Vector Database

A vector database stores high-dimensional embeddings and finds the nearest ones to a query vector, enabling search by meaning rather than by keyword. It is the retrieval backbone of RAG: a prompt is embedded, the closest passages are fetched, and those passages are added to the model's context to ground its answer.

Because it decides which content is pulled into a prompt at runtime, the vector store is a governance point as much as an infrastructure one. Retrieved passages can carry sensitive data into a request or injected instructions the model will act on, so scoping what a given identity can retrieve - and inspecting what comes back - keeps the retrieval step from becoming an ungoverned data path.