Check AI Partner Status and Choose the Next Step

The AI Partners page helps you see who is working, who can take a new task, and who needs attention. A runtime state is only the current operating signal. You still need to verify task completion in the original task or delivered result.

Before you start

• Note the partner you want to check and the task or channel it was expected to use.
• Do not restart a partner or submit the same important task again only because it is taking longer than expected. Check recent activity and the original task first.

1. Review the overall status

Open AI Partners from the sidebar. The counters at the top group all partners by Working, Idle, Offline, and Error. When the list is long, select one state to focus on that group.

Manage mode shows full partner cards for detailed checks. Switch to Patrol mode when you only need a quicker overview.

2. Understand the four runtime states

• Working: the partner is running a task. Check recent activity before deciding whether to wait or open the original task.
• Idle: the partner is online with no active task. It can take new work. If you expected a task to be running, confirm that it was actually submitted.
• Offline: the partner is offline or disconnected. Leave it alone when that is intentional; otherwise continue with a health check.
• Error: the partner has a runtime problem. Pause repeated submissions, keep the failure details, and start recovery.

3. Read the partner card in context

Confirm the partner name and role first so you do not troubleshoot the wrong one. Then review recent activity, channel chips, model and usage, plus the memory, knowledge base, task, and skill counts.

These details explain the current state. For example, a channel marked Setup is not ready for live messages. A changing task count or token total also does not prove that a task finished.

4. Choose the next step

1. Working: open recent activity or the original workspace and confirm that it is handling the intended task. Review the result against the task criteria when it reaches your approval point.
2. Idle: after checking that nothing was missed, hand over one clearly scoped task with its input, expected result, and stopping point for your approval.
3. Offline: open the management actions and run a health check. Confirm that the partner was supposed to be online before following the recovery prompt. Do not resend the original task yet.
4. Error: keep the latest failure message and open Repair. Complete the checks and quick repair, then retry the smallest useful part of the original task.

5. Verify in the original task

Return to the task, file, or channel you originally needed. Confirm that the result exists, is reviewable, and still stops before actions that require your decision. A move from Offline or Error to Working only proves that the partner is running again; it does not replace this acceptance check.

Done when

You can explain why each important partner is in its current state, every partner that should be working has recovered or has a clear next step, and the key task has been accepted in its original location.

Troubleshooting

• Working for a long time with no result: inspect recent activity and the original task. Run a health check only when activity has stopped; do not submit duplicates.
• Idle even though a task should be running: confirm that the task was submitted, its schedule is enabled when applicable, and required input is available.
• Offline but not currently needed: leave it offline. Do not start a partner only to make the status look healthy.
• Healthy after recovery, but the task still fails: return to Repair, report that the issue remains, and keep the original failure details for the next check.

Related pages

See Connect and Manage Channels for an AI Partner when a channel needs setup. Continue with Check and Recover an AI Partner After a Problem when a partner is offline or in an error state.

Data boundary

Work data stays on your device by default; only channels you connect and model calls go online as the task requires.