Vector Memory Schemas
Reference for embedding upsert, search, and deletion payloads including metadata and retention tags.
Upsert payloads
`VectorUpsertRequest` includes document IDs, vector arrays, metadata, and retention policies. Responses confirm inserted IDs and index stats.
- `dimension` validation protects against malformed vectors.
- `metadata` supports nested JSON for flexible tagging.
- `compliance_hold` flag prevents automatic deletion until cleared.
Search payloads
`VectorSearchRequest` allows specifying namespaces, score thresholds, and re-ranking hints.
- `include_vectors` toggles returning raw embeddings when debugging.
- `similarity_metric` currently supports cosine with dot-product planned.
- `trace_id` helps correlate searches with orchestrator runs.