i.
Idle-aware scheduling.
The daemon yields the moment a user touches the keyboard, mouse, or any compute-heavy process they care about.
Squadron packages your code, ships it to every idle workstation in your lab, and runs your jobs in parallel. No Docker, Slurm, or AWS invoices, just the GPUs you already paid for.
↳Squadron packages your repo against a base image, builds the container for you, and pushes it to a private registry. Your colleagues’ idle machines pull and run. Logs and artifacts come back to you.
The daemon yields the moment a user touches the keyboard, mouse, or any compute-heavy process they care about.
Pick a base image and your job runs the same on every machine. Dependencies stay isolated and runs stay reproducible, with no setup to manage.
Volumes mount as ordinary filesystems inside the job, fetching files only when read and then caching for future reads. Open a TB-sized dataset the same way on every machine, without syncing or copying anything up front.
Tail stdout from any job in any terminal. Outputs upload to S3 as they're written.
Clone a working starter project and run it locally before shipping it to the cluster.
squadron examples clone hello-worldPackage your project directory into a container and run it on any idle workstation.
squadron project upload && squadron job queueUpload a dataset once, then reference it from any project manifest.
squadron volume upload ./datasetsTail logs and outputs from your terminal while the job runs on someone else's machine across the lab.
squadron job tail --latest