Kale¶
From Jupyter Notebook to Kubeflow Pipeline — Zero Boilerplate
Kale (Kubeflow Automated pipeLines Engine) turns annotated Jupyter notebooks into production-ready Kubeflow Pipelines without requiring you to write a single line of KFP SDK code.
Tag cells in your notebook, let Kale figure out the data dependencies between them, and compile the whole thing into a KFP v2 pipeline you can run on any Kubeflow cluster.
Why Kale?¶
Annotate cells, compile, run. Kale generates the KFP v2 DSL for you — no need to learn components, artifacts, or Python decorators.
Variables flow between steps as if you were still in a single notebook. Kale’s type-aware marshalling handles numpy, pandas, scikit-learn, PyTorch, Keras, TensorFlow, XGBoost and more.
Tag cells visually, define step dependencies, and submit pipelines from the Kale side panel inside JupyterLab 4.
Compiles to the modern KFP v2 pipeline DSL with full artifact support. Runs on any compliant Kubeflow Pipelines backend.
Get started¶
Compile and run your first notebook on Kubeflow Pipelines in a few minutes.
Understand cell annotations, data marshalling, and how Kale compiles to KFP.
Practical walkthroughs for annotating, parameterizing, and running pipelines.
Python API for the Pipeline, Step, Compiler and marshalling modules.
Kale in the Kubeflow ecosystem¶
Kale is part of the Kubeflow ML Experience Working Group, alongside the Kubeflow SDK, Kubeflow Pipelines and Kubeflow Notebooks. It lives at the notebook layer — where data scientists prototype — and bridges the gap to the pipeline layer, where production workloads run.
If KFP is the “how” of running ML pipelines on Kubernetes, Kale is the “what you meant”: take the notebook you already have, and turn it into a pipeline without rewriting anything.
Community¶
GitHub: kubeflow/kale
Slack: #kubeflow-ml-experience on the Kubeflow workspace
Working group: ML Experience WG meetings — see the Kubeflow community calendar
Issues & feature requests: github.com/kubeflow/kale/issues