Monitor and support a Computer Agent
Keep Computer Agents reliable, useful, recoverable, and appropriately supervised.
Computer Agents work across external systems, so reliable operation requires more than checking whether the Agent can reply in chat.
Verify the capability that matters
Use the Computer settings to review overall status and health. Then verify the specific capability involved:
- inspect channel status for messaging problems
- verify the skill after changing its setup or credentials
- review Task run history for repeating work
- inspect the linked work conversation for tool activity and results
- test a browser workflow after a website changes
- confirm files, records, or messages in the destination system
Troubleshoot with evidence
Start with recent activity and service logs. Use structured diagnostics when the cause is not clear. Avoid repeatedly restarting or changing settings without understanding the failure.
If GetVoiceBot offers a repair for a detected issue, review what it will do and confirm it intentionally. Verify health and the affected capability afterward.
Back up before risky changes
Create a backup before major upgrades, restores, or destructive lifecycle work. A restore can temporarily stop the Computer while its saved state is replaced.
Keep alerts useful
A routine check should stay quiet when nothing meaningful changed. Ask the Agent to alert people only when:
- a real issue remains unresolved
- a deadline or customer commitment is at risk
- user approval is required
- the Agent found a specific, measurable improvement worth acting on
Transient tool errors that recover during the same run should not become customer-facing alerts.
Review access over time
Remove unused skills, channels, credentials, and browser sessions. Revisit the Agent's instructions when responsibilities change, and keep approval requirements explicit for destructive, financial, or externally visible actions.