Spin up isolated clusters per PR, per experiment, or per notebook. No setup. Automatic cleanup. GPU-ready.
Multiple teams deploying incompatible custom resources crash each other's workloads.
Long-running jobs starve other experiments. Quotas don't provide true resource isolation.
Leftover resources, failed deployments, and cluster state from other teams interfere with your work.
Shared clusters mean slow testing, unreliable results, and manual troubleshooting.
Ghostcluster creates an ephemeral, fully-isolated Kubernetes cluster for every PR, experiment, or notebook. Each cluster has its own API server, scheduler, and etcd—no shared state, no conflicts.
All clusters share the underlying host GPUs and nodes efficiently. When your TTL expires, the cluster auto-destructs. No cleanup. No orphaned resources.
Standard kubectl. Standard workloads. Instant isolation.
Your PR / Notebook
↓
ghostctl up
↓
┌──────────────────┐
│ vCluster Instance│
│ (isolated) │
│ API + scheduler │
│ + local etcd │
└──────────────────┘
↓
Shared GPU Node Pool
↓
Auto-destroy on TTL
One CLI command spins up a complete K8s cluster with GPU quotas in seconds.
Full control plane isolation. CRDs, RBAC, and resources scoped to your cluster only.
GPU quotas guaranteed. Other experiments can't starve your workload.
TTL-based expiration. Cluster tears down automatically—zero manual teardown.
Then deploy with standard kubectl. Cluster auto-cleans when TTL expires.
Real Kubernetes in seconds. Test operators, CRDs, GPU workloads without fear.
GPU quotas guaranteed. No competition. No starvation. Run experiments in parallel.
TTL handles teardown. No orphaned resources. No surprise costs. No toil.
CI/CD integration. Cluster per PR. Code → review → isolated test → merge.
Clusters only live as long as needed. Auto-expiration = no wasted resources.
Full Kubernetes isolation without overhead. Built on proven vCluster technology.
Start with open-source. Upgrade to managed automation when you're ready. Same ghostctl CLI. No lock-in.
Open-source. Self-hosted. Available now.
Managed. Hands-off automation. Coming soon.
vCluster-powered: Production-grade cluster isolation used at scale by enterprises.
Kubernetes-native: Standard API, standard tools, standard workloads. No vendor lock-in.
Open-source: MIT licensed. Community-driven. Your infrastructure, your control.
Ghostcluster is free, open-source, and production-ready. Deploy on your infrastructure today.
MIT Licensed • Community-driven • Self-hosted on your infrastructure