← All updates Databricks Update One-Pager Navy version
databricks Platform · Updated Jun 29, 2026

Govern LLMs in Unity Catalog with Model Services

Centralized governance for LLM endpoints across workspaces.

Beta Open docs ↗release note ↗
What it does

This feature allows users to govern Databricks-hosted LLMs through model services in Unity Catalog, enabling centralized management of LLM endpoints. Model services serve as securable objects that define, share, and govern access to LLMs across multiple workspaces. Users can create and manage these services using the Unity AI Gateway UI, Catalog Explorer, or REST API, and can query them from both inside and outside Databricks.

Why it matters

By centralizing the governance of LLMs, organizations can reduce duplication of effort and streamline access management across workspaces, enhancing productivity and governance. This feature allows for better tracking of usage and costs associated with LLMs, which is crucial for budgeting and resource allocation.

Key capabilities
Centralized Governance
Govern access to LLMs centrally using Unity Catalog privileges.
Ready-to-Use Model Services
Access system-provided model services for Databricks-hosted foundation models.
Cross-Workspace Querying
Query model services from any workspace attached to the same metastore.
Standard Privilege Model
Utilize Unity Catalog's standard privilege model for managing access.
Prerequisites
Account admin access to control feature access from the account console.
Unity Catalog must be enabled for your Databricks workspace.
Recommended use cases
Centralized management of LLM endpoints across multiple teams.Tracking usage and costs of LLM services for budgeting.Simplifying access control for LLMs in large organizations.
Limitations · remember
!Provisioned throughput models and external model providers are not supported.
!Creating and managing model services with Terraform or SQL is not available.
!Discovering model services requires more than just the 'BROWSE' privilege.
!Lineage and global search for model services are not available.
Architecture
Unity Catalog → Model Services → Unity AI Gateway → Databricks-hosted LLMs
Get started
1
Access Account Console
Navigate to the account console to manage Databricks previews.
2
Create Model Service
Use the Unity AI Gateway UI or Catalog Explorer to create a model service.
3
Define Access Privileges
Assign Unity Catalog privileges to govern access to the model service.
4
Query Model Service
Invoke the model service using its fully qualified name.
5
Monitor Usage
Track usage and costs through Unity Catalog system tables.
Key takeaway
Govern LLMs centrally in Unity Catalog to enhance productivity and streamline access management across workspaces.
2026-06-29-govern-llms-in-unity-catalog-with-model-services-beta Generated by the Databricks Updates Agent · browse all updates