What is Kubeflow?

Kubeflow is the open source machine learning toolkit on top of Kubernetes.

Kubernetes is the industry standard for software delivery at scale and Kubeflow provides the cloud-native interface between K8s and data science tools - libraries, frameworks, pipelines, notebooks - bringing the Ops to ML.

Try Kubeflow Kubeflow operators

Contributors to Kubeflow

What's inside Kubeflow?

Kubeflow dashboard

Multi-user dashboard with role-based access control (RBAC) allows data scientists to focus on creating great models, while the ops team takes care of the backend.

Jupyter notebooks

The notebook of choice for data scientists. Quickly spin-up Jupyter notebook servers from the Kubeflow dashboard, with allocated GPUs and storage.

Kubeflow pipelines

Pipelines map dependencies between components in the ML workflow where each component is a containerised piece of ML code. Learn more ›


Kubeflow started as part of TensorFlow Extended (TFX) and naturally, it includes TensorFlow training, TensorFlow serving and even TensorBoard.

ML libraries & frameworks

For training - PyTorch Training, MXNet XGBoost MPI for distributed training and scikit-learn For model serving - Seldon Core, KFserving and more.

Experiment tracking

Every time you run a Kubeflow pipeline with specific parameters the results of this run are stored so they can be easily compared and replicated.

Hyperparameter tuning

Kubeflow includes Katib for hyperparameter tuning. Katib runs pipelines with different hyperparameters (e.g. learning rate, # of hidden layers) optimizing for the best ML model.


Save, compare and share generated artifacts - models, images, plots - through the feature store. Connect with Prometheus for monitoring and alerting. Integrate with Pachyderm for data versioning.

Kubeflow, well packaged

Charmed Kubeflow integrates the 30+ apps that make up Kubeflow with ops code to provide the best Kubeflow experience, from deployment to day-2 operations.

Visit Charmed-Kubeflow.io

Why MLOps?

Bringing AI solutions to market can involve many steps: data pre-processing, training, model deployment or inference serving at scale... The list of tasks is complex and keeping them in a set of notebooks or scripts is hard to maintain, share and collaborate on, leading to inefficient processes.

In the study, Hidden Technical Debt in Machine Learning Systems, Google describes that only about 20% of the effort and code required to bring AI systems to production is the development of ML code, while the remaining is operations. Standardizing ops in your ML workflows can hence greatly decrease time-to-market and costs for your AI solutions.

Area = effort & code

Who uses Kubeflow?

Thousands of companies have chosen Kubeflow for their AI/ML stack.

From research institutions like CERN, to transport and logistics companies - Uber, Lyft, GoJek - to financial and media industries with Spotify, Bloomberg, Shopify and PayPal.

Forward-looking enterprises are using Kubeflow to empower their data scientists.

Get started today

Try-out Kubeflow on your K8s deployment. Or on MicroK8s - Zero-ops Kubernetes with high availability.

Single-command deploy locally on your desktop, public cloud VM or on-prem server.

Try Kubeflow on any K8s Kubeflow on MicroK8s