Release 9.2.42.1¶

This release adds discovery for Databricks Agent Bricks Information Extraction agents.

AI Assets Collector¶

Discovery for Agent Bricks Information Extraction agents

Databricks Agent Bricks Information Extraction agents are now discovered as their own asset type, alongside the existing Agent Bricks Custom LLM support.
Each agent is captured with its console URL, backing model and serving endpoint, and the Unity Catalog volumes and tables it extracts from, so it is governed with the same depth as other agent types.

Enhanced custom-agent trace ingestion

Custom-agent traces from customer-managed object storage (Langfuse OTLP) now use a standardized schema for consistent and reliable trace ingestion.
Trace processing now provides more consistent extraction of token usage, model, and provider information across different trace formats.
Improved user and agent identification enables more accurate association of traces with the corresponding identities and AI assets.

Information Extraction discovery requires a broader-access credential

Discovering Databricks Agent Bricks Information Extraction agents currently requires a credential with broad access: a personal access token issued with all API scopes, or a service principal granted Admin access in the workspace, or Can Manage permission on each Information Extraction agent.
The standard, narrower asset permissions used for other agent types are not sufficient for this agent type alone. See Personal access token and Service principal (OAuth M2M) for details. A more scoped permission model for this agent type is planned.