Easy Kubeflow operations

Pure open source Python operators put Kubeflow on rails, for bare metal, VMware and multi-cloud or edge.

Composable operators for data scientists to stand up and integrate the Kubeflow applications they need, on a laptop, workstation or cluster. Upgrades and security updates all supported in the free version.

Install in 5 minutes Watch Operations Guide

Public cloud

  • GPU acceleration
  • Amazon EKS
  • Google GKE
  • Azure AKS


  • Runs on any Kubernetes
  • Bare metal deployment
  • NVIDIA GPU autoconfig
  • Included in MicroK8s


  • Lightweight config
  • Windows
  • macOS
  • Ubuntu


  • Inference-only config
  • Distributed training
  • Fast-clustering
  • Optimised for x86

Industrial MLops from lab to enterprise

Professionalise data science with industrial data pipelines

Experiment with Kubeflow on MicroK8s, from desktop to large scale. Highly available with 3+ nodes.

Unleash your data scientists with the latest ML tools in a single dashboard. All-in-one toolkit, one-time-config.

No more siloed notebooks and scripts, reproducible and shareable AI workflows with Kubeflow pipelines.

Kubeflow, well packaged

Charms wrap the 30+ apps that make up Kubeflow with ops code.

Charmed Kubeflow integrates these charms to provide the best Kubeflow experience, from deployment to day-2 operations.

Automatic GPU acceleration

Detect, configure and accelerate all your machine learning

Accelerate your AI to market. Pass GPUs to your ML pipeline components, for deep learning dev and test.

Train faster on the latest silicon. Charmed Kubeflow will automagically detect, configure and use available GPUs.

On desktop, rack, or any public cloud.

Data lake integration

Kafka, Spark, Cassandra. Turn data into predictions.

Bring the analytics power of Kubeflow to your data lake, with seamless integration.

Looking to modernise your MLOps, but still running on Hadoop? Need to integrate with Spark or Kafka? We’ll connect the dots.

The right community driven proxy-charms model your current data solutions and integrate them with Kubeflow.

All–you–need Kubeflow services

Enterprise support, deployment, training and fully managed Kubeflow

Rely on 24/7 enterprise support with the SLAs you need.

Provide specialised MLOps training to your sysadmins, devops engineers and data scientists. Any infra and level of expertise, tailored to your data.

Off-load the complexity of deployment and management of Kubeflow to Canonical engineers.

Scale data pipelines to thousands of jobs

Offer data scientists a platform for large-scale continuous analytics

With Kubernetes at the core, scaling your ML compute with high availability is painless.

From experimentation with MicroK8s on desktop or one cloud node, to large scale multi-cloud Kubernetes.

Multi-cloud Kubeflow delivers the elasticity of the public clouds and the cost-effective compute of on-prem.

Deploy Kubeflow at scale

Open source operators in Python

Small, composable, model-driven Kubernetes charms

Make use of open source integration code and compound the benefit of 100+ available operators.

Charms make deploying and managing Kubeflow simple to your IT ops team. No hard-coded services, evergreen operations.

Engineered for flexibility around complex scenarios, charms allow you to tailor Kubeflow to your needs with custom integrations.

Kubeflow charms and models

Public cloud optimisation

Fast-deploy on every major public cloud with GPU acceleration

We work with Amazon, Microsoft, Google, Oracle and IBM to simplify multi-cloud GPU accelerated workload development.

Easy setup on public cloud Kubernetes for a multi-cloud Kubeflow. Get the elasticity of the public clouds and the cost-effective compute of on-prem.

On-rails deployment and management with Python operators on public cloud Kubernetes services — AKS, EKS, GKE.

Kubeflow on multi-cloud K8s

Lightweight workstation deployments

Run Kubeflow locally. No resource limits.

With the expansion of Kubeflow as a full ML toolkit, running it on desktops with up to 16Gb of RAM has become a painful experience.

Kubeflow-lite is a lightweight version of Kubeflow with the ML services you typically need on your desktop, on a smooth experience.

Install Kubeflow lite Learn more about Kubeflow-lite ›

Bare metal Kubeflow operations

Deploy and manage on-rails. No YAML-fu.

We work with bare metal vendors like Dell, Lenovo and Supermicro to bring the best out-of-the-box Kubeflow experience on-prem.

Enjoy a public cloud like experience on-prem with Metal As A Service.

Charms integrate your networking and storage of choice with logging, monitoring and alerting solutions, Kubernetes and Kubeflow.

Take Kubeflow to the edge

Well-integrated Kubeflow components for IoT devices

Install Kubeflow-edge, a lightweight ML toolkit for inference and distributed training at the edge, with one command.

Alternatively, build your custom Kubeflow bundle with the charmed apps you need, well integrated.

On MicroK8s or any conformant Kubernetes.

Install Kubeflow edge Learn about Kubeflow-edge ›

The team behind Charmed Kubeflow

Charmed Kubeflow is built by the Kubernetes team at Canonical. Besides Kubeflow, this team delivers two K8s products. MicroK8s is on-rails and opinionated, for the zero-ops experience. Charmed Kubernetes for all the permutations of Kubernetes components.

Need help?

Get in touch with one of our engineers.

Contact us