Run workloads and keep their results connected
Run training jobs, model services, and interactive sessions on your Kubernetes clusters. Manage shared compute, track experiments, automate pipelines, and version the outputs teams reuse.
Product capabilities
GPU scheduling
Set priorities and resource limits for training, notebooks, sandboxes, and batch jobs on your Kubernetes clusters.
Compute clusters
Submit jobs and services through one control plane, with queues that route work to the Kubernetes clusters and namespaces you operate.
Experiment tracking
Record parameters, metrics, logs, and artifacts together, whether your experiment runs on a laptop, in a notebook, or as a managed Kubernetes job.
Hyperparameter tuning
Define a search space, run parallel trials on your Kubernetes clusters, and compare parameters, metrics, and artifacts before selecting a candidate.
Pipelines and automation
Define dependencies, pass outputs between jobs, and control when a workflow continues—with schedules, conditions, and approval steps.
Model and artifact registry
Register selected outputs with their source runs, manage lifecycle stages, and give downstream jobs and services a named version to load.
Model serving
Package a model server as a Polyaxon service, choose its resources and model inputs, and inspect its status and logs alongside the rest of your workloads.
Notebooks & apps
Launch JupyterLab, TensorBoard, and custom web apps on your Kubernetes compute. Choose the image, data connections, and CPU or GPU resources; open each session from its Polyaxon run.
Sandboxes
Give agents and developers command, file, and terminal access to a service on your Kubernetes cluster. Choose the image, resources, storage, and connections; control the session with Python or the CLI.
Choose where the platform runs
With Polyaxon Cloud, Polyaxon manages the control plane and your workloads run on the Kubernetes compute clusters you connect. With self-hosted Enterprise, you also operate the control plane. Community Edition provides a self-hosted runtime with core tracking and experimentation features.
Queues, connected-cluster management, and the managed workflow engine belong to the commercial offering. Each capability page explains its scope; the pricing pages list plan limits and access controls.
Start with the work you need to do
See how the capabilities fit together for sharing GPUs, training and evaluating models, serving predictions, processing datasets, and executing agent code. The use-case pages describe the workflow; the documentation and integrations cover its configuration.