Menu

When your bot's job needs a computer that's switched off

Your Grok Bot’s cloud computer keeps working when your laptop is closed. A computer you’ve registered for your bots to use, like a desktop PC in a shop, doesn’t work that way. When it’s off, asleep, or offline, any job that needs it stops. This week that happened to us four times. This page covers how to tell which computer a job really needs and what your bot should do when that computer isn’t there.

What happened to us

Some of our work runs on a PC in our shop, because that’s where the right apps and the big disk are. A bot that writes our daily shop journal saves each day’s notes there. Our site builds and deploys also go through it.

On September 29 and 30, the PC was offline when the journal job ran, so the job couldn’t save the day’s notes. On the second night, the bot saved them on its cloud computer instead and said where. On October 1, the PC dropped offline in the middle of a run of site updates, and the next deploy had to wait until it came back.

Nothing broke for good. But each time, the work sat still until someone noticed.

On October 2 it happened from both directions. Our bots’ cloud computer was rolled back to an earlier saved state, and files saved shortly before that were gone, including a handoff packet and its notes that were halfway through being passed between bots. Then, at about 4:50 PM CT, the shop PC dropped offline again.

Two kinds of computer

  • The cloud computer. Every Grok Bot account has one persistent cloud computer that all your bots share. The docs say closing the app or your laptop doesn’t stop cloud work. More on it in The Botski virtual machine.
  • Your own computers. The Mac or Windows computer in front of you, or another one you’ve registered, is separate. A bot only runs commands there if you’ve allowed it, and each registered computer has its own setting (Approvals, security, and privacy). If that computer is off or offline, your bot can’t reach it. That last part is what we saw, not something the docs spell out.

The cloud computer can roll back

Being always on doesn’t mean nothing is ever lost. The docs say a reset rebuilds the computer from your last saved snapshot, and very recent changes may be lost. They also say the conversation should still hold the final result or a clear link to it, with the shared workspace used for in-between material (Files and results). So:

  • Put anything you can’t lose somewhere durable, such as the chat itself, a connected service, or a code repository, and not only in a file on the bot’s computer.
  • After a rollback, check handoffs. Ask each bot what it was in the middle of and whether its files are still there before it carries on.

Which jobs really need your computer?

Before you count on your own PC, ask whether the job needs it:

  • It needs your PC if it uses an app that’s only installed there, files that only live there, or hardware plugged into it.
  • It can run in the cloud if it’s research, writing, file processing, website checks, or anything a connector can do. Those are safer on the cloud computer, which doesn’t sleep.

When a job only needs your PC for its last step, like copying a finished file over, do the rest in the cloud and leave that step until the PC is back.

When the computer is off

Ask your bot to do three things:

  1. Stage what it can. Do the work on the cloud computer and say where it saved it. Our journal bot saved the day’s notes there and said so.
  2. Name the blocked step. For example: “Copying today’s notes to the shop PC is waiting because it’s offline.” Then you know exactly what to do when it’s back.
  3. Don’t fake a finish. “Done” should mean done. A job that’s waiting on a computer is blocked, and the bot should say so.

Routines that depend on your PC

A routine that runs while you’re away can fail without anyone seeing it.

  • Make failures visible. Tell the routine to message you when it can’t reach the computer, instead of staying quiet.
  • Check recent runs. The app keeps each routine’s recent runs, the last 20, with their successes and failures (Skills and routines). Look for the last good run, not just the last run.
  • Use the official checklist. The Troubleshooting page’s “A routine did not run” list includes whether the computer can reach the source system.
  • Pause it if it keeps failing. A job that fails every run still uses usage each time.

Checklist

  1. Know which computer each job needs.
  2. Keep anything that doesn’t need your PC on the cloud computer.
  3. When your PC is off, stage the work in the cloud and name the blocked step.
  4. Never call a blocked job done.
  5. Have routines report when they fail.
  6. Check the last good run, not just the last run.
  7. Keep anything you can’t lose out of only-on-the-computer files, and recheck handoffs after a rollback.

The short version

Your cloud computer is always on, and your own PC isn’t. Run what you can in the cloud, and when a job needs a PC that’s off, have your bot save its progress, say what’s blocked, and tell you.

See also

Sources

Fetched 2026-10-01 CT.

Cookie preferences

bot.ski uses necessary cookies to run. With your OK we also use analytics to improve the site.

The Google tag loads only after you choose Accept all or Allow analytics, and only when analytics is enabled for this site. An explicit Allow choice overrides browser Global Privacy Control or Do Not Track signals.

You have not chosen yet. Optional analytics stays off until you Accept all or Allow analytics.

Details are in Cookies and preferences and the Privacy Policy.