Nvke

Open source alternative to NVIDIA Run:ai

560 stars
Apache-2.0
Active this week
Last commit 1 day ago

Last verified Nov 8, 2026

Deploy a managed Kubernetes service on your own NVIDIA GPU bare-metal infrastructure with API-driven cluster provisioning and lifecycle management.

Visit website GitHub560 Deploy on Hostinger

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who Nvke is for#

AI Factory Operators Offering Kubernetes as a Service

Operators who run GPU fleets and want to offer isolated Kubernetes clusters to their customers gain API-driven provisioning, isolated control planes per customer, automatic node replacement, and an artifact chain they control. NVKE eliminates the work of assembling and maintaining a bespoke cluster management system through every Kubernetes and GPU generation update.

Skip if:

Teams running a small number of clusters for internal use only, where the operational overhead of a management plane exceeds the benefit of automation. At that scale, managing clusters by hand with kubeadm or a lightweight distribution is simpler.

NVIDIA Cloud Partners Building Managed Inference Platforms

NVIDIA Cloud Partners that offer inference or fine-tuning services as a managed product can use NVKE as the Kubernetes management layer beneath their service. NVKE and NVKR handle artifact signing, cluster provisioning, and node health, while the partner builds the customer-facing service on top.

Skip if:

Teams whose offering does not include managed Kubernetes and whose customers interact only with an application API, not a Kubernetes API. NVKE adds infrastructure complexity that is unnecessary if the cluster layer is not customer-facing.

Platform Engineers Integrating Cluster API Components

Kubernetes ISVs and platform teams that want specific NVKE components, such as CAPNICo (the Cluster API provider for NICo) or the OpenTelemetry-based NKTP telemetry pipeline, can integrate them individually without adopting NVKE whole. Both components are open source and designed to work in other Cluster API environments.

Skip if:

Teams not running on NVIDIA or NICo infrastructure. CAPNICo is specific to NICo; the telemetry pipeline works more broadly, but NVKE's other components are designed for the NVIDIA/NICo hardware stack.

Security-Focused GPU Cluster Operators

Teams that require verifiable node identity and an auditable software supply chain benefit from NVKE's TPM-based SPIFFE/SPIRE attestation, SBOM-recorded builds from NVKR, and signed release artifacts. Each node presents its TPM attestation before joining; build provenance is checked against operator policy.

Skip if:

Teams that cannot use TPMs or run virtualized nodes where vTPM support is unavailable. The node attestation path requires TPM hardware or virtual TPM support and is not optional in the production flow.

the problem

The problem it solves#

AI factories use Kubernetes to operate GPU fleets and as the foundation for managed inference, training, and fine-tuning services. Offering Kubernetes to customers as a managed service means building a full cluster management system: API-driven provisioning, isolated control planes per customer, upgrades, node replacement, node identity, and observability across a fleet of GPU machines. Without a dedicated tool for this, an operator or Kubernetes ISV assembles and maintains that system from open-source and commercial components, keeping it working through every Kubernetes release and GPU hardware generation.

The gap is not just the initial build. Cluster management must handle node attestation before machines join, etcd encryption per customer, certificate renewal, node pool scaling, and fault remediation when NVSentinel reports a faulty GPU node. Each Kubernetes upgrade potentially touches all of these concerns. Teams maintaining this stack from scratch carry that toil indefinitely on top of running the actual AI workloads.

how Nvke solves it

How it solves it#

API-Driven Cluster Provisioning

Request workload clusters through the NVKE API or the `nvkectl` CLI. NVKE calls your infrastructure provider (NICo) through Cluster API (CAPNICo), provisions the bare-metal nodes, runs kubeadm to form the cluster, and delivers the customer a kubeconfig. Each workload cluster has its own Kubernetes control plane and etcd.

Workload Cluster Lifecycle Management

Create, scale, and delete node pools, upgrade Kubernetes versions, and replace faulty nodes through the same API. With NVSentinel installed, NVKE automatically reboots or replaces nodes that NVSentinel flags as faulty. Node pools share an instance type, a release, and a Kubernetes version.

