Capabilities

• Dense search over Milvus float-vector collections.
• Automatic baseline collection and HNSW creation for development.
• search, searchIds, and agent asTool() integration.
• Multiple embeddings per document with logical-result collapsing.
• Scalar metadata stored alongside reserved Anvia fields.
• Translation of eq, gt, lt, and, and or filters.

The ingestion method calls Milvus insert using deterministic physical IDs; it does not issue a provider upsert operation. Plan repeated-ingestion behavior against the configured primary-key policy.

The adapter loads collections but does not manage partitions, replicas, aliases, compaction, client shutdown, or index rebuilds. See the API reference.