Capabilities

• Dense vector upsert using deterministic physical IDs.
• Namespace-scoped ingestion and retrieval.
• Cosine, Euclidean, or dot-product metric configuration.
• Equality, range, and, and or metadata filters.
• Multiple embeddings collapsed into one logical document result.
• search retrieval through the core VectorStore interface, with retrieveDocuments and createVectorSearchTool integration from @anvia/core/vector-store.

Metadata keys beginning with __anvia_ are reserved for document reconstruction.

The adapter does not determine index dimension, capacity, deletion policy, or tenant authorization. Its automatic create path is a convenience and cannot express a production index's complete dimension and deployment policy; pre-provision instead. See the API reference.