TPM-Based Node Identity and Encryption

On bare metal, each node attests with its TPM through SPIFFE/SPIRE before it receives a join credential, so only verified nodes enter the cluster. Per-cluster etcd encryption at rest is available. These controls apply at bootstrap time and are not add-ons.

Verifiable Build Artifact Chain (NVKR)

The companion NVKR project builds node OS images, binaries, container images, and Helm charts from pinned sources and records a software bill of materials (SBOM) for each artifact. Releases from NVKR's hosted pipeline are signed and carry build provenance you check against your own policy before any artifact reaches a cluster.

OpenTelemetry Fleet Telemetry (NKTP)

The NVKE telemetry pipeline runs OpenTelemetry Collectors that gather metrics, logs, and traces from each cluster and export them to your backend over OTLP. NKTP is installed on the management cluster and can run on any other Kubernetes cluster independently.

CNCF-Conformant Kubernetes Clusters

NVKE creates clusters certified as conformant Kubernetes by the Cloud Native Computing Foundation. Customers' workloads, tools, and skills transfer from other Kubernetes environments without modification. NVKE is a Certified Kubernetes AI Platform.

strengths · trade-offs

Strengths and trade-offs#

Strengths

Operator Owns the Full StackYou control the build, the registries, and the supply chain. NVKE can be adopted whole, integrated piece by piece (for example, CAPNICo or the telemetry pipeline alone), or extended. No dependency on NVIDIA-managed infrastructure beyond NICo for bare-metal provisioning.
Apache 2.0 LicenseLicensed under Apache 2.0, NVKE can be run internally, modified, and deployed as part of a commercial service without licensing fees. NVKR, the companion build system, is also Apache 2.0 and accepts outside contributions.
Upstream Kubernetes, Not a ForkNVKE uses upstream Kubernetes, kubeadm, and Cluster API without forking them. CNCF conformance certification means no NVKE-specific behavior to unlearn. Teams migrating workloads from other Kubernetes clusters need no tool or manifest changes.
Actively Developed by NVIDIACreated in August 2026 and reaching 561 GitHub stars and 90 forks in its first few months, NVKE is under active development by NVIDIA and the DSX AI Factory. The project's CI dashboard shows the last 30 runs of every workflow and refreshes hourly.

Trade-offs

-Preview Status: Not Production-ReadyNVKE is explicitly a preview. APIs, configuration, and behavior can change between releases without notice. The management cluster cannot upgrade, scale, or replace itself, and there is no management-plane backup or restore procedure. The team recommends against production workloads until these gaps close.
-Requires NVIDIA Infra Controller (NICo)Real installations depend on NICo as the sole supported infrastructure provider for bare-metal machine provisioning. Teams without NICo access cannot run NVKE against real hardware. Local evaluation uses mock-NICo on a Linux host, but that path is for testing only and requires Docker, virtual TPM support, and at least 6 CPUs and 12 GiB of free memory.
-API Authentication Is Off by DefaultAfter bootstrap, the NVKE API runs over plain HTTP with authentication disabled. The operator must configure API authentication and protected access before admitting callers. Skipping this step leaves the management plane open on the network.
-108 Open Issues at PreviewThe repository had 108 open issues at the time of indexing, expected for a preview project. Teams evaluating NVKE should review open issues for blockers in their specific hardware, networking, or Kubernetes version combination before committing to an integration path.

versus alternatives

Nvke vs alternatives#

NVKE vs NVIDIA Run:ai

Both NVKE and NVIDIA Run:ai address GPU computing on Kubernetes, but they operate at different layers. NVKE manages cluster lifecycle: it creates, provisions, upgrades, and operates Kubernetes clusters themselves, with each workload cluster isolated for each customer. NVIDIA Run:ai is an enterprise GPU orchestration platform that manages resource scheduling and policy enforcement inside existing Kubernetes clusters.

