Skip to main content
Deployment is optional. An agent can work on code without a cluster or hosted application. When you want to deploy, you choose the destination: a Docker host, a VPS, Kubernetes, or another service reachable from the selected worker.

Prepare access

  1. Install the destination’s CLI and project tooling on the worker.
  2. Verify network access from that worker to the intended destination.
  3. Have the installation operator configure scoped account or project credential bundles for the intended execution environment.
  4. Describe the destination, intended account/context, namespace or application conventions, and deployment procedure in account instructions or the project’s AGENTS.md. Refer to credential names and file paths without including values.
  5. Ask the agent to inspect the destination and verify the intended target before making deployment changes.
For Kubernetes, a kubeconfig can be delivered through a scoped credential bundle as a protected secret file whose path is exposed through KUBECONFIG. It should not be pasted into a chat, committed to a repository, or assembled by having an agent read a secret’s value. Validate access with a read-only command such as kubectl get namespaces after choosing the intended context.

Networking, domains, registry access, and storage

Your deployment provider determines ingress configuration, public addresses, TLS, available storage classes, quotas, and application persistence. There is no required r5d registry, Kubernetes runtime, storage class, or public hostname. Use the provider’s supported configuration and back up your application data separately from workspace synchronization. Use GitHub Container Registry to build and publish images. Worker push credentials, deployment pull credentials, and CI deployment access are separate concerns. Configure CI credentials or workload identity for your destination, and maintain their rotation and revocation through that provider.