What you will verify
You will configure one Agent, pass its environment check, and run a small Chat task. The real task matters: a successful test proves that Rudder found the runtime, but not that the runtime can complete your work.Before you start
Install and authenticate one supported runtime on the same computer as Rudder:
Run the command once in your terminal to confirm it opens without a missing
command or login error.
Connect it to an Agent
- Open the Agent, then edit its Runtime configuration.
- Choose the local runtime you installed.
- Choose a model and thinking level when that runtime offers them.
- Leave advanced permissions, sandbox bypass, custom commands, and instruction-file paths at their defaults unless your task needs them.
- Select Test runtime chain.
- Read the result:
pass: Rudder can start the runtime.warn: setup may work, but read the named limitation.fail: fix the named command, login, model, path, or network problem.
- Save the Agent.
- In Chat, ask for a safe, concrete result such as a two-line summary.
- Open the Agent Run and confirm the selected runtime produced the reply.
Check the result
The setup is ready when the environment test passes, or has a warning you understand, and the real Chat task finishes. The Run should show the intended runtime plus a result or clear blocker.Common problems
- Command not found: install the CLI or set the correct supported command.
- Authentication failed: sign in through the runtime outside Rudder, then test again.
- Model unavailable: choose a model your provider account can use.
- Instructions or workspace cannot be read: use an absolute path and fix host permissions.
- The test passes but the task fails: open the Run and fix the specific provider, model, permission, or project-access error.
