Vector Memory Service Guide
Store and retrieve embeddings locally for rapid recall without leaking sensitive documents to third parties.
Local persistence
Embeddings are persisted on disk with optional encryption and namespace scoping to separate teams or projects.
- Support for multiple backends such as SQLite, FAISS, and Qdrant adapters (roadmap).
- Automatic cleanup for expired entries via retention policies.
- Index statistics appear in the observability dashboard.
Search operations
`vector.search` supports cosine similarity with adjustable top-K and score thresholds to avoid noisy matches.
- Metadata filters let you scope searches to projects, confidentiality levels, or owners.
- `hybrid` mode combines keyword and vector scoring for precision.
- Requests return reference IDs to hydrate context windows later.
Data governance
Deletion and quarantine workflows ensure regulated data can be removed or isolated when policies change.
- `vector.delete` requires confirmation and emits an audit record.
- `vector.quarantine` (roadmap) will isolate suspect entries for review.
- Namespace access controls integrate with upcoming role-based security.