A conventional database is built to find exact matches: a customer ID, a date, a keyword. A vector database is built for a different question: given this new piece of text, which stored pieces of content are most similar in meaning? It stores embeddings and is optimised to search millions of them for the closest matches in milliseconds.
This is the storage layer that makes retrieval-augmented generation practical at scale. Without it, comparing a new query against every document in a large knowledge base one by one would be far too slow to use in a live system.
Choosing and configuring a vector database is an infrastructure decision with real governance implications: what’s indexed, who can query which slice of it, and how outdated content gets removed all directly affect what an AI system built on top of it is able to say, and to whom.