CLI commands¶

Commands listed here are the ones Arena shows in the UI (Profile and Agents deployment snippets). Install the Python CLI package first, then authenticate with Authentication.

Authentication¶

Command	Shown on
`arena login`	Agents Arena CLI tab (preamble), Profile (alternative to a PAT)

arena login

Run once per machine (or again after session expiry). Inference snippets assume you are already logged in when they say so in the comment line.

Set ARENA_API_KEY instead when you cannot open a browser. Profile does not print a separate arena subcommand for keys; create the PAT in the UI and export the variable.

Inference¶

Shown on the Agents page under Deployment code → Arena CLI, after a deployment is Deployed.

Classic RL: run a step¶

Gym, PettingZoo, and recurrent PPO deployments use the same command shape. Replace the deployment name and observation placeholder with your values.

# Make sure you're logged in with `arena login` before running this command.

arena inference run YOUR_DEPLOYMENT_NAME --obs '<YOUR_OBS_HERE>'

YOUR_DEPLOYMENT_NAME: deployment name from the Agents list (same string as in the Arena client tab).
--obs: observation payload the CLI sends for one get_action-style call. Format matches what the Manual HTTP tab serializes; copy the pattern from the UI snippet for your env type.

For a full rollout loop, use the Arena client tab (ArenaClient + open_inference_agent) or wire Manual HTTP yourself. The CLI tab is a minimal one-shot example.

LLM: list deployment metadata¶

Advanced (LLM) deployments do not use arena inference run in the UI. They show:

arena inference list --name YOUR_DEPLOYMENT_NAME

Use this to inspect deployment metadata from the terminal. For generate calls, switch to the Arena client or Manual HTTP tab on the same panel.

Projects, experiments, environments¶

The Arena web app does not currently show terminal commands for creating projects, submitting training jobs, or uploading custom environments. Do that work in the UI: