Gabba Works

Gabba Builds · 11 November 2026

The day my OpenBot VM needed more room

Conceptual AI illustration: a builder slides a larger desk panel toward a small white robot working on brass gears.

As I added agents and worked through the OpenBot setup, a low-memory warning appeared in OpenBot. On 9 November we raised the virtual machine's memory from 8 GiB to 12 GiB. After a clean shutdown and restart, the guest reported about 11.66 GiB. We haven't claimed the warning is gone for good, and we haven't measured a productivity change from this step.

That sounds like a quick settings change. In practice it meant thinking about the host computer, the guest running OpenBot, and who could bring the system back if it stopped.

Why wasn't this just a settings change?

The warning showed up inside OpenBot itself. I'm paraphrasing it rather than quoting it, because I don't have the exact wording on record. I don't want to dress it up as more precise than it was. It was a signal that the guest was short of room as the team grew, and that we needed to plan a change rather than keep pushing.

Adding memory looks like picking a number, and the number is the easy part. A virtual machine doesn't apply a change like that while it's running in a way you can take for granted. You have to know which machine you're changing, whether the host has headroom, and how you'll start the guest again from outside if it doesn't come back cleanly.

How did we make the change?

The team checked the host had room to spare, saved the work, shut the VM down cleanly, changed the memory allocation and arranged a restart. I'm leaving the host's own figures out. They were a snapshot from that moment and they won't mean much by the time you read this.

I care about that order. Save first. Power the guest off fully before a memory-only change. Have a person, or a supported route, that can restart it from outside the guest. Keep a rollback plan if the new allocation isn't what you expected.

After the restart, the guest read back 12,229,032 kB, which works out to about 11.66 GiB. That's a little under the 12 GiB we allocated. I'd rather show the number the guest actually reported than round it up to make the change look tidier.

The team also checked the guest's memory and sampled continuity after the restart. I don't have a full list of what was sampled, so I'll leave it there. The point of those checks wasn't to declare the job finished. It was to see that the guest had come back and that work could carry on.

What can we say after the restart?

The memory change was applied, and the guest came back. What we can't say is that the warning is cured, or that every peak workload is stable. I don't have evidence either way on whether the warning has come back under ordinary use, so I'm not claiming it has gone.

The work was to make the change in a way that left us able to recover, and to check what actually happened.

If you need to change a VM's memory, this is the sequence I'd use. Check the host has sufficient headroom and identify the exact VM. Save current work and have a person or supported route available to restart it from outside the guest. Confirm the VM is fully powered off before a memory-only change, keep a rollback plan, and after restart check both the guest's reported memory and the services you depend on.

Your hypervisor's steps may differ. Use its current documentation rather than copying someone else's commands or identifiers.

A prompt that keeps the plan on paper until you're ready: "Prepare a memory-only VM upgrade and rollback plan. Separate host and guest memory, include a clean shutdown, a way to restart from outside the guest, and post-restart checks. Don't execute it."

Applying the setting and seeing a new number is only part of it. Confirming that the team can actually work again is the check I care about.

What's next

Next is seeing how the guest behaves under real workloads. Until we've watched it in ordinary use, I won't claim the warning is gone for good, and I won't claim the larger allocation made the team faster. We haven't measured a productivity change from this step.

This change sits inside the same VM I wrote about in Why I put OpenBot in a VM, and why I kept it. It's the machine I use while building the Gabba Works apps, including PDF Desk. The memory change was a setup step on that machine. It isn't a change to the product, and it doesn't affect how PDF Desk keeps files on your device.

The guest came back with about 11.66 GiB, which is what I wanted to see. I'll watch ordinary use before I say more about the warning.

Thanks for reading,
Gabba