Feature	NVKE	NVIDIA Run:ai
License	Apache 2.0	Proprietary
Self-hosting	Yes, fully operator-owned	No: managed service
Primary function	Cluster provisioning and lifecycle	GPU workload scheduling and policy
Multi-tenancy model	Isolated per-customer control planes	Namespace/quota isolation within a shared cluster
Infrastructure	NVIDIA NICo (bare metal)	Any Kubernetes cluster
Node identity	TPM-based via SPIFFE/SPIRE	Not applicable
Artifact chain	SBOM and signed builds via NVKR	Vendor-managed

NVKE is the right choice when you need to offer isolated Kubernetes clusters to multiple customers from your own GPU hardware, build a verifiable artifact supply chain, or run a Kubernetes service without commercial licensing fees. The Apache 2.0 license means you own the full deployment.

NVIDIA Run:ai fits when you already have Kubernetes clusters running and need policy-based GPU scheduling, quota management, and multi-tenant workload isolation inside those clusters, with an enterprise support contract. Run:ai also supports a wider range of Kubernetes infrastructure than NVKE in its current preview state, which requires NICo for bare-metal provisioning.

Teams running AI workloads at scale often need both layers: cluster management (NVKE's domain) and workload scheduling within those clusters (Run:ai's domain). NVKE does not replace Run:ai's scheduling and policy features; Run:ai does not provision the Kubernetes clusters it schedules on.

install · quick start

Quick start#

bash

Install NVKE from source by building the management plane binaries with Go 1.27.3 and GNU Make 4.
```bash
git clone https://github.com/dsx-ai-factory/nvke.git
cd nvke
make build
```

tech stack · detected from GitHub

What it's built on#

Languages: GoPython
Infrastructure: Kubernetes

frequently asked

FAQ#

Is NVKE ready for production use?

No. NVKE is currently in preview. APIs, configuration, and behavior can change between releases without notice. The team explicitly recommends against production workloads. The management cluster has no upgrade path and no backup or restore procedure in preview. Review the Preview status and known limitations guide in the repository docs before planning a production timeline.

Does NVKE require NVIDIA hardware?

For real installations, yes. NVKE currently supports only NICo (NVIDIA Infra Controller) as its infrastructure provider, and NICo manages NVIDIA GPU bare-metal machines. Local evaluation and development run on any Linux host (amd64 or arm64) without GPU hardware, using mock-NICo and a local quickstart path that requires Docker and virtual TPM support.

What is the relationship between NVKE and NVKR?

NVKR (NVIDIA Kubernetes Recipes) is the companion build system. It builds the node OS images, binaries, container images, and Helm charts that NVKE deploys. NVKR produces signed, SBOM-recorded releases from pinned sources; NVKE deploys those artifacts into clusters. They are separate open-source repositories designed to work together as part of the DSX AI Factory stack.

How does NVKE differ from a standard kubeadm install?

kubeadm creates a single Kubernetes cluster from machines you prepare by hand. NVKE manages a fleet: a management cluster that provisions many isolated workload clusters on demand through an API. NVKE adds bare-metal provisioning via NICo, TPM-based node attestation, automatic node replacement with NVSentinel, and certificate renewal across the fleet, things kubeadm leaves to the operator.

What license does NVKE use?

Apache 2.0. You can run NVKE internally, modify it, include it in commercial products, and distribute it, with no licensing fee. Apache 2.0 requires preserving copyright notices and the license text; it does not require distributing source modifications when you run a modified version as a service (unlike AGPL). NVKR uses the same Apache 2.0 license.

also worth a look

Similar open-source tools#

FckSignups

InCustomer Support

Open-source tools that work instantly, no signup required

4,557 GitHub stars
Licence GPL-3.0

Pangolin

Alternative toCloudflare Zero Trust

Zero-trust network access, reverse proxy, and AI gateway, self-hosted.

23,346 GitHub stars

Portainer

Alternative toSUSE Rancher Prime

One self-hosted console for Docker, Kubernetes, Swarm, and Podman.

