What the emergency stop does
Your settings are preserved
What the emergency stop does
The emergency stop is a single control that halts everything the agent does on its own. Hitting it does two things at once: it pauses autonomy across your whole workspace, so from that moment every action routes to your approval queue instead of running, and it voids anything currently sitting in its soft-commit safety window before it can fire.
Reach for it whenever something looks wrong and you want to freeze the agent immediately rather than fix settings one at a time. It is the brake, not a setting — it takes effect at once and asks no further questions.
Open the Agent settings and find the emergency stop.
Activate it. Autonomy pauses and any in-flight actions in their safety window are cancelled.
Optionally add a short reason, which is recorded with the pause.
Work through what needs you in the approval queue while autonomy is off.
Your settings are preserved
Pausing does not erase or reset anything. Your per-feature toggles and spend limits are left exactly as they were, so a pause is safe to use freely — you are not going to lose a careful configuration by hitting it. While paused, the agent still surfaces work and drafts actions; it simply will not execute any of them without you.
Because the emergency stop is a safety brake, it is intentionally easy to reach. Turning autonomy back on is the more deliberate step, so resuming is what is gated, not stopping.
When you are ready, resume from the same place. Resuming lifts the pause and restores your exact prior per-feature configuration — the settings were never changed, so nothing needs rebuilding. The agent picks back up within the same limits it had before.
One thing resume does not do is un-void the actions that were cancelled when you paused. Anything that was in its safety window at the moment you hit stop stays cancelled; if you still want it, you re-initiate or approve it fresh. Both the pause and the resume are written to your audit trail with who did it and when.