Agents
nohup doesn't ship a model — it drives external ACP agents that you install and log into. Two are configured out of the box.
| Agent | Launch command | Auth |
|---|---|---|
| Claude Code | claude-code-acp | your Claude subscription |
| Codex | codex-acp (-c sandbox_mode=workspace-write …) | your ChatGPT/Codex subscription |
Install & log in
The agents are Node CLIs. Install them on the machine running nohup (or in the Docker image — the provided image attempts this for you):
npm i -g claude-code-acp
npm i -g codex-acp
Then log in once with each tool so its credentials are stored:
claude # then /login
codex login
This writes to ~/.claude and ~/.codex. In Docker, mount those directories so the agent inside
the container is authenticated:
volumes:
- ${HOME}/.claude:/root/.claude:ro
- ${HOME}/.codex:/root/.codex:ro
Configure in the UI
Open Settings → Agents to change an agent's launch command, arguments, or enable/disable it.
The command must be on the server's PATH.
nohup runs each session in an isolated git worktree and auto-approves the agent's tool prompts, so
Codex is seeded with sandbox_mode=workspace-write to let it work freely inside that worktree.