Deploys are not a separate system to learn: pushing to your default branch deploys. Your pushes, the agent’s merged pull requests, a revert — anything that lands on main goes out the same way, built and run on your own box.
The Deployments tab
Every deployment records its status, checks, logs and the commit it shipped. From the tab you can:
- Redeploy the current commit,
- Retry a failed build,
- Roll back in one click — every build tags the previous images, so rollback is starting what is already on the box, not rebuilding history.
When a deploy fails
The part that matters at 3am — in both failure modes, nothing retries in a loop and nothing pages you:
- A failed build never replaces the running app. The old containers keep serving; your users see the previous version, not an error page.
- A deploy that lands and then fails its health checks leaves the new version up — you are alerted, and rollback is one click away.
- Either way, the failure lands in your Inbox with the log a click away, and the agent pauses and reports instead of trying again.
Health of the machine itself
Server-level operations — restart the HTTPS proxy, prune Docker, reboot — live on the Servers page, not per project.