skypilot
The AI Compute Platform for frontier teams. SkyPilot turns fragmented AI compute into one AI supercomputer, so frontier AI teams build custom intelligence faster.
SkyPilot is an agent skill that provides AI assistants with the ability to manage and orchestrate compute workloads across diverse infrastructure environments. It enables agents to launch jobs, provision GPUs, and manage AI workloads on Kubernetes clusters, Slurm systems, and over 20 cloud providers through a unified interface. Designed for AI teams building custom models and applications, the skill allows agents to handle resource provisioning, job scheduling, and cluster management tasks without requiring manual infrastructure configuration. The skill translates natural language requests into SkyPilot commands for compute orchestration.
Key Features
Use Cases
- 01Training large language models across multiple GPU clusters with automatic resource failover
- 02Deploying and serving inference endpoints for models like Llama, DeepSeek-R1, or Qwen on owned infrastructure
- 03Running reinforcement learning experiments with independent scaling of training and inference workloads
- 04Orchestrating batch processing jobs across reserved GPUs, spot instances, and multiple cloud providers
- 05Managing interactive development environments on Kubernetes with simplified pod access and code syncing
- 06Executing distributed PyTorch, DeepSpeed, or Ray training jobs with gang scheduling and multi-node coordination
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
skypilot — FAQ
What is the SkyPilot agent skill?+
The SkyPilot agent skill enables AI assistants like Claude Code and Codex to manage AI compute workloads across Kubernetes, Slurm clusters, and multiple cloud providers. It provides a unified interface for provisioning GPUs, launching training jobs, and orchestrating inference serving without manual infrastructure setup.
How do I install the SkyPilot skill for my AI agent?+
Install SkyPilot using uv or pip with your desired cloud providers, then instruct your agent to fetch and follow the installation guide at https://github.com/skypilot-org/skypilot/blob/HEAD/agent/INSTALL.md. The agent will then have access to SkyPilot commands for compute management.
Which AI assistants work with the SkyPilot skill?+
The SkyPilot skill is designed to work with agent-based coding assistants including Claude Code, Codex, and similar tools that can execute shell commands and follow installation instructions. Any agent capable of running Python CLI tools can integrate with SkyPilot.
Do I need API keys or cloud credentials to use SkyPilot?+
Yes, you need credentials for the infrastructure you want to use. This includes cloud provider API keys (AWS, GCP, Azure, etc.), Kubernetes cluster access, or Slurm credentials. SkyPilot operates within your own cloud accounts and clusters (BYOC - Bring Your Own Cloud).
Is SkyPilot free to use?+
Yes, SkyPilot itself is free and open-source software. However, you pay for the underlying compute resources (GPUs, VMs, cloud instances) through your own infrastructure providers. SkyPilot helps optimize costs by finding the cheapest available resources and auto-stopping idle instances.
What types of AI workloads can SkyPilot handle?+
SkyPilot supports training jobs (PyTorch, DeepSpeed, Jax/TPU), model serving (vLLM, SGLang, Ollama), reinforcement learning, batch inference, fine-tuning LLMs, and distributed multi-node workloads. It works with existing code without modifications.
How do I install skypilot?+
Open the source repository on GitHub and follow its README. skypilot is a skill — MCP Agents Market links you directly to the official repo.
Is skypilot free?+
skypilot is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.