AI can inspect platform information and invoke operations through the same authorization rules as the console and API. A model connection, a message channel, and AI on-call are separate settings.
Choose a workflow
| Workflow | Configuration |
|---|---|
| Ask questions through a message channel | Model, platform channel, personal identity binding |
| Analyze events and run scheduled checks | Model and AI on-call |
| Connect an external AI client | MCP and an API key |
| Receive ordinary operational notifications | Channel and personal subscription; no model required |
An Admin manages model credentials and on-call settings. A conversation uses the bound user’s current role. AI on-call uses its own configured role, defaults to Operator, and has no user’s personal account scope.
Set up and verify
Start with model configuration. Connect a message channel, bind your own identity, and ask a read-only question. Confirm the answer refers to the intended deployment.
Enable AI on-call only with a role appropriate to unattended work. Writable roles permit changes within their permissions; notifications are not an approval gate for every action.
Data and failure handling
Using an external model sends relevant conversation and operational context to that provider. Choose credentials and provider settings consistent with the data you allow it to process.
If replies fail, inspect model connectivity and tool support separately from message delivery. A model appearing in discovery does not guarantee suitable tool-calling capability. Check operation results and actual platform state rather than relying only on a natural-language reply.