A regular database searches for exact matches: "Find records where name = 'Smith'." A vector database searches for similarity: "Find documents that are thematically similar to this question."
How does it work? Each text (sentence, paragraph, document) is converted into a vector – a list of numbers representing its "meaning." Similar texts have similar vectors.
Vector databases are the backbone of RAG systems. When a customer support agent searches through your 500 internal documents for relevant information, it does so via a vector database – finding thematically relevant sections even when the question doesn't contain the exact words from the documents.
Popular vector databases: Pinecone, Weaviate, Chroma, pgvector (as a PostgreSQL extension).
For most enterprise AI projects, the vector database operates "in the background" – you don't see it as a client, but it's part of every reliable RAG solution.