Skip to main content
The catalog examples reproduce agent and service installations as editable manifests. These examples require v0.13 and explicitly select image deployment.

​
Choose an execution model

CLI examples declare kind: cli: Claude, Codex, OpenCode, Junie, Kilo Code, and Pi. They install an executable for use inside the workspace. They do not expose a port or have an HTTP health check. Service examples expose a port and health endpoint: Jupyter, Marimo, Streamlit, pgweb, Excalidraw, File Browser, OpenTerminal, Firecrawl, Hermes, T3 Code, Open WebUI, and OpenClaw Gateway. Read the selected README for provider credentials and resource needs.

​
Deploy from your machine

After installing the CLI, run:
These examples use reproducible: true and the image strategy. Retain the printed workspace/app IDs. Incremental CLI deployments have executable verification requirements; this tutorial uses image installation explicitly.

​
Verify inside the workspace

Connect using SSH access, then run the installed executable’s version/help command, such as codex --version. Installation success does not establish provider authentication. Configure managed AI or the provider credentials required by your selected tool before making an AI request. For an HTTP service, deploy from its catalog-apps/service/NAME directory instead, inspect rig app health --app APP_ID, and open the reported URL. Check the manifest’s visibility before sharing; do not assume a product’s own authentication is enabled. If a CLI installation fails, inspect rig app logs --app APP_ID --install and the image build logs. Do not debug a CLI recipe by adding a fake HTTP port.

​
Clean up

Export workspace files you need, then remove only the disposable workspace:
For a managed installation instead of editing a recipe, see Catalog apps.