AI Coding Agents
Four AI coding agents are pre-installed. Each is started with its own command:
| Agent | Command | Version | Official site |
|---|---|---|---|
| Claude Code | claude | 2.1.283 | claude.com/claude-code |
| Codex | codex | 0.158.0 | github.com/openai/codex |
| Gemini CLI | gemini | 0.61.0 | github.com/google-gemini/gemini-cli |
| Grok | grok | 1.0.41 | x.ai |
Signing in
Run the command in /workspace (your project). On first run, each agent walks you through signing in with its own account or API key. Follow its prompts; the details belong to each vendor, so refer to their official documentation above.
Each agent stores its login and settings in your home directory inside the container:
| Agent | Location |
|---|---|
| Claude Code | ~/.claude/ and ~/.claude.json |
| Codex | ~/.codex/ |
| Gemini CLI | ~/.gemini/ |
| Grok | ~/.grok/ |
These are removed together with the container. To stay signed in across restarts, keep these directories in Docker volumes, as described in Data & Persistence.
Tips
- Run several agents side by side in tmux panes. tmux highlights windows with new output, so you can see which agent needs attention. See tmux.
- Get notified and approve remotely with yeci-notify: agents can send you a Telegram message when they finish or need permission.
- Memory:
NODE_OPTIONSis set to--max-old-space-size=4096, so Node-based agents have enough memory for large tasks. - Versions: agent versions are fixed per image. To get newer versions, pull a newer image (see Images & Tags) instead of updating inside the container, where changes are lost when the container is removed.