This path uses an installed Claude Code or Codex CLI. That vendor CLI keeps its own sign-in; Maxxwell does not copy the credential into a separate service.

​
1. Install the CLI

On macOS or Linux x86_64:
If you set MAXXWELL_INSTALL_DIR, add that directory to PATH instead.

​
2. Check your available runtimes

Install and sign in to Claude Code or Codex using that tool’s own setup, then ask Maxxwell what it can see:
If both are installed, save the one you want Maxxwell to use:
This saves a runtime name, not a credential. Run maxxwell runtime again at any time to inspect the selection.

​
3. Start an interactive session

Run Maxxwell from the project directory the agent should work in:
Bare maxxwell opens one interactive conversation. Type help or ? inside the session for local help; it does not send a paid model request. Type exit, quit, or press Ctrl-D to leave.
The current release keeps the conversation in the running process. Closing the process ends that interactive conversation; it is not resumable after a restart.

​
4. Open the desktop app or fleet surface

  • Run maxxwell ide to open the desktop surface.
  • Set MAXXWELL_FLEET_DIR=/absolute/path/to/fleet and run maxxwell fleet for the read-only terminal fleet view.
  • Run maxxwell --help for the complete command list in your installed build.
The fleet view reads existing records. It does not create a fleet or prove that an automatic supervisor is running.