Agents on a worker
How the worker detects agents, enables or disables them, and which settings each agent takes.
The worker looks for each supported agent's executable on the worker user's PATH — claude, codex,
gemini, opencode, aider — and reports what it finds to the control plane.
| Agent | Executable | Page |
|---|---|---|
| Claude Code | claude | Claude Code |
| OpenAI Codex CLI | codex | Codex |
| Gemini CLI | gemini | Gemini CLI |
| OpenCode | opencode | OpenCode |
| Aider | aider | Aider |
Enable and disable
Worker UI → Agents: turn detected agents on or off for this machine. Policies can also allow, block or prefer agents per organization, project or task.
Sign in once, as the worker user
Agents use their own login by default. Sign in once from a terminal as the user the worker runs as. If an agent is not signed in, tasks that land on it end with authentication required, and the worker tries another compatible agent first.
Test agent
A deterministic mock agent exists for testing the platform itself; enable it per worker. It is not a coding agent and is not listed in the directory.