# SkyPilot SkyPilot is a system for running, managing, and scaling AI workloads on any AI infrastructure. SkyPilot gives AI teams a simple interface to run jobs on any infra. Infra teams get a unified control plane to manage any AI compute — with advanced scheduling, scaling, and orchestration. ## Why SkyPilot SkyPilot is easy to use for AI users: - Quickly spin up compute on your own infra - Environment and job as code — simple and portable - Easy job management: queue, run, and auto-recover many jobs SkyPilot makes Kubernetes easy for AI & Infra teams: - Slurm-like ease of use, cloud-native robustness - Local dev experience on K8s: SSH into pods, sync code, or connect IDE - Turbocharge your clusters: gang scheduling, multi-cluster, and scaling SkyPilot unifies multiple clusters, clouds, and hardware: - One interface to use reserved GPUs, Kubernetes clusters, or 16+ clouds - Flexible provisioning of GPUs, TPUs, CPUs, with smart failover - Team deployment and resource sharing SkyPilot cuts your cloud costs & maximizes GPU availability: - Autostop: automatic cleanup of idle resources - Spot instance support: 3-6x cost savings, with preemption auto-recovery - Intelligent scheduling: automatically run on the cheapest and most available infra ## Getting Started - [Overview](https://docs.skypilot.ai/en/latest/overview.html.md): SkyPilot architecture and core concepts - [Installation](https://docs.skypilot.ai/en/latest/getting-started/installation.html.md): Install SkyPilot with pip for your cloud providers - [Quickstart](https://docs.skypilot.ai/en/latest/getting-started/quickstart.html.md): Launch your first cluster and run a job - [Agent Skills](https://docs.skypilot.ai/en/latest/getting-started/skill.html.md): Install the SkyPilot skill for your AI agents - [Examples](https://docs.skypilot.ai/en/latest/examples/index.html.md): Browse example use cases - [Sky Computing](https://docs.skypilot.ai/en/latest/sky-computing.html.md): Learn about the Sky Computing paradigm ## Team Deployment (API Server) - [API Server Deployment](https://docs.skypilot.ai/en/latest/reference/api-server/api-server.html.md): Deploy SkyPilot as a service for teams - [Remote API Server Setup](https://docs.skypilot.ai/en/latest/reference/api-server/api-server.html.md#sky-api-server-remote): Multi-user team deployment - [Admin Deployment Guide](https://docs.skypilot.ai/en/latest/reference/api-server/api-server-admin-deploy.html.md): Production deployment instructions - [Multi-tenancy](https://docs.skypilot.ai/en/latest/reference/api-server/api-server.html.md#multi-tenancy): Share clusters and jobs securely - [Authentication (SSO) and RBAC](https://docs.skypilot.ai/en/latest/reference/auth.html.md): User authentication (SSO, basic auth, service accounts) and authorization - [Workspaces](https://docs.skypilot.ai/en/latest/admin/workspaces.html.md): Manage team workspaces ## Clusters - [Interactive Development](https://docs.skypilot.ai/en/latest/examples/interactive-development.html.md): SSH, sync code, and connect IDEs on cloud and Kubernetes - [Cluster Jobs](https://docs.skypilot.ai/en/latest/reference/job-queue.html.md): Job queue and scheduling - [Auto-failover](https://docs.skypilot.ai/en/latest/examples/auto-failover.html.md): Automatic failover across regions/clouds - [Auto-stop](https://docs.skypilot.ai/en/latest/reference/auto-stop.html.md): Automatically stop idle clusters ## Jobs - [Managed Jobs](https://docs.skypilot.ai/en/latest/examples/managed-jobs.html.md): Auto-recover from job failure and preemptions. - [Multi-Node Jobs](https://docs.skypilot.ai/en/latest/running-jobs/distributed-jobs.html.md): Run distributed training and inference - [Many Parallel Jobs](https://docs.skypilot.ai/en/latest/running-jobs/many-jobs.html.md): Run many jobs efficiently - [Model Training Guide](https://docs.skypilot.ai/en/latest/reference/training-guide.html.md): Best practices for training ## Model Serving - [Getting