Docs menu

Documentation

Check and Recover an AI Partner After a Problem

Open Health Check from the partner card, review issues and repair scope, then verify recovery with the original task.

Updated Doc version 1.1

When an AI partner stops responding or a task fails, start with that partner’s Health Check and identify what needs attention. After the corrective action, return to the original conversation or task to confirm that work can continue.

Before you start

  • Note the partner, failed task, channel, and latest message.
  • Pause duplicate submissions. A repair may affect other partners sharing the same runtime, so check that they are not running important work.

1. Open Health Check from the partner card

Open AI Partners in the sidebar and find the affected partner. Select Health Check from its card’s action menu and confirm the partner name in the dialog. An Offline or Error card may also offer a direct health-check action.

MotiClaw AI Partners shows partner names, runtime states, and channel controls
Find the affected partner, open Health Check, then verify the original task after the corrective action.

Find the affected partner in AI Partners first. Review the check result and the original task result separately.

2. Read the results and next step

Wait for the check to finish, then review partner setup, runtime, channels, and recent replies. Items needing attention include guidance. When a model-related item appears, check the model settings as prompted.

Not Connected means the corresponding channel has not been connected. Set up only the channels your task needs. Read the guidance for Needs Review or Failed items before deciding whether to change setup or run a repair.

3. Review the impact before repair

When the results offer Diagnose & Repair, review the listed issues and their impact scope first. If the action affects other partners sharing the runtime, confirm that their work can be interrupted before continuing. Keep the results and seek help when the scope needs confirmation or you cannot judge the guidance.

Wait for the repair stages to finish. The dialog attempts a fresh check automatically. If the partner is still recovering or the follow-up check did not finish, wait briefly and select Run Again. When no repair action is available, complete the setup step indicated by the check instead of repeatedly trying unrelated controls.

4. Check again and verify the original task

Review the latest results and confirm that issues affecting your task are resolved. Return to the conversation or task that failed and retry the smallest useful part. Confirm that the partner actually replies, the task finishes, and the result meets your expectations.

If the check passes but the task still fails, keep the new message and failure point, then inspect the relevant channel, model, or input. Pause repairs that repeatedly fail. For urgent work, use another partner you have verified is ready.

Done when

The check issues affecting your task are resolved, the original task completes one small retry in the agreed location, and you have reviewed any output needing your judgment. Resume normal submissions after these conditions are met.

Troubleshooting

  • Repair is missing from the sidebar. Open AI Partners and select Health Check from the affected card’s action menu.
  • The check is healthy, but no reply arrives. Confirm the conversation belongs to this partner, then check the channel, model, and input as prompted.
  • The partner is still recovering after repair. Wait briefly and select Run Again. Keep any remaining failed items before deciding what to do next.
  • Repair affects other partners. Confirm their work can be interrupted. Stop and seek help when you cannot establish the impact scope.

Related pages

Open Run a Daily Check from AI Partners in the Docs navigation for daily inspection, Connect and Manage Channels for an AI Partner for conversations, or Quickstart: Create Your First AI Partner and Verify a Task if you have not created a partner yet.

Data boundary

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