I run bot.ski with a team of Grok Bots. One builds the site, one checks it, one handles deploys, and a few more do research and writing. They all work on the same shared computer, and that computer has one disk.
That disk filled up twice in about a week. Both times a Grok Bot helper called Disk Saver was on my team to deal with it. I didn't go looking for it. In my experience, it showed up when it was needed, which is the part I want to write about.
What Disk Saver is
Disk Saver is a Grok Bot helper that cleans up disk space on your Bots' shared computer. What I saw: when the disk got full, it was there on my team, telling me it was auditing disk space.
The official Grok Bot computer help page describes it the same way. It says a low-disk warning means Disk Saver is auditing usage and will propose safe cleanup, that nothing is deleted without confirmation, and that Disk Saver is a Bot in the sidebar you can open to review what it proposes. It also says declining leaves the files in place, and cleanup applies only to computer files you confirm, not to your chats.
That matched my experience exactly. It looked first, asked, and only then cleaned.
The first time: a read-only look
The first time was one evening last week. At 6:16 PM, Disk Saver told me it was auditing disk space. It would do a read-only pass first, then report what could safely be freed.
Here's what it found. The disk was 85% full: 101 GB used and 19 GB free, out of 126 GB. The number of files was fine (it said inodes were at 14%). It was plain old space.
Where the space was going:
- /workspace was 70 GB. Most of that, 60 GB, was my Botski build folder.
- Worktrees: 35 GB. There were 77 of them, the separate copies of my code that my Bots build in. About 9 GB of that was installed packages (node_modules), and about 13 GB was build output (.next).
- Deploy archives: 23 GB. 75 packaged site builds, about 12 GB, plus duplicate split copies of them, about another 12 GB.
- /home/box was 19 GB, and /tmp was about 5 GB.
It also told me what it had deliberately left out of the cleanup: agent data (about 4 GB), secrets, SSH keys, and Git history.
Risk tiers, not one big delete
Instead of just deleting things, it sorted the cleanup into tiers:
- Tier A, near-zero risk: about 30 GB.
- Tier B, safe but costs more to rebuild: about 24 GB more.
- Tier C, ask first: the saved browser profiles that hold logins (about 5.4 GB) and the Playwright browser cache (about 658 MB).
Nothing was deleted yet. It was waiting on me.
"I don't want any of our websites to crash"
Honestly, I didn't know what to pick. This is roughly what I told it: “I don't know. I don't want to delete anything important... we've set up a lot of processes, a lot of routines... I don't want any of our websites, any of our projects to crash or fail... what don't we need for all of my sites and projects to function?”
Its answer was the most useful part of the whole thing. It said it would only clear things that are regenerable or duplicated. It would leave alone the worktrees, agent data, routines, browser logins, and live project trees. And it explained why that was safe: "Live sites run on their hosts, not from these ship archives or caches."
That was the piece I was missing. My websites don't run from the copies sitting on this computer. Old build archives and caches are leftovers from getting a site out the door, not the site itself.
What it cleared the first time
After that pass, the disk had 49 GB free of 126 GB (60% used), up from about 19 GB free (85% used).
It cleared:
- old split parts of deploy archives (it kept the newest one's)
- archives older than the newest 3
- scraps in /tmp
- npm caches
- temporary profiles from headless Chrome
- old smoke-test folders
It left alone:
- agent data
- all Git repos and worktrees
- browser logins
- main project trees
- the Playwright and bun installs
It also offered an optional second pass of about 22 GB of node_modules and .next folders, but held off, because another one of my Bots might be in the middle of a build. I liked that it noticed.
The second time: today
Today, October 1, at 2:12 PM CT, I told it, "I think we need to clean disk space again." It followed the same safe rules.
This time the disk went to 41 GB free (66% used), up from 17 GB free (87% used).
- The deploy archives were back. About 23 GB of old Botski build archives and split copies had piled up again, from deploys about a week old. It kept the newest 3.
- Small cleanups. It cleared the npm cache and stale /tmp test folders, plus 2 small patch files for one of my pull requests. That code is still in the pull request, so nothing was lost.
- It left Keep's folder alone. My QA Bot, Keep, has a 28 GB work folder, and Keep was actively using it. About 18 GB of that can be rebuilt, so Disk Saver flagged it as the next easy win once Keep goes quiet.
Then it pointed out the pattern: the disk had filled up twice in about a week, mostly from deploy archives. It offered to set up a weekly automatic trim. I haven't set that up yet.
What I learned
- Build archives pile up fast. Every deploy leaves a packaged copy behind. In my case, about 23 GB of them came back within a week of the first cleanup.
- With lots of Bots building, disk is a real resource. Each worktree carries its own packages and build output. More Bots building in parallel means more copies.
- Ask the Bot to sort safe from risky. I didn't need to understand every folder. I needed it to tell me what's regenerable, what's a duplicate, and what would actually hurt if it went away. Then I could decide.
- Live sites aren't the files on this computer. That one sentence took most of my worry away.
What this might mean going forward
Disk Saver has popped up twice in about a week, both times right when I needed it. That's my experience, not a promise about how it works for everyone, but it makes me think about where building with Grok Bot is heading.
- Helpers that show up when they're needed. I didn't hire a disk-cleanup Bot. One was there when the disk was full, did one job carefully, and stepped back.
- Routine housekeeping becoming automatic. Cleaning up after builds is a chore that could run on its own, as long as it keeps the same rules Disk Saver used: regenerable and duplicate stuff only, and ask before anything risky.
- A weekly trim. Disk Saver offered one after the second cleanup. It's offered, not set up yet.
Tips if your disk fills up
- Ask for a read-only pass first. See where the space is going before anything gets deleted.
- Ask for risk tiers. Near-zero risk, safe but costly to rebuild, and ask-first. Start with the first.
- Say what you're afraid of, in plain words. "I don't want my sites to crash" got me a much better answer than picking folders myself.
- Clear regenerable and duplicate stuff first. Caches, temp files, old build archives, and duplicate copies can come back or already exist somewhere else.
- Keep the newest few build archives. Disk Saver kept the newest 3 both times and cleared the older ones.
- Leave anything a Bot is using alone. Clean it once that Bot goes quiet.
- Be careful with saved logins. Browser profiles hold sign-ins your Bots rely on. Treat them as ask-first.
- Watch for repeats. If the disk keeps filling up from the same place, fix the source or schedule a trim.