What is vCluster?
vCluster helps AI cloud providers deliver highly scalable, resilient compute clusters as self-service tenant products, including managed Kubernetes, Slurm, inference, Ray, and other cluster types. Each tenant gets a dedicated tenant clusterTenant clusterA fully isolated Kubernetes environment provisioned for a single tenant. Each tenant cluster has its own API server, controller manager, and resource namespace, backed by a virtualized control plane hosted on a control plane cluster. From the tenant's perspective it behaves exactly like a standard Kubernetes cluster., provisioned on your infrastructure or directly on bare metal, with its own control plane and RBAC. Tenants get an isolated cluster experience, with the isolation level set by the worker node model you choose.
The control plane is completely invisible to tenants. There are no shared control plane nodes, no in-cluster agent pods, and no lateral path between environments. vCluster suits any environment where isolation is a hard requirement, from developer platforms and CI/CD pipelines to GPU cloud infrastructure serving paying tenants.
Ready to deploy? See the Quick Start to choose the path that fits your environment.
How it works​
vCluster works across a range of infrastructure configurations. Each tenant cluster is backed by a dedicated virtualized control plane, built on a certified Kubernetes distribution. That's what lets tools built for the standard Kubernetes API, such as kubectl, Helm, Argo, and Crossplane, work against a tenant cluster without modification. It's the same foundation that higher-level cluster types like Slurm, Ray, and inference-serving stacks run on as workloads.
Where the control plane runs
- On an existing Kubernetes cluster — the control plane runs as a pod in a dedicated namespace. Your existing cluster becomes the Control Plane Cluster and no additional infrastructure is required.
- Standalone — a complete, zero-dependency Kubernetes distribution that runs as a self-contained binary on bare metal or VMs. Use it as a full Kubernetes environment for any workload. When connected to vCluster Platform, it can act as a Control Plane Cluster for tenant isolation.
- vind — the full stack runs in Docker containers with no Kubernetes dependency. Suited for local development and CI/CD pipelines.
How tenant workloads run
On cluster-based deployments, you choose how tenant workloads land on compute:
- Shared nodes — tenants share the existing node pool, with multiple tenant clusters running on the same physical nodes under full API-level isolation. Each tenant on shared nodes is a nested cluster, a tenant cluster whose control plane and workloads both run on the control plane cluster's shared node pool. This is the tenancy model vCluster OSS provides. It isn't a security boundary for untrusted tenants, so it suits internal, trusted tenants such as developer platforms, CI environments, and testing rather than externally facing or resold offerings.
- Private nodes — each tenant cluster gets dedicated nodes enrolled through a token-based process. Network, storage, and compute are fully isolated per tenant, with no cross-tenant visibility at the infrastructure level. This model requires a vCluster license and is the isolation model for GPU workloads, regulated industries, and AI cloud platforms serving paying tenants. Nodes can come from bare metal (vMetal), cloud VMs, or any Linux machine.
Building a customer-facing or other production offering? Choose a worker node model walks through the questions that decide which one your offering requires.
Features and plans​
vCluster is available as open source and as an enterprise-grade platform with additional features across paid tiers:
Free, Dev, Prod, and Scale are vCluster Platform license plans. Open source does not need a license or a Platform connection. See Compare open source and free tiers.
Next steps​
- Deploy vCluster — control plane modes, worker node types, and settings locked at deploy time
- Architecture — control plane internals, syncer behavior, and networking
- Private Nodes — dedicated infrastructure for GPU tenants and regulated workloads
- vCluster Standalone — zero-dependency Kubernetes for bare metal and edge
- Building a GPU cloud platform — deployment models for AI cloud providers offering managed Kubernetes
- Building an inference platform — tenancy models and capacity planning for managed inference
- vcluster.yaml reference — full configuration reference