New feature
GA
Model services
Model services is a Databricks ai / ml capability, introduced June 2026.
A Unity Catalog securable that represents a governed LLM endpoint, so an endpoint is defined once and shared across workspaces through Unity Catalog privileges instead of being duplicated per workspace.
- It turns a model endpoint into a catalog object, which means GRANT is now a reasonable way to hand someone access to an LLM.
- Databricks pre-populates them for you in a schema literally called system.ai, so the hosted foundation models show up as governable objects without anyone creating anything.
- Within two months of launch they were pulled into attribute-based access control, where a policy can grant privileges to any model service whose governed tags match a condition.
Limitations: Provisioned throughput models are not supported as destinations, model services cannot be created or managed with SQL, they cannot be discovered with only the BROWSE privilege, and global search does not cover them.
Open in REbricked →- Category
- AI / ML
- Also known as
- model service, model API, model APIs, system.ai model services
- Verified
- 2026-08-20
Sources
- Official Official Databricks / Microsoft docs
- Official Databricks docs: Govern model APIs (model services)
- Official Databricks release notes: Unity AI Gateway is now generally available (August 4, 2026)
- Official Databricks docs: model services named among the capabilities made generally available by the Unity AI Gateway GA
- Official Databricks docs: Unity AI Gateway is generally available, with service policies the beta capability enabled separately