ZenML

ZenML is an open-source MLOps framework that lets teams build portable ML pipelines in pure Python — @pipeline and @step decorators — and run them on any stack: local processes, Kubernetes, Airflow, or SageMaker, switching backends without rewriting code. A stack abstraction separates pipeline logic from infrastructure components like orchestrators, artifact stores, and experiment trackers.

ZenML emphasizes governance for enterprise ML: audit trails, step caching, artifact lineage, model versioning, and a Model Control Plane for champion/challenger rollouts and one-command rollbacks. The core is Apache 2.0, and it composes with MLflow, Weights & Biases, Kubeflow, and KServe rather than replacing them.

Limitations: the abstraction layer adds concepts to learn versus running Airflow directly; the ZenML server is another component to operate for teams wanting the dashboard and lineage UI; and very large-scale feature-engineering workloads may still need dedicated data infrastructure underneath.

pip install zenml; zenml init; define @step functions and a @pipeline, then python run.py — switch stacks with zenml stack set.