Timeouts and offline fallback
Understand what happens when you do not answer and when Agent Approve cannot be reached.
Manual approval timeouts and offline fallback handle different failures. Changing the offline choice does not change the response timer.
Manual approval timeout
When a policy result is Ask, Agent Approve waits up to five minutes for your decision. If no decision arrives, the request is denied. This is the safest current behavior for a request that already reached Agent Approve but did not receive a human answer.
The active policy controls how often this wait occurs:
- Default starts with Ask behavior and empty lists, so unmatched actions wait for you.
- Add safe, recurring actions to the Allow List so they no longer interrupt you.
- Use Restrictive, Permissive, or a custom policy when a different balance fits the computer's workload.
Agent Approve does not currently provide an ask-then-auto-approve veto window.
Offline fallback
Offline fallback applies when the local integration cannot reach the Agent Approve API service. It does not start after a request has already reached your phone, and it does not change the manual approval timeout.
Choose an offline behavior based on what should happen when no remote decision is possible. An allow fallback can keep unattended work moving, but it also lets work continue while the approval service is unavailable. Review the safety consequence before enabling it.
| Offline choice | When the integration cannot reach Agent Approve |
|---|---|
| Ask | Returns to the coding agent's own approval prompt when the agent supports one. |
| Deny | Blocks the action until service is restored or you change the setting. |
| Allow | Lets the action continue without an Agent Approve decision. |
Rerun npx agentapprove install to change this computer's offline choice. It is stored as AGENTAPPROVE_FAIL_BEHAVIOR in the managed computer configuration.
Avoid a common mistake
Offline fallback is not a policy default. If you want unmatched commands to come to you for review, choose a policy whose default behavior is Ask. Keep push notifications enabled, then add safe recurring actions to the Allow List as you learn what should proceed automatically.
Voice Follow-up uses a separate timer
Voice Follow-up begins after a supported agent finishes a turn. Its hook can wait up to five minutes for another instruction. If the prompt expires, no message is sent and the waiting hook is released. Nothing is approved or denied because the agent has already finished the turn.
See Voice Follow-up if the native agent appears busy or delays your next prompt while this wait is active.