38,792 GitHub stars
Licence Zlib

Code Server

Alternative toGitHub Codespaces

VS Code in the browser, self-hosted, with no per-seat pricing.

79,785 GitHub stars
Licence MIT

Self Host

Alternative toEasypanel

CLI tool for 93 self-hosted apps with an agent-ready install workflow.

1,501 GitHub stars
Licence Apache-2.0

Homeio

Alternative toUnraid

Browser desktop for your self-hosted home server

103 GitHub stars
Licence AGPL-3.0

by category

Open source tools in these categories#

129

Self-host Nvke on a Hostinger VPS

Keep it running 24/7, even when your laptop is off. A Hostinger VPS gives it a server of its own.

01
Choose a plan
Starts from $6.49/mo
02
Apply
Extra 10% off yearly plans
03
Install Nvke
On a server that stays on
04
Start building
Focus on the app, not the server

Deploy on Hostinger

Inbox Zero

Alternative toSuperhuman

AI email assistant that drafts replies and clears your inbox

12,915 GitHub stars
Licence AGPL-3.0
#Email

rakazo

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

7,028 GitHub stars
Licence Apache-2.0
#AI Agents

SnitchFeed

Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.

Marketing & Growth

Advertise your product

The projects we add and every new guide. Free, unsubscribe any time.

Find open source alternatives to paid software

NVKE vs NVIDIA Run:ai

Feature	NVKE	NVIDIA Run:ai
License	Apache 2.0	Proprietary
Self-hosting	Yes, fully operator-owned	No: managed service
Primary function	Cluster provisioning and lifecycle	GPU workload scheduling and policy
Multi-tenancy model	Isolated per-customer control planes	Namespace/quota isolation within a shared cluster
Infrastructure	NVIDIA NICo (bare metal)	Any Kubernetes cluster
Node identity	TPM-based via SPIFFE/SPIRE	Not applicable
Artifact chain	SBOM and signed builds via NVKR	Vendor-managed

Who Nvke is for#

AI Factory Operators Offering Kubernetes as a Service

Skip if:

NVIDIA Cloud Partners Building Managed Inference Platforms

Skip if:

Platform Engineers Integrating Cluster API Components

Skip if:

Security-Focused GPU Cluster Operators

Skip if:

The problem it solves#

How it solves it#

API-Driven Cluster Provisioning

Workload Cluster Lifecycle Management

TPM-Based Node Identity and Encryption

Verifiable Build Artifact Chain (NVKR)

OpenTelemetry Fleet Telemetry (NKTP)

CNCF-Conformant Kubernetes Clusters

Strengths and trade-offs#

Strengths

Trade-offs

Nvke vs alternatives#

NVKE vs NVIDIA Run:ai

Quick start#

What it's built on#

FAQ#

Similar open-source tools#

Open source tools in these categories#

Cloud & Hosting

DevOps & CI/CD

IT Management

AI & Machine Learning

Find open source alternatives to paid software

Who Nvke is for#

AI Factory Operators Offering Kubernetes as a Service

Skip if:

NVIDIA Cloud Partners Building Managed Inference Platforms

Skip if:

Platform Engineers Integrating Cluster API Components

Skip if:

Security-Focused GPU Cluster Operators

Skip if:

The problem it solves#

How it solves it#

API-Driven Cluster Provisioning

Workload Cluster Lifecycle Management

TPM-Based Node Identity and Encryption

Verifiable Build Artifact Chain (NVKR)

OpenTelemetry Fleet Telemetry (NKTP)

CNCF-Conformant Kubernetes Clusters

Strengths and trade-offs#

Strengths

Trade-offs

Nvke vs alternatives#

NVKE vs NVIDIA Run:ai

Quick start#

What it's built on#

FAQ#

Similar open-source tools#

Open source tools in these categories#

Cloud & Hosting

DevOps & CI/CD

IT Management

AI & Machine Learning

Find open source alternatives to paid software