Installation¶

This page walks you through the prerequisites, installation options, and a quick sanity check for a working Kale setup.

Prerequisites¶

Requirement	Version	Notes
Python	3.11 or later	Kale uses modern typing features; older Pythons are not supported.
Kubeflow Pipelines	v2.16.0+	Older KFP servers reject the `securityContext` field Kale emits for step pods.
Kubernetes cluster	any	`minikube`, `kind`, Docker Desktop, or a managed cluster all work.
JupyterLab	4.0+ (for the UI)	Only required if you want to use the Kale side panel inside JupyterLab.

Install KFP by following the official Kubeflow Pipelines installation guide. Make sure to set PIPELINE_VERSION=2.16.0 or later. If you are upgrading an older environment, ensure your Python deps include kfp[kubernetes]>=2.16.0.

Install from PyPI¶

Kale v2.0 pre-release

Kale v2.0 is available on PyPI as a release candidate. Install it with the --pre flag until the final release is tagged.

pip install --pre "jupyterlab>=4.0.0" "kubeflow-kale[jupyter]"
jupyter lab

Install from source¶

Clone the repository and use the make targets provided by the project:

git clone https://github.com/kubeflow/kale.git
cd kale

make dev      # Install Python + Node deps, build and link the labextension
make jupyter  # Start JupyterLab with the Kale panel

make dev takes care of:

Installing uv if it is not already present.
Running uv sync --all-extras to create the virtual environment and install the Python package in editable mode.
Building the JupyterLab 4 extension and linking it so it appears in your JupyterLab UI.
Installing pre-commit hooks for linting.

See Contributing for a full breakdown of the available make targets and the development workflow.

Try Kale in Docker (no cluster required)¶

If you just want to play with Kale in an isolated environment, the repo ships with a Dockerfile based on the official Kubeflow jupyter-scipy image:

make docker-build   # Build Kale wheels and bake them into the image
make docker-run     # Start JupyterLab on http://localhost:8889

To also connect to a KFP cluster, follow the multi-terminal setup in Running Pipelines.

Verify your installation¶

Start a cluster and KFP port-forward (minikube example):

minikube start
kubectl port-forward -n kubeflow svc/ml-pipeline-ui 8080:80

Test the CLI against an example notebook:
```
kale --nb examples/base/candies_sharing.ipynb \
     --kfp_host http://127.0.0.1:8080 \
     --run_pipeline
```
This compiles the notebook into .kale/<name>.kale.py, uploads the pipeline, and starts a run.
Test the JupyterLab extension:
- Open JupyterLab (make jupyter or jupyter lab).
- Open any notebook from examples/base/.
- Click the Kale icon in the left sidebar.
- Toggle the Kale panel on — you should see cell type dropdowns appear on each notebook cell.

If any of these steps fail, head to Troubleshooting — the most common issues are covered there.