Connect a cluster without changing its purpose
Install Polyaxon Agent (compute cluster) in the Kubernetes environment where you want workloads to run. It checks the queues assigned to it and manages the operations dispatched there, including jobs, distributed jobs, and services.
A single control plane can connect to multiple compute clusters. You can keep on-premises training capacity and a cloud batch environment separate while managing their runs through Polyaxon.
Make the destination explicit
Queues connect workload submission to an execution destination. Choose a queue on an operation, apply it through a preset, or set a project default. Project restrictions determine which queues a team can use.
The destination still needs the right container images, drivers, resource capacity, and network access. Moving a workload to another queue does not copy its storage or make its dependencies available on another cluster.
Queue routing and restrictions · Reusable scheduling presets
Configure storage and access per cluster
Each deployment has an artifacts store for run outputs and logs, plus a catalog of connections for resources such as data volumes, object storage, Git repositories, and container registries. Clusters can use different catalogs or share access to selected resources.
Configure the credentials and permissions at each destination. A run can refer to a named connection without embedding credentials in its training component, but the connection must exist and be authorized where that run executes.
Keep infrastructure ownership clear
Polyaxon manages operation metadata, queue dispatch, and workload lifecycle through the connected compute cluster. Kubernetes schedules the resulting workloads on nodes. Your infrastructure team maintains capacity, drivers, networking, storage, and any additional operators the workloads require.
With Polyaxon Cloud, the control plane is managed for you. With self-hosted Enterprise, your team operates it too. Connecting a cluster does not provision cloud GPUs, configure cross-cluster failover, or replace your backup and upgrade procedures.