Databricks¶
Agent Management discovers supervisor agents, knowledge assistants, agent serving endpoints, and agent Apps in a Databricks workspace.
Connection and discovery¶
Use a Databricks model deployment connection with the workspace URL and supported Databricks authentication, such as a personal access token or OAuth2 credentials. The identity needs access to the workspace resources and, for monitoring, their MLflow experiments and traces.
Discovery includes serving endpoints whose task starts with agent/ and Apps whose name starts with agent. Dedicated supervisor and knowledge-assistant endpoints are represented by their corresponding agent records.
Supported capabilities¶
Inventory: records discovered agents and the descriptive metadata returned by Databricks.
Interaction logs and traces: collects MLflow traces for compatible agents associated with a resolved experiment.
Tokens and cost: derives model usage from trace data. Reported DBU usage can be priced with the platform’s configured currency per DBU.
Uptime tests and agent reviews: supported through agent invocation.
Limitations¶
Generic Agent Bricks records do not support log collection. Supervisor agents, knowledge assistants, custom serving endpoints, and Apps require a resolved MLflow experiment. Apps also require a resolved MLflow model. Missing or ambiguous experiment or model associations leave log collection disabled.
Active-user and conversation usage metrics, separate platform-metric fetching, and platform enable or disable actions are unavailable through this connector.