Kubeflow Pipelines is the machine-learning workflow component of the Kubeflow platform. It lets you build, deploy, and manage portable, scalable ML pipelines on Kubernetes, from data preparation through training to serving.
Pipelines are authored in Python with the Kubeflow Pipelines SDK and compiled into workflow definitions that run on Kubernetes, so the same pipeline can move between clusters and clouds. The web UI provides experiment tracking, run comparison, and visualization of each step’s inputs, outputs, and metrics, while compiled components can be shared and reused across teams. Step outputs are cached so unchanged stages are skipped on re-runs, and artifact lineage is recorded through ML metadata. Kubeflow Pipelines is one of the most widely adopted open-source MLOps orchestration tools, is released under the Apache-2.0 license, and has over 4,200 stars on GitHub. It remains under active development within the Kubeflow community, with updates continuing through 2026.
Kubeflow Pipelines is free and open source under the Apache-2.0 license. It runs on your own Kubernetes infrastructure, and there is no paid edition or per-run fee.
MLOps engineers and ML platform teams who need reproducible, auditable machine-learning workflows on Kubernetes, from experimentation to production training.
Kubeflow Pipelines can be deployed on its own or as part of a full Kubeflow installation, and managed distributions are offered by major cloud and platform vendors. Its SDK supports both lightweight Python components and fully containerized components, and pipelines can be triggered from notebooks, CI systems, or schedules, which has made it a common backbone for production machine-learning platforms that need repeatable, auditable training workflows.