skypilot

Why we included this project

SkyPilot is for teams whose AI workloads spread across more than one cloud, Kubernetes cluster, or Slurm fleet, and who are tired of stitching that infrastructure together by hand. You get one CLI for launching jobs and scaling GPU capacity, with a scheduler that handles placement, spot instance preemption, and prioritized job queues across providers. People use it for everything from short-lived dev boxes and sandboxes for LLM-generated code to large pre-training runs and online RL jobs spanning thousands of GPUs. It runs on top of infrastructure you already have, whether that is Kubernetes, Slurm, on-prem VMs, or public clouds like AWS, GCP, Azure, and CoreWeave, so you are not locked into one vendor. For small teams, the CLI keeps the learning curve low compared with hand-rolling YAML and cluster wiring.