A database built to store embeddings and quickly find the content closest in meaning to a query.
A vector database stores the vectors produced by embeddings alongside the original text and answers questions like which ten passages are closest in meaning to this question. That is a different kind of search from the keyword or number lookups in a regular database.
You need one when you build an assistant that answers from your documents, smart product search or recommendations. Some well-known databases, such as PostgreSQL, can add this capability through extensions, so you do not always need to run a separate system. Practical tip: start with the simplest option that fits your data size. With a few hundred documents you rarely need complex infrastructure. Remember to refresh vectors whenever the source content changes, or your assistant will answer from outdated information.