Databricks
Discover registered models and serving endpoints in a Databricks workspace.
Airia discovers the AI assets in a Databricks workspace: registered models (Unity Catalog) and model serving endpoints.
Authentication model
A Databricks personal access token (PAT) plus the workspace URL.
Required permissions
The PAT inherits the permissions of the principal that created it. For full results the principal needs:
| Capability | Privilege on the Databricks side |
|---|---|
| Enumerate registered models and serving endpoints (core discovery) | Read access to the workspace's models and serving endpoints |
| Unity Catalog grant diagnostics | USE_CATALOG on the model's catalog and USE_SCHEMA on its schema |
| Serving endpoint ACL diagnostics | CAN_VIEW or higher on the serving endpoint |
Only the first row affects what Airia discovers. The other two power the diagnostic checks on the connection test, so without them the connection reports a warning while authentication and discovery still succeed normally.
Grant diagnostics report a missing Unity Catalog grant as HTTP 400
If the principal lacks USE_CATALOG and USE_SCHEMA, the grant-diagnostics check returns
"The provider rejected the request as invalid (HTTP 400)".
If you see a 400 on that check while every other check passes, grant the two privileges above.
What the connection test checks
The test runs five checks and reports the worst result. Each one carries its own remediation and a link to the relevant Databricks API documentation.
| Check | Category | What it proves |
|---|---|---|
config | Connectivity | The workspace URL and token are both present |
auth | Auth | Airia reached the workspace and authenticated with the token |
data.agents | Data | How many assets were found — registered models and serving endpoints |
permission.model_grants | Permission | Unity Catalog grants are readable for a registered model |
permission.serving_acl | Permission | The access-control list is readable for a serving endpoint |
Only config and auth are critical. The other three are warning-level, so a connection that
authenticates will still save and still discover assets even when they do not all pass.
Customer-side setup steps
- In Databricks, create a service principal (recommended) or choose an integration user with read access to the models and serving endpoints in scope.
- Generate a personal access token for that principal (User Settings → Developer → Access tokens, or the service-principal token API).
- Copy the workspace URL (for example
https://yourworkspace.cloud.databricks.com). - In Airia, go to Discover → Connect → Databricks and create a new connection with the workspace URL and token.
Form fields the Airia UI asks for
| Field | Required | Notes |
|---|---|---|
| Workspace URL | Yes | https://<workspace>.cloud.databricks.com |
| Access token | Yes | Stored encrypted, masked on edit |
Multiple environments
Multiple Databricks connections per tenant are supported: one per workspace.