OpenWorkerDeveloper
DocumentationHarness SDK

GCP

The GCP reference targets GKE Standard, Cloud SQL PostgreSQL, Cloud Storage and Artifact Registry, with NVIDIA OpenShell providing tool sandboxes.

The application uses the same OpenWorker service API and session model as other deployments.

Setup flow

  1. Select a project, region, billing/budget boundary and deployment lifetime.
  2. Review the separate setup, executor, workload and cleanup identities.
  3. Run bootstrap/preflight checks and review the generated infrastructure plan.
  4. Establish independent cleanup before creating the paid workload.
  5. Qualify the cluster, private installer route, OpenShell, agent execution and cleanup.

Use scoped service-account impersonation and workload identity instead of copying administrator credentials into pods. Validate the chosen node OS/kernel and networking with OpenShell; do not assume that a different GKE mode is interchangeable.

This deployment path is experimental. A production deployment requires validation of live GCP permissions, OpenShell compatibility and independent cleanup in your project.

Start with the service CLI and deployment options.

Search guides, examples and deployment instructions.