Vector Store & Embedding Operations

Operational management of the vector index underpinning retrieval — building it, keeping it current as documents change, and handling the re-embedding required when the embedding model changes.

Any retrieval system over a corpus that changes.

Small static corpora, where scanning is simpler than indexing.

Index access must respect document permissions. An index built without access metadata cannot filter at retrieval, and the only remaining option is filtering after the model has already seen the content.

Index drift. Documents change, the index does not, and retrieval returns confidently stale content that the source system has already superseded.

An evidence repository where a re-uploaded corrected bank statement must displace the superseded one in the index, not sit alongside it.