Databricks launches ai_decide beta for governed AI decisions

Databricks has launched ai_decide, a beta AI Function that the company says can turn governed text or structured data into decisions through SQL and a REST API.

The release makes the decision format available in two Databricks workflows: SQL queries and REST API requests. It follows the general availability of Genie One MCP in Unity Gateway.

Databricks says availability is limited to some regions. Workspace administrators can control access from the Previews page, and the function is not available on Databricks SQL Classic, according to the Databricks SQL documentation.

Each call sends one shared input, called state, to one or more questions. A question can return a probability for a true-or-false judgment, select from named choices, or assign a score on an ordered scale. The SQL function accepts STRING or VARIANT input; the REST request accepts a string, JSON object or array.

For API use, Databricks documents POST /api/2.0/ai-functions/ai-decide with the ai-functions scope. The Databricks REST reference lists version 1.0 and returns answers keyed to the question identifiers supplied in the request.

On data handling, Databricks says document data is processed within its security perimeter and that it does not store parameters passed into AI function calls. It does retain run metadata such as the Databricks Runtime version.

The function requires Databricks Runtime 15.4 LTS or newer, and Databricks recommends Runtime 18.2 or newer.