SkyPilot
AI compute platform that runs, manages, and scales AI workloads across Kubernetes, Slurm, and 20+ clouds.
At a Glance
Self-hosted SkyPilot core released under the Apache License 2.0.
Engagement
Available On
Alternatives
Listed Nov 2026
About SkyPilot
SkyPilot is an AI compute platform, originally an open-source project from UC Berkeley's Sky Computing Lab, that gives AI teams one interface to run jobs on any infrastructure. Infra teams get a unified control plane for Kubernetes, Slurm, VMs, on-prem clusters, and 20+ clouds. It is available as an Apache-2.0 open-source project and as a managed SkyPilot Platform.
What It Is
SkyPilot unifies fragmented GPU, TPU, and CPU compute into a single pool. Users describe a task in YAML or the Python API, covering resources, synced data, setup commands, and run commands. SkyPilot then finds available infrastructure across clusters or clouds, provisions it with auto-failover on capacity errors, syncs the working directory, runs setup, and streams logs. It runs inside the user's own cloud accounts, VPCs, and clusters (BYOC).
Workloads It Covers
The platform page describes devspaces (remote GPU dev environments over SSH or IDE), a trainer for large multi-node runs with gang scheduling and hot swapping of faulty GPUs, sandboxes for agent and RL code with gVisor and microVM isolation, and inference endpoints with prefill-decode disaggregation and KV cache-aware routing. Batch inference and post-training/RL are also listed. An official SkyPilot Skill lets coding agents such as Claude Code and Codex launch and manage workloads.
Fleet Management and Governance
The platform page lists quota borrowing, priorities and preemption, topology-aware scheduling, a fleet GPU map, automatic GPU health checks, and metrics and logs. Governance features include usage and cost tracking, admin policies, audit logs, SSO and RBAC, secrets management, and SOC 2 Type II (report on request). Slack notifications and webhooks report job status changes, GPU failures, and preemptions.
Open Source vs Platform
Per the vendor's comparison, open source includes multi-cloud/K8s/Slurm orchestration and auto-failover. The managed Platform adds a high-availability control plane, team quotas, GPU/NCCL failure recovery, cost tracking, fleet observability, SSO/RBAC, Devspaces, Sandboxes, Endpoints, and 24x7 support from the core team.
Community Discussions
Be the first to start a conversation about SkyPilot
Share your experience with SkyPilot, ask questions, or help others learn from your insights.
Pricing
Open Source
Self-hosted SkyPilot core released under the Apache License 2.0.
- Apache License 2.0 (OSI-approved)
- Run, manage, and scale AI workloads on any AI infrastructure
- BYOC and BYOK (bring your own cloud/keys)
Contact sales
SkyPilot Platform; demo request available, pricing not published.
- SkyPilot Platform
- Request a Demo
Capabilities
Key Features
- Unified interface for Kubernetes, Slurm, VMs, on-prem, and 20+ clouds
- YAML and Python API task definitions
- Auto-failover across infrastructure on capacity errors
- Autostop of idle resources
- Gang scheduling and multi-node jobs
- Multi-cluster support
- Quotas, priorities, and preemption
- Topology-aware scheduling
- GPU health checks with auto-remediation
- Devspaces with SSH or IDE access
- Sandboxes with gVisor and microVM isolation
- Inference endpoints with KV cache-aware routing
- Cost tracking and per-user GPU-hours
- SSO, RBAC, admin policies, and audit logs
- Slack notifications and webhooks
- Official SkyPilot Skill for coding agents
