Skip to content
d_devboxby 508.dev/ DocumentationDashboard ↗
Browse documentation
DOCUMENTATION / Backups & recovery

Persistence, backups and recovery

A devbox keeps work on a remote disk, but a persistent disk and a tested backup solve different problems. Decide how you will recover before relying on the host for irreplaceable work.

What survives

EventFilesRunning interactive sessions
Laptop disconnectStay on the hostHost tmux sessions can keep running
Host rebootStay if the persistent disk is intacttmux and agent processes stop
Host or disk lossDepend on backups or a recoverable provider volumeLost
Platform release rollbackUser files remain outside the platform releaseService recovery does not reconstruct every agent session

Services configured with your normal service manager can restart after a reboot. An interactive tmux launch is not automatically converted into a reboot-persistent service. Vendor-supported conversation continuation is separate from restoring a process.

Back up the right data

Protect repositories with remote Git history and back up unpushed work, application development data and private configuration through your own approved backup mechanism. Treat vendor authentication backups as sensitive; do not copy them into a public archive or shared image.

Record how to recover the host account, persistent volumes and SSH identity. A backup that exists only on the same disk does not survive losing that disk.

Test restoration safely

Restore into an isolated target first. Check representative files and development databases, not just whether the backup command exited successfully. Keep the original host and disk intact while investigating an outage.

The hosted owner pilot has a verified local control-plane database restore. Offsite backup remains pending; this is not a broad production durability claim for users' machines.

Reconnect after an outage

TERMINAL
devbox connection list --json
devbox doctor
devbox workspace refresh --computer home

If the host identity changed, verify the new fingerprint independently before updating trust. A stale heartbeat means unknown status, not a completed agent task. Check the actual remote processes and workspace contents.

If automatic workspace matching is incomplete, select a currently observed workspace explicitly. Never reset or delete a tree just to make it match a stale laptop checkout.

Development machines. Your code. Your control.
Private pilot · Feature availability is called out in each guide.