Aquanode DocsAuto-pause
Get warned when a GPU sits idle, then have your setup saved and the machine released so it stops billing. Off by default everywhere — here's how to turn it on.
Auto-pause watches GPU utilization on a deployment, warns you when the box has been idle for a while, and then saves your setup and releases the machine so it stops billing — the same save-then-release mechanism manual Pause uses, just firing on a timer instead of a click.
Off by default — here's how to turn it on
Auto-pause is off unless you ask for it. Nothing you deploy is watched by default, on any surface.
To turn it on for a deployment, switch Auto-pause when idle on in the console deploy sheet before deploying, or turn it on afterward from the deployment's Idle tab. Through the API or the aq CLI, set the idle policy explicitly.
It is per deployment, never account-wide, and nothing is ever enabled retroactively on deployments you already have.
How it works
- Switch Auto-pause when idle on when deploying, or turn it on later from the deployment's Idle tab. It is off until you do.
- Set (or keep the defaults for) three thresholds: how low GPU utilization has to be to count as idle, how many minutes of that before you get a warning, and how many minutes before it acts.
- If utilization stays at or under the threshold for the warn period, you get an email — the box is still running and still billing.
- If it stays idle past the act threshold, your setup is saved first, then the machine is released.
The warn threshold always has to be lower than the act threshold — the console won't let you save a configuration where the warning would fire after the box was already paused.
The four states
The Idle tab and the Overview page show one of four states for a deployment with auto-pause configured:
| State | Meaning |
|---|---|
| In use | GPU utilization is above the idle threshold, or someone is present on the box. |
| Idle | Utilization has been at or under the threshold long enough to warn, but not yet long enough to act. |
| Idle past the auto-pause threshold | Past the act threshold — the box will be saved and paused on the next check. |
| We can't read utilization on this box | An explicit unknown, not a guess. |
Unknown is never treated as idle
Presence on a box is read from the on-box agent, which sees SSH logins and browser-terminal sessions. If utilization can't be read, or the agent can't be reached, the deployment shows as unknown rather than being silently counted as idle. The system doesn't invent evidence that nobody is on the box — an unreadable signal never triggers an auto-pause.
What gets saved
When auto-pause acts, it saves the deployment's setup to storage before releasing the machine — the same save mechanism used for a manual Pause. It saves files, not a running process: nothing that was loaded in memory survives, and restoring the save puts your files back on a box rather than picking a job back up.
Getting back to an auto-paused deployment
Resume it the same one-click way you'd resume a deployment you paused yourself. Resume rents a fresh machine, possibly on a different provider, and restores your saved setup onto it — it is never the same physical machine coming back to life.
Changing thresholds
Open a deployment and go to its Idle tab to turn auto-pause on or off, or change the utilization threshold, warn minutes, or act minutes at any time.
Portable Workspaces
A named directory that follows a VM deployment across providers. Attach it, work, sync, detach — restore it onto the next box.
ComfyUI
Deploy ComfyUI, the node-based Stable Diffusion UI, as a template on a GPU instance. Aquanode provisions and starts it for you behind a private, password-protected URL.