Platform Architecture

The managed platform has a hosted control plane and an agent in your Kubernetes cluster. The control plane accepts run definitions, assigns work, and serves the dashboard. The agent connects outward to collect assignments, starts the workloads as Kubernetes resources, and reports their status.

The control plane does not need a kubeconfig or inbound access to your cluster. The agent needs outbound access to the control plane. Evaluation Jobs and Pods use resources in your cluster, including GPUs where required. During early access, image versions, resource requests, and object storage are set up with each team.

From a run to workloads

A run names models and tasks. The platform creates one workload for each model–task pair; a try at that workload is an attempt. It derives the model servers, environment workers, and runtime processes needed to execute them. Use the dashboard to start and inspect a run. Upload a Model or Environment covers the path from a local container to the catalog.

Two paths for results

Live episode results and logs flow from the cluster to the dashboard. The agent buffers them for at-least-once delivery, so a connection interruption can delay the live view.

Separately, each attempt writes a manifest and artifacts to object storage. The control plane reads the manifest to finalize the workload. If the live view and stored artifacts disagree, the artifacts are authoritative.