Skip to main content

Wonderfence

WonderFence provides real-time guardrails for AI-generated content. APIs for evaluating and moderating AI-generated content and interactions to protect against harmful outputs and prompt attacks.

GETList policies

Lists the WonderFence enforcement policies available to the project, the same catalog the Policies page shows, including vector (custom-model) policies.

GETGet policy-to-application assignments

Maps each policy id to the ids of the applications it is enabled for. A policy enabled for no application is absent or maps to an empty list. To get one application's policies, collect the policy ids whose list contains its appId.

PUTSet the policies an application has enabled

Replaces the whole set of policies enabled for one application, as the Apply Policies page does. To enable or disable a single policy, read the application's current policies from `GET /wonderfence/application-policies`, add or remove the policy id, and send the full list. An empty `policyIds` disables every policy for the application. Other applications are never affected.

POSTWonderFence evaluate message

Evaluate messages with application-specific policy enforcement. Every request must identify the application with `app_id`; only policies assigned to that application will be applied.

POSTEvaluate a Databricks Unity AI Gateway external-policy event

External policy endpoint for the Databricks Unity AI Gateway. Configure it on a Unity Catalog HTTP connection and Databricks calls it around every governed model request, sending an `{event, config}` envelope and acting on the `ALLOW`/`DENY` verdict it gets back.