Skip to content

Finish Runner Setup

Open the WebUI, go to Settings, and use Runner setup to finish the part that is easiest to manage in one screen.

This field is required for real remote work.

The shell prelude runs before every remote command, so it should contain the environment setup that your non-interactive SSH sessions need. Your prelude would probably look roughly like this:

Terminal window
export NVM_DIR="$HOME/.nvm"
[ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"
. "$HOME/.cargo/env"
export PATH="$PATH:/home/<your-user>/.local/bin"

Keep it focused on PATH and toolchain setup. If it prints extra text, remote command parsing becomes much less reliable.

Choose the default runner for new work:

  • Codex if you want new task dispatches and PR reviews to default to Codex
  • Claude if your remote machine is set up for Claude-first runs

This is only the default. You can still override it later per task dispatch or per review request.

If you plan to use the PR review workflow, also set:

  • Main GitHub user: the user name the bot should refer to in review output, and the only account whose review comments can trigger automatic follow-ups
  • Default review prompt: reusable guidance that gets appended to every manual review request

If you also want automatic review follow-up after relevant PR activity, turn on Review follow-up. Manual reviews only need the main GitHub user; automatic follow-up needs both the user and the toggle.

The save button stays disabled until:

  • the remote agent itself has already been configured from the CLI
  • the shell prelude is non-empty
  • the main GitHub user is present when review follow-up is enabled