Capabilities

• Dense vector ingestion using Chroma upsert.
• Query search, ID-only search, and asTool() integration.
• Logical document reconstruction from stored text and metadata.
• Multiple embeddings per logical document; results collapse to one best logical result.
• Equality, greater-than, less-than, and, and or filter translation.
• Collection creation with custom metadata and configuration.

Scores are normalized as 1 - distance, so compare rankings within the same collection rather than assuming a universal threshold across backends.

The adapter does not generate embeddings, provision authentication, manage collection retention, or close an injected client. See Embeddings and the API reference.