A database specialized in storing and indexing vector embeddings to power semantic, similarity-based search retrieval.
A vector database stores and indexes vector representations so an application can search for similar items. The stored vectors are commonly associated with content identifiers and metadata. This gives the application a way to retrieve candidate passages or items without requiring the query to use exactly the same words.
The database is one component of a larger system. An embedding model creates the representations, retrieval finds candidates, and a separate application may display them or pass them to a language model. Storing documents in a vector index does not itself produce a reliable answer.
Appropriate filters and data management still matter. A conceptually similar document may belong to the wrong product, region or permission group. Refreshing the source collection, retaining useful metadata and enforcing access rules are operational responsibilities, not automatic consequences of using vectors.
Takeaway: Evaluate the whole retrieval workflow rather than choosing a system only because it uses a vector database. Confirm that the right documents can be found and inappropriate ones excluded.
Example
A service company indexes equipment instructions alongside model numbers and revision dates. Similarity search can propose relevant passages, while model and version information helps narrow the candidates. Without those checks, the nearest-looking passage could come from an incompatible machine.
What to check
Test retrieval separately from generated answers so errors can be located. Keep a process for source updates and removals, and verify access controls with realistic user roles. Similarity retrieval is not a substitute for authorization.
For more detail, see Vector search in Azure AI Search.