Vector store
A vector store keeps embeddings and returns nearby items according to a similarity function. Useful records also keep the original text, source location, permissions, tenant, and other metadata.
Tiny example
Section titled “Tiny example”A query embedding is compared with stored passage embeddings. The store returns five passage IDs, their similarity scores, and the metadata needed to fetch and cite the source text.
What it is not
Section titled “What it is not”A vector store does not make retrieved content correct, current, or authorized. It also does not replace lexical search for exact identifiers.
FDE note
Section titled “FDE note”Apply tenant and document-permission filters inside retrieval. Filtering results after search can leak metadata and reduce the useful result count.