Pipelines on a cluster
Training and serving steps run as jobs on Kubernetes.
We have no published Kubeflow build. We have shipped a Docker multi-stage deployment and we have no published Kubernetes build, so we say plainly that a Kubeflow job starts with a cluster.
Training and serving steps run as jobs on Kubernetes.
Multi-stage Docker builds, as on TruckFleet.
Runs and results recorded, with MLflow where it fits.
Moves onto the current Community Distribution.
Illustration. A dashed outline marks where the technology would sit. We have no published build on it.
Source: Kubeflow documentation, Releases, kubeflow.org/docs/releases, checked 6 November 2026.
A FastAPI back end deployed with a Docker multi-stage build and Docker Compose.
Kubernetes
MLflow
Apache Airflow
Docker
PyTorch
TensorFlowThe releases page lists Kubeflow Community Distribution 26.03 first. It gives no release date and no cadence.
No. We say so on the first call and show the nearest work, which is a Docker multi-stage deployment, with no published Kubernetes build.
A banner on its releases page says Kubeflow is now a CNCF graduated project.
You do, from the first commit. Repositories and cloud accounts stay in your name.
Thirty minutes on your cluster, your pipelines and your models. No sales pitch.