Virtual Kubernetes Control Planes

it's a cluster f*ck.

Run thousands of isolated Kubernetes control planes on a single API server. ~3 MB overhead each. Native semantics — no syncing, no translation layers.

Get StartedHow It Works

Getting started

Three commands.

01Install the CLI
$ brew tap kplane-dev/tap$ brew install kplane
02Create management plane
$ kplane up
03Create a virtual cluster
$ kp create cluster example
04Full Kubernetes API, ready immediately
$ kubectl get namespaces --context=example# Full Kubernetes API, ready immediately

How it works

kplane virtualizes control planes at the API server level. Instead of separate infrastructure per tenant, one process serves all clusters with path-based isolation.

  • Each control plane gets its own Kubernetes API endpoint
  • Full RBAC, CRD, and namespace isolation per control plane
  • Controllers and informers share caches — adding a cluster adds ~3 MB, not another cache tree
  • Workloads schedule directly on shared or dedicated nodes — no syncer copying resources between layers
  • kubectl, Helm, ArgoCD, Terraform, and every Kubernetes tool works unchanged
Shared API Server

Single process, path-isolated storage

/cp-1
/cp-2
/cp-3
/cp-4
...
Shared Node Pool

A better multi-tenancy primitive

Namespace Isolation

  • Namespaces + RBAC + NetworkPolicy on a shared cluster
  • No CRD isolation, no cluster-scoped resource separation
  • Fine for trusted teams, breaks when tenants need cluster-admin or custom CRDs

Synced Virtual Clusters

  • Separate API server + syncer per tenant
  • Syncer copies Pods and Secrets to a host cluster for scheduling
  • Higher-level resources stay in the virtual cluster and don't sync
  • ~128 MB+ overhead per cluster, 30-60 s provisioning
  • Behavioral gaps for resource types the syncer doesn't handle

Virtual Control Planes

  • One shared API server, path-isolated per tenant
  • No syncer, no per-cluster process, no per-cluster datastore
  • ~3 MB overhead per control plane, ~2 s provisioning
  • Full Kubernetes API — nothing is translated or approximated

Use Cases

Who runs kplane

Platform Engineering

Self-service control planes for internal developers. Provision in seconds, tear down when done. One management plane to operate instead of a fleet.

Learn more

SaaS Multi-Tenancy

Isolated Kubernetes environments per customer without running separate clusters. Control plane isolation at namespace cost.

Learn more

CI/CD & Testing

Ephemeral control planes per PR or test suite. Hundreds running concurrently on modest infrastructure. Full Kubernetes API — not a mock.

Learn more

AI / GPU Infrastructure

AI agents are moving from VMs to managing entire clouds. Give each agent its own control plane and scheduling view over shared GPU pools — full Kubernetes API without duplicating expensive hardware.

Learn more

Density

~3MBper control plane
~2sprovisioning
Nconcurrent planes

Give everyone a control plane.

Get started