Restart a box without destroying it.
The escape hatch from ERROR — before this endpoint, a box with a stale heartbeat (CPU spike / OOM / agent crash) could only be destroyed by the user, which loses the EBS volume and forces a fresh setup. Now the user can:
systemctl restart bux-tg. ~5s.
Fixes most cases (agent process crash, wedged long-poll).Allowed when status is READY or ERROR. PROVISIONING / RESIZING / AWAITING_OAUTH / DESTROYED → 409 (no instance to restart, or a long-running operation is already in flight).
Returns the box view; status is unchanged here. The next bux_health tick + box-agent heartbeat is what authoritatively flips ERROR → READY when the agent comes back up. The FE polls /me to watch.
Documentation Index
Fetch the complete documentation index at: https://docs.browser-use.com/llms.txt
Use this file to discover all available pages before exploring further.
How aggressive a /me/restart should be.
service — fastest, fixes ~95% of cases (agent process crash, OOM,
wedged TG long-poll). SSM systemctl restart bux-tg. ~5s.
reboot — fallback for kernel-hung / SSM-unreachable boxes. AWS
rebootInstances. ~30-60s. IP + EBS preserved.
service, reboot Successful Response
small, medium, large User-visible spec sheet for one size tier. Returned alongside BoxView so the UI can render the tooltip ("2 vCPU · 4 GB · 20 GB") without a second round trip + so the picker on /bux can show all three rows without us hardcoding prices in the frontend.
provisioning, awaiting_oauth, ready, resizing, error, halted, destroyed