MLflow AI assets

October 26, 2026

The MLflow AI integration uses the following asset types. The asset types come out of the box with Collibra Platform.

Important The functions of MLflow integration is limited if you do not enable AI Governance. When you enable AI Governance, you can add custom attributes or custom relations, for example, AI Model Version, to the MLflow AI Model Version asset.

MLflow object	Asset type	Description
Registered model (only when it isn't detected as an agent, a registered model produces either the model or agent pair, never both)	AI Base Model	The foundational business concept that defines the model's purpose and governing principles, remaining constant across all technical iterations.
Each READY model version	MLflow AI Model Version	A subtype of AI Model Version that represents AI model versions in MLflow. Only model versions with a `READY` status are ingested.
Registered model detected as an agent	AI Agent	An autonomous system that perceives its environment, makes decisions based on predefined rules or learned patterns, and takes actions to achieve specific goals.
Each READY version of that agent model	AI Agent Version	A specific, immutable implementation of the model logic and parameters that serves as the audited "raw" model version before it is activated.
Training dataset of the version's run, only if the source is a Delta table; created once and shared across versions	Database	A collection of data that is systematically organized or structured to make it easy to create, update, and query the information. Only available if the Ingest training data lineage for AI model versions option is turned on in the integration settings.
Training dataset of the version's run, only if the source is a Delta table; created once and shared across versions	Schema	An asset that contains the location of specific data. It provides all the details that are required for setting up a connection to a database or server. Only available if the Ingest training data lineage for AI model versions option is turned on in the integration settings.
Training dataset of the version's run, only if the source is a Delta table; created once and shared across versions	Table	An implementation of data entities in columns and rows, in a given database system. It is the basic structure of a relational database. Only available if the Ingest training data lineage for AI model versions option is turned on in the integration settings.