Roadmap

Kale is under active development again after a hiatus, with a re-energized maintainer team and a growing roster of contributors. This page captures the current direction in broad strokes — it is intentionally not a detailed spec. Things will move around as the community gives feedback.

Where we are

Kale v2.0 is the headline release for this cycle. It brings the project back in sync with the Kubeflow ecosystem:

  • Full compatibility with Kubeflow Pipelines v2 (the KFP v2 DSL and YAML IR, artifact model, and component spec).

  • A modernized JupyterLab 4.x extension rewritten for the current labextension API.

  • Python 3.11+ support across the backend.

  • A cleaner, testable compiler pipeline with golden-file fixtures for the generated KFP DSL.

Tracking issue: kubeflow/kale#457 — Road to 2.0.

What we’re focused on now

GSoC 2026: composable notebooks

Kale is participating in Google Summer of Code 2026 under the Kubeflow umbrella. The accepted project focuses on multi-notebook coordination and composable pipelines — letting users build pipelines that span more than one notebook, and composing them into larger workflows. This is a concrete, community-driven effort landing in Kale over the coming months, and it shapes a lot of the near-term priorities below.

If you want to contribute, the GSoC project is a great way to get involved. Subscribe to the GitHub milestone for that work, drop a note in the ML Experience WG meeting, or ping us on Slack.

Directional themes

These are high-level directions the maintainers are aligning around. None of them are committed features yet; they indicate where we expect Kale to go, not a fixed schedule.

Deeper notebook experience

The notebook is where Kale differentiates, and we want to make that experience richer:

  • Incremental execution — compile and run only the parts of a pipeline that changed since the last run, without recomputing upstream steps.

  • Dependency visualization — surface Kale’s DAG view directly in the side panel so you can see the pipeline shape while editing.

  • Run sweeps and experiment comparison — easy fan-out across parameter grids, with side-by-side metrics comparisons in the KFP UI.

From development to production

  • Local execution mode — run a Kale pipeline end-to-end on the local machine for fast feedback, without a Kubernetes cluster in the loop.

  • Kubeflow SDK integration — tighter alignment with the emerging Kubeflow SDK, so that Kale-generated pipelines can interop with hand-written SDK pipelines and components.

  • Artifact management and model registry — better default support for typed artifacts, model cards, and registries so that pipelines can hand off to serving and evaluation infrastructure without glue code.

Community and ecosystem

  • Documentation — that’s what this site is about. Expect the docs to grow in scope as features land.

  • Contributor onboarding — simpler setup, better make targets, clearer contribution docs, more “good first issue” labeling.

  • Ecosystem alignment — closer collaboration with the Kubeflow ML Experience WG, KFP, Notebooks, and Katib maintainers.

There’s much more to come

The maintainers are happy to say: this is the most active Kale has been in years, and we’re treating the roadmap as a living document. Expect more detail to appear on this page as concrete designs emerge from the WG.

Get involved in shaping the roadmap

  • GitHub issues and discussions — file issues for bugs, feature requests, and design discussions. Label suggestions welcome.

  • Milestones and project boards — see the milestones page for in-flight work.

  • ML Experience Working Group meetings — bi-weekly on the Kubeflow community calendar. Kale roadmap updates are a recurring agenda item.

  • Slack — #kubeflow-ml-experience on the Kubeflow Slack workspace.

The best way to influence the roadmap is to show up with a use case.