Started](https://docs.skypilot.ai/en/latest/serving/sky-serve.html.md): Production-ready model serving with SkyServe - [Autoscaling](https://docs.skypilot.ai/en/latest/serving/autoscaling.html.md): Scale replicas based on load - [Authentication](https://docs.skypilot.ai/en/latest/serving/auth.html.md): Secure your serving endpoints - [HTTPS Support](https://docs.skypilot.ai/en/latest/serving/https.html.md): Enable HTTPS for serving - [Spot Instances](https://docs.skypilot.ai/en/latest/serving/spot-policy.html.md): Use spot instances for cost savings - [Service Updates](https://docs.skypilot.ai/en/latest/serving/update.html.md): Update running services ## Infrastructure Choices - [Kubernetes](https://docs.skypilot.ai/en/latest/reference/kubernetes/index.html.md): Deploy on any Kubernetes cluster - [Kubernetes Getting Started](https://docs.skypilot.ai/en/latest/reference/kubernetes/kubernetes-getting-started.html.md): Quick start guide - [Multi-Kubernetes](https://docs.skypilot.ai/en/latest/reference/kubernetes/multi-kubernetes.html.md): Deploy across multiple clusters - [Using Existing Machines](https://docs.skypilot.ai/en/latest/reservations/existing-machines.html.md): Connect your own machines - [Reserved Instances](https://docs.skypilot.ai/en/latest/reservations/reservations.html.md): Use cloud reserved instances - [Using Cloud VMs](https://docs.skypilot.ai/en/latest/compute/cloud-vm.html.md): Launch cloud virtual machines - [GPU Support](https://docs.skypilot.ai/en/latest/compute/gpus.html.md): GPU availability and pricing ## Data - [Storage Overview](https://docs.skypilot.ai/en/latest/reference/storage.html.md): Manage cloud storage and volumes - [Volumes](https://docs.skypilot.ai/en/latest/reference/volumes.html.md): Attach persistent volumes - [Syncing Code](https://docs.skypilot.ai/en/latest/examples/syncing-code-artifacts.html.md): Sync code and artifacts efficiently ## User Guides - [Async API](https://docs.skypilot.ai/en/latest/reference/async.html.md): Asynchronous operations - [Environment Variables](https://docs.skypilot.ai/en/latest/running-jobs/environment-variables.html.md): Configure environments - [Docker Containers](https://docs.skypilot.ai/en/latest/examples/docker-containers.html.md): Use custom Docker images - [Ports](https://docs.skypilot.ai/en/latest/examples/ports.html.md): Expose and access ports - [Logging](https://docs.skypilot.ai/en/latest/reference/logging.html.md): Logging configuration - [FAQ](https://docs.skypilot.ai/en/latest/reference/faq.html.md): Frequently asked questions ## Administrator Guides - [API Server Deployment](https://docs.skypilot.ai/en/latest/reference/api-server/api-server.html.md): Deploy SkyPilot for teams - [Admin Deployment Guide](https://docs.skypilot.ai/en/latest/reference/api-server/api-server-admin-deploy.html.md): Production deployment - [Authentication (SSO) and RBAC](https://docs.skypilot.ai/en/latest/reference/auth.html.md): User authentication (SSO, basic auth, service accounts) and authorization - [Workspaces](https://docs.skypilot.ai/en/latest/admin/workspaces.html.md): Manage team workspaces - [Cloud Permissions](https://docs.skypilot.ai/en/latest/cloud-setup/cloud-permissions/index.html.md): Required IAM permissions - [Admin Policies](https://docs.skypilot.ai/en/latest/cloud-setup/policy.html.md): Configure cloud policies - [External Logging Storage](https://docs.skypilot.ai/en/latest/cloud-setup/external-logging.html.md): Configure logging storage ## References - [Task YAML](https://docs.skypilot.ai/en/latest/reference/yaml-spec.html.md): Complete YAML specification - [CLI](https://docs.skypilot.ai/en/latest/reference/cli.html.md): All CLI commands and options - [Python API](https://docs.skypilot.ai/en/latest/reference/api.html.md): Complete Python API documentation - [Config](https://docs.skypilot.ai/en/latest/reference/config.html.md): Configuration options - [SkyPilot State](https://docs.skypilot.ai/en/latest/reference/architecture/state.html.md): System