Run your first agent

Outcome: an agent window running a real task on your machine. Time: ~5 minutes. You’ll open an agent, task it, watch the transcript, and check status — the loop you’ll use all day.

Prerequisites

At least one agent CLI installed and signed in (Claude Code is the default). Settings → Agents shows what’s detected.

Steps

1. Open an agent. In the command bar at the bottom, type open an agent and press Enter (or press Ctrl+Shift+A, or use the left-toolbar agent button to pick a specific brand). A window appears with an auto-assigned name — say, Alex · Claude Code.
2. Give it a task. Two ways:
   • Type into the agent window’s own input: summarize what’s in this folder → Send.
   • Or route from the command bar: tell Alex to summarize what’s in this folder. A meteor animation flies to the window that received it.
3. Watch it work. The transcript streams the agent’s output; the pane glows while it runs; the token meter (top right of the pane) tracks usage.
4. Stop anytime. Click Stop in the pane, or say stop Alex.
5. Check on everyone. Ask who’s working