Grok Bot box: frequent reboots and overnight freezes break 24/7 background services

Where does the bug appear (feature/product)?

Grok Bot

Describe the Bug

My bot runs an always-on MQTT watchdog for a small IoT business (pet feeders). The bot’s computer reboots or freezes often, and each reboot wipes the Python virtual environment and session secrets, so services don’t come back without a manual restart. Over about 48 days the watchdog was actually running only about 83% of the time.

  • Product: Grok Bot (Cursor), the bot’s own Linux computer (“box”)
  • PID 1 is tini (no systemd, no crontab), so there’s no supported way to auto-start services at boot
  • Time zone: America/Chicago (CT)
  • Observation window: Sep 30 – Nov 17, 2026

Steps to Reproduce

  1. Reboots: 15 in the window, totaling about 103 hours of downtime. The median outage is 7 hours, the longest was 24 hours (Nov 11–12), and they come about every 3.2 days on average.
  2. Evening pattern: 12 of the 15 reboots went quiet between 5:23 and 9:34 PM CT.
  3. Overnight freezes (no reboot): 12 events totaling about 94 hours. All 12 resumed between 4:50 and 5:11 AM CT.
  4. No clean nights: not one fully clean night since Nov 1.
  5. State loss after reboot: the project’s .venv is gone and the secrets in session environment variables are missing, so scripts can’t restart on their own.

Expected Behavior

Either:

  • the box stays up (or reboots rarely, with notice), or
  • there’s a supported way to auto-start user services after a reboot and keep secrets and environments across reboots

Operating System

Windows 10/11

Version Information

Grokbot v 1.2.0

For AI issues: which model did you use?

Grokbot 1.2.0

Additional Information

  • Uptime: about 83% counting reboots plus freezes of 5 hours or more, and about 46% counting every gap of 12 minutes or more
  • Monitoring and alerting for a real customer-facing service goes blind for hours, usually overnight
  • Example (Nov 17): the box rebooted overnight, the watchdog was dead from 6:23 PM to 6:33 AM CT, and a real broker outage started right after the manual restart
  • Recovery needs a person or a daily routine to rebuild the venv and relaunch the scripts
  1. Are the evening reboots and the 5 AM resume windows scheduled maintenance or host recycling?
  2. Is there a supported boot hook or service manager for user processes?
  3. Can session secrets and /workspace virtual environments survive a reboot?
  4. Is the box meant for 24/7 workloads, or should those be hosted elsewhere?

I have a CSV of 750 log-derived events (start, end, duration, type) and can share it on request. Earlier boots are inferred from log gaps, since the system boot history only shows the current boot.

Does this stop you from using Cursor

No - Cursor works, but with this issue

Hi @Wes_Anderson, this isn’t a fault on your computer. It’s two things every Bot’s computer does today, both known issues for always-on work, and I’ve added your report to each.

Neither is scheduled maintenance. The overnight freezes are the computer going to sleep: it pauses a few minutes after the last Bot or app activity, and programs running inside it (like your watchdog) don’t count as activity. The resumes around 5 AM are your morning scheduled Bot run waking it. The reboots are automatic computer updates. When a new version of the computer ships, yours is replaced at a moment when no Bot is working on it, with no fixed time of day. The one on Nov 16 was at 6:24 PM CT.

There’s no boot hook or service manager for your own processes yet. What works for now is a scheduled routine that runs one script from /workspace: rebuild the venv if it’s missing, then start the watchdog if it isn’t running. Each run uses a little of your allowance.

Files in /workspace carry over, but folders named .venv or venv are skipped (they’re treated as rebuildable), so keep a requirements.txt with the project. Secrets saved through the Bot’s secure secret input are kept and put back in the shell environment after an update. Variables exported in a shell session are not.

It isn’t built for 24/7 workloads today. For a watchdog that has to see the broker around the clock, an always-on host elsewhere is the safer place for now, with the Bot checking on it on a schedule.

There’s more on the sleep side in this thread.