architecture - [Developer Guide](https://docs.skypilot.ai/en/latest/developers/index.html.md): Contribute to SkyPilot ## Serving Examples - [vLLM](https://docs.skypilot.ai/en/latest/examples/serving/vllm.html.md): Serve LLMs with vLLM - [TGI](https://docs.skypilot.ai/en/latest/examples/serving/tgi.html.md): Text Generation Inference - [SGLang](https://docs.skypilot.ai/en/latest/examples/serving/sglang.html.md): SGLang serving - [Ollama](https://docs.skypilot.ai/en/latest/examples/serving/ollama.html.md): Ollama integration - [LoRAX](https://docs.skypilot.ai/en/latest/examples/serving/lorax.html.md): LoRA serving - [Cog](https://docs.skypilot.ai/en/latest/examples/serving/cog.html.md): Cog containers ## LLM Examples - [Llama 4](https://docs.skypilot.ai/en/latest/examples/models/llama-4.html.md): Deploy and serve Llama 4 models - [Llama 3.2](https://docs.skypilot.ai/en/latest/examples/models/llama-3_2.html.md): Run Llama 3.2 multimodal models - [Llama 3.1](https://docs.skypilot.ai/en/latest/examples/models/llama-3_1.html.md): Deploy Llama 3.1 405B and smaller - [Llama 3](https://docs.skypilot.ai/en/latest/examples/models/llama-3.html.md): Serve Llama 3 models - [Llama 2](https://docs.skypilot.ai/en/latest/examples/models/llama-2.html.md): Run Llama 2 models - [DeepSeek R1](https://docs.skypilot.ai/en/latest/examples/models/deepseek-r1.html.md): Deploy DeepSeek R1 reasoning models - [DeepSeek R1 Distilled](https://docs.skypilot.ai/en/latest/examples/models/deepseek-r1-distilled.html.md): Run distilled R1 models - [DeepSeek Janus](https://docs.skypilot.ai/en/latest/examples/models/deepseek-janus.html.md): Multimodal DeepSeek models - [Mixtral](https://docs.skypilot.ai/en/latest/examples/models/mixtral.html.md): Serve Mixtral MoE models - [Pixtral](https://docs.skypilot.ai/en/latest/examples/models/pixtral.html.md): Deploy Pixtral vision models - [Qwen](https://docs.skypilot.ai/en/latest/examples/models/qwen.html.md): Run Qwen/Qwen2 models - [Kimi K2](https://docs.skypilot.ai/en/latest/examples/models/kimi-k2.html.md): Deploy Kimi K2 models - [Yi](https://docs.skypilot.ai/en/latest/examples/models/yi.html.md): Serve Yi models - [Gemma](https://docs.skypilot.ai/en/latest/examples/models/gemma.html.md): Run Google Gemma models - [GPT-OSS](https://docs.skypilot.ai/en/latest/examples/models/gpt-oss.html.md): Deploy OpenAI GPT-OSS models - [CodeLlama](https://docs.skypilot.ai/en/latest/examples/models/codellama.html.md): Code generation with CodeLlama - [Vicuna](https://docs.skypilot.ai/en/latest/examples/models/vicuna.html.md): Serve Vicuna models - [DBRX](https://docs.skypilot.ai/en/latest/examples/models/dbrx.html.md): Deploy Databricks DBRX ## Fine-tuning - [Llama 4 Fine-tuning](https://docs.skypilot.ai/en/latest/examples/training/llama-4-finetuning.html.md): Fine-tune Llama 4 models - [Llama 3.1 Fine-tuning](https://docs.skypilot.ai/en/latest/examples/training/llama-3_1-finetuning.html.md): LoRA and full fine-tuning - [Llama 2 Fine-tuning](https://docs.skypilot.ai/en/latest/examples/training/llama-2-finetuning.html.md): Fine-tune Llama 2 - [GPT-OSS Fine-tuning](https://docs.skypilot.ai/en/latest/examples/training/gpt-oss-finetuning.html.md): Fine-tune GPT-OSS models - [Axolotl](https://docs.skypilot.ai/en/latest/examples/training/axolotl.html.md): Use Axolotl for fine-tuning - [Unsloth](https://docs.skypilot.ai/en/latest/examples/training/unsloth.html.md): Fast fine-tuning with Unsloth - [DeepSpeed](https://docs.skypilot.ai/en/latest/examples/training/deepspeed.html.md): Distributed training with DeepSpeed - [VeRL](https://docs.skypilot.ai/en/latest/examples/training/verl.html.md): RL training for LLMs with VeRL ## Training Frameworks - [PyTorch Distributed](https://docs.skypilot.ai/en/latest/examples/training/distributed-pytorch.html.md): Multi-node PyTorch training - [TensorFlow Distributed](https://docs.skypilot.ai/en/latest/examples/training/distributed-tensorflow.html.md): Distributed