Real Kubernetes in Docker. Load balancers out of the box, Platform UI, pull-through image cache, external cloud nodes, pause & resume. Everything KinD doesn’t have.
KinD served the community well, but as workflows grow more complex, its limitations become daily friction. vīnd was built to eliminate every one of them.
Run kind load docker-image every time you build locally. One extra step that compounds into hours of wasted time.
Services of type LoadBalancer never get an external IP. You need MetalLB or cloud-provider-kind just to test a basic service.
Need to upgrade Kubernetes or take a break from a project? The only answer is delete and recreate, losing everything you configured.
No cloud nodes, no GPU workloads, no way to share tenant clusters with teammates. KinD is an island.
KinD is CLI-only. No visual management, no observability, no dashboard. Just kubectl all the way down.
Ingress, DNS, and networking require brittle manual config that breaks between setups. Every demo is a gamble.
vīnd was designed from the ground up to fix every major friction point in local Kubernetes development.
vīnd isn’t just a drop-in replacement. It’s a step change in what local Kubernetes can do.
vīnd’s integrated cache means every image in your local Docker daemon is immediately available inside the cluster. Build locally, use instantly. No kind load ever again.
# KinD: $ kind load docker-image my-app:latest
$ kubectl run my-app --image=my-app:latest
✓ Pulled from local daemon instantlyWorking across multiple projects? Put any vīnd cluster to sleep and all containers stop consuming resources immediately. Wake up in seconds with full state intact. No waiting, no rebuilding.
$ vcluster sleep my-cluster
✓ Successfully paused vCluster my-cluster
$ vcluster wakeup my-cluster
✓ Successfully resumed vCluster my-clusterDeploy a service with type: LoadBalancer and get a real, locally accessible IP instantly. No MetalLB, no tunnels, no workarounds. Works on Mac and Linux out of the box.
spec:
type: LoadBalancer
✓ EXTERNAL-IP: 127.0.0.1 assigned
✓ curl http://127.0.0.1 → 200 OKJoin any cloud instance (AWS EC2, GCP, Azure VMs, or bare-metal GPU nodes) directly into your local vīnd cluster with a single token. Local laptop and cloud GPU in one cluster.
$ vcluster token create my-cluster
✓ Join token ready
# Run on any cloud instance:
$ bash <(curl join-token-url)
✓ EC2 node joined your clusterLaunch the vCluster Platform as a Docker container alongside vīnd and get a full web UI to manage all your clusters visually. Observe workloads, inspect nodes, and share access with teammates.
$ vcluster platform start
✓ Platform UI at https://...
✓ Free tier activated
# Manage everything from any browserChange the Kubernetes version, add worker nodes, swap the CNI: just update your vcluster.yaml and run upgrade. No cluster deletion, no state loss, no reconfiguration from scratch.
$ vcluster create my-cluster \
--upgrade -f vcluster.yaml
✓ K8s upgraded to v1.31
✓ Worker node addedSave your entire cluster, Deployments, PVCs, ConfigMaps, everything, into a single portable file or push to an OCI registry. Share it with your team. Restore on any machine in one command.
$ vcluster snapshot create my-cluster \
oci://ghcr.io/team/snapshots:ml-stack \
--driver docker
✓ Snapshot pushed to OCI registry
# Your teammate, any machine:
$ vcluster restore my-cluster \
oci://ghcr.io/team/snapshots:ml-stack \
--driver docker
✓ Full environment restored in secondsvīnd removes the daily friction of local Kubernetes so your team can focus on building, not babysitting infrastructure.
The pull-through cache means every local Docker image is automatically available in your cluster the moment you need it.
Sleep your cluster when switching tasks. Wake it up in under 3 seconds, full state intact, no rebuild required.
Join AWS, GCP, Azure, or on-prem nodes into your local cluster via the built-in VPN. Test GPU workloads without a full cloud setup.
Real load balancers mean your local environment matches production. What works locally actually works when deployed. Fewer surprises at ship time.
Share your local vīnd cluster with teammates via the platform UI. No manual VPN setup, no kubectl context exports. They log in and they’re in.
vīnd is part of the vCluster open source project. No license fees, no feature gates on core functionality, no vendor lock-in whatsoever.
“After switching to vīnd from minikube, it’s really a big help. It allows local images to pull directly from the Docker cache. In kind there’s always that extra step to pull the image into the environment. vīnd removes that hurdle completely from my workflow.”
Community developer, vCluster GitHub
From inner-loop development to production-like testing, vīnd adapts to how you and your team actually work.
Build an image, apply your manifest, see it running in seconds. No imports, no tunnels.
Spin up a real cluster in your pipeline, run integration tests, tear it down. No cloud costs, no flaky shared environments.
Test platform configurations locally with the free UI before rolling out to production.
Create repeatable local environments that work offline, on conference Wi-Fi, or in the air. No cloud provisioning needed.
Join a cloud GPU node into your local cluster. Run model training or test inference, then detach when done. No EKS setup needed.
Run it on your home server, pause when idle, resume when needed. Full Platform UI, VPN, and auto nodes included free.
All you need is Docker and the vCluster CLI. No existing Kubernetes cluster required.
Available via Homebrew, direct download, or binary for Mac, Linux, and Windows.
One command tells the CLI to use Docker instead of Kubernetes as the deployment target.
Your cluster is ready. Kubeconfig is automatically configured. Start deploying.
bash
# Install vCluster CLI (Linux)
$ curl -L -o vcluster "https://github.com/loft-sh/vcluster/releases/latest/download/vcluster-linux-amd64"
$ chmod +x vcluster && sudo mv vcluster /usr/local/bin
# Or via Homebrew
$ brew install loft-sh/tap/vcluster
# Enable Docker driver (one-time setup)
$ vcluster use driver docker
# (Optional) Start the free Platform UI
$ vcluster platform start
# Create your first cluster
$ vcluster create my-cluster
✓ Cluster "my-cluster" ready
✓ kubeconfig updated, kubectl is readyvīnd is completely open source, part of the vCluster project. No license, no credit card, no catch. Just better local Kubernetes, starting today.
Works on Mac and Linux. Windows support is available but LoadBalancer services are not yet supported on Windows.
