Browse the guide

Your first chat

On this page

Start a chat to work with one agent in its own terminal. You can choose the agent directly or use a saved role with its own instructions and settings.

Start and finish a small task

  1. Install and sign in to an agent, then open Runner.
  2. Press ⌘N / Ctrl+N to open Start a chat. If an empty pane has focus, the chat will fill it; otherwise it starts in a new tab.
  3. Choose Direct and select your agent, or choose Role and select a saved role such as Coder. The seeded Coder role requires Codex. Model and effort choices depend on the agent; leaving them at their defaults uses the agent or role configuration.
  4. Set Working directory to a folder you want the agent to read. A project selection can supply this path. Check the path before starting.
  5. Click Start chat, or press ⌘Enter / Ctrl+Enter.
  6. In the terminal, send a small prompt such as: “List the top-level files in this folder and explain what each is for. Do not change any files.” Complete any sign-in, trust or approval prompts in the agent's own terminal.
  7. Wait for the agent's reply. Its terminal remains available for follow-up questions. When finished, use Stop in the session controls, or ⇧⌘X / Ctrl+Shift+X with that session focused.

A stopped chat remains in Runner. Resume, or ⇧⌘R / Ctrl+Shift+R, continues its saved conversation when the runtime supports it. Sessions explains the difference between stopping, resuming and closing.

Add another pane

Press ⌘D / Ctrl+D to split right, or ⇧⌘D / Ctrl+Shift+D to split down. Use New chat in the empty pane to select another agent or role. Each pane holds a separate session; splitting does not copy the first agent's conversation or create a mission.

Open a shell

Press ⌥F12 / Alt+F12 to show the chat tab's terminal drawer. Use it for commands such as git status while keeping the agent visible. Press the shortcut again to hide it. You can also start a standalone terminal tab with ⌘T / Ctrl+Shift+T.

Close and return later

Runner's background service owns the sessions. In Settings → General → When Runner quits, choose whether to ask, keep sessions running, or stop them. With Keep running, reopening Runner reconnects to the live terminals. Resume running agents on launch, in the same pane, controls recovery of agents that were live when Runner stopped them; it does not automatically restart every chat you stopped manually.

For a team working together on one task, continue to your first mission.

View source on GitHub ↗