Databricks puts a staging lane between dashboard metrics and Unity Catalog

Databricks has added local metric views to Azure Databricks AI/BI dashboards, creating a middle ground between ad-hoc dashboard SQL and a globally shared semantic object. The new view type keeps fields, measures, joins, filters and parameters inside one dashboard while retaining semantic aggregation behavior and query-time transformations. Databricks’ documentation is dated September 17.

A staging area for semantic logic

The practical change is organizational as much as technical. Teams can define dashboard-specific metrics without first publishing them to Unity Catalog. Databricks explicitly positions the feature for prototyping, small-team analysis, validation before promotion and users who lack write access to a Unity Catalog catalog or schema. A user still needs CAN USE permission on a SQL warehouse.

That creates a clearer lifecycle for metric definitions: experiment locally, test the behavior in a real dashboard, and export the definition when it is stable enough to become a Unity Catalog metric view. Once promoted, Databricks says the metric view becomes available to downstream dashboards, Genie Agents and other tools that support Unity Catalog metric views.

Local views can start from one or more tables or extend an existing Unity Catalog metric view with dashboard-specific fields and measures. The latter needs only read access to the cataloged view. Teams can therefore preserve a shared core definition while adding a local calculation that is not yet appropriate for every consumer.

What remains local—and what does not

The boundary matters. Databricks recommends Unity Catalog metric views when metrics must be shared across dashboards, agents or tools. Local views are scoped to their dashboard until they are exported. They can be edited through the same visual model used for cataloged metric views, or through YAML, and support CLUSTER BY and PARTITION BY settings for materialized layouts. The current limitations include no support for IDENTIFIER parameters or SQL containing GRANT statements.

This also changes the handoff around agent-authored dashboards. Genie Code’s dashboard authoring mode can already search permitted tables, create datasets, build visualizations and organize pages after user approval. Its access and actions follow the user’s permissions. Local metric views give that workflow a semantic destination that does not require every experimental definition to enter the global catalog immediately.

For platform teams, the useful policy is straightforward: allow local views for iteration, require review before export, and reserve Unity Catalog for definitions that other dashboards or agents are expected to reuse. The feature does not eliminate semantic governance; it gives unfinished metric logic an explicit place to live before governance makes it shared infrastructure.