A vector database stores numerical representations of content — embeddings — and finds items that are semantically similar rather than an exact text match. That is what lets “find documents about cancelling a subscription” return the right help article even if it never uses those words.
How similarity search works
Each piece of content becomes a point in high-dimensional space. Related content lands close together, so the database can answer “what is nearest to this query” very quickly using specialized indexes.
Do you need a dedicated one?
For a few thousand items, a vector extension on your existing database is plenty. Dedicated vector databases earn their keep at large scale or when you need advanced filtering and high query throughput.