TensorFlow - [Ray Train](https://docs.skypilot.ai/en/latest/examples/training/ray.html.md): Distributed training with Ray - [NeMo](https://docs.skypilot.ai/en/latest/examples/training/nemo.html.md): NVIDIA NeMo framework - [TPU Training](https://docs.skypilot.ai/en/latest/examples/training/tpu.html.md): Train on Google Cloud TPUs - [GPT-2 Training](https://docs.skypilot.ai/en/latest/examples/models/gpt-2.html.md): Train GPT-2 from scratch ## Performance & Optimization - [AWS EFA](https://docs.skypilot.ai/en/latest/examples/performance/aws_efa.html.md): High-performance networking on AWS - [GCP GPU Direct TCPx](https://docs.skypilot.ai/en/latest/examples/performance/gcp_gpu_direct_tcpx.html.md): GPU Direct on GCP - [Nebius InfiniBand](https://docs.skypilot.ai/en/latest/examples/performance/nebius_infiniband.html.md): InfiniBand on Nebius ## Applications - [RAG Systems](https://docs.skypilot.ai/en/latest/examples/applications/rag.html.md): Build retrieval-augmented generation - [Batch Inference](https://docs.skypilot.ai/en/latest/examples/applications/batch_inference.html.md): Run batch inference jobs - [LocalGPT](https://docs.skypilot.ai/en/latest/examples/applications/localgpt.html.md): Deploy LocalGPT - [Stable Diffusion](https://docs.skypilot.ai/en/latest/examples/applications/stable_diffusion.html.md): Image generation - [Tabby](https://docs.skypilot.ai/en/latest/examples/applications/tabby.html.md): Code completion service - [Vector Databases](https://docs.skypilot.ai/en/latest/examples/applications/vector_database.html.md): Deploy vector DBs ## Orchestration - [Airflow Integration](https://docs.skypilot.ai/en/latest/examples/orchestrators/airflow.html.md): Use with Apache Airflow - [GitHub Actions](https://docs.skypilot.ai/en/latest/examples/orchestrators/github_actions.html.md): CI/CD integration - [Cron Jobs](https://docs.skypilot.ai/en/latest/examples/orchestrators/cron.html.md): Schedule recurring jobs - [Jupyter Notebooks](https://docs.skypilot.ai/en/latest/examples/frameworks/jupyter.html.md): Run Jupyter on clusters - [DVC Integration](https://docs.skypilot.ai/en/latest/examples/frameworks/dvc.html.md): Data version control - [MPI](https://docs.skypilot.ai/en/latest/examples/frameworks/mpi.html.md): Run MPI applications ## Cloud Providers - [AWS Setup](https://docs.skypilot.ai/en/latest/cloud-setup/cloud-permissions/aws.html.md): Amazon Web Services configuration - [GCP Setup](https://docs.skypilot.ai/en/latest/cloud-setup/cloud-permissions/gcp.html.md): Google Cloud Platform setup - [Kubernetes Setup](https://docs.skypilot.ai/en/latest/cloud-setup/cloud-permissions/kubernetes.html.md): Kubernetes cluster configuration - [Nebius Setup](https://docs.skypilot.ai/en/latest/cloud-setup/cloud-permissions/nebius.html.md): Nebius cloud configuration - [vSphere Setup](https://docs.skypilot.ai/en/latest/cloud-setup/cloud-permissions/vsphere.html.md): VMware vSphere configuration ## Advanced Topics - [Comparison](https://docs.skypilot.ai/en/latest/reference/comparison.html.md): Compare with other tools - [Troubleshooting](https://docs.skypilot.ai/en/latest/reference/kubernetes/kubernetes-troubleshooting.html.md): Common issues and solutions - [TPU Support](https://docs.skypilot.ai/en/latest/reference/tpu.html.md): Using TPUs - [Quota Management](https://docs.skypilot.ai/en/latest/cloud-setup/quota.html.md): Manage cloud quotas ## Community & Resources - [Release Notes](https://github.com/skypilot-org/skypilot/releases): Version history - [Contributing](https://github.com/skypilot-org/skypilot/blob/master/CONTRIBUTING.md): Contribution guidelines - [Community Slack](https://slack.skypilot.co/): Join the community - [Blog](https://skypilot.ai/blog): News, tutorials, and announcements - [Case Studies](https://skypilot.ai/case-studies): Real-world usage - [GitHub Repository](https://github.com/skypilot-org/skypilot): Source code and issues