runs in production at millisecond speed. If an AI product can answer questions about your own private documents, there is almost certainly a vector database quietly doing the retrieval underneath.
