FAQ

Will it work?

Which platforms? Linux x86-64 only. There are no macOS, Windows, Arm or GPU runners. KVM is planned.

Public repositories? No. Only private repositories can use the runners.

Does a job move to another cloud when one is short? No. Each label runs in one cloud. When a VM type has no spot capacity, the pool launches the next VM type in the same cloud.

What happens when a spot VM is reclaimed? Its jobs fail. When you have to rerun them, backmesh posts one check run per workflow run that lists every reclaimed job, and you rerun them from the Actions tab. backmesh doesn’t rerun them itself yet, and there are no on-demand runners yet.

Can I add a cloud later? The clouds are chosen at install.

TODO(owner): say whether adding a cloud later needs an uninstall and a fresh install.

Is there a Docker layer cache? Not between jobs: each job starts with an empty Docker data directory.

Is it safe?

See the security model: what runs where, the GitHub App’s permissions, what a stolen key can do, and the one trust level that jobs from every repository share.

What does it cost?

backmesh’s price: TODO(owner): decide and state the price.

Your clouds bill you directly for the VMs, disks and the checkpoint bucket. Install prints each pool’s monthly cost before it creates anything.

Before any job runs: until scale-to-zero ships, each cloud’s pool keeps one spot VM running. With the management VM and bm-control, one AWS cloud costs about $245–260 a month, and three clouds about $635–750. After scale-to-zero, the idle cost is the management VM and bm-control, about $80 a month on AWS. These are estimates from cloud catalog prices on 2026-11-05; spot prices change by the hour.

TODO(owner): confirm that scale-to-zero ships in anycloud 0.1.73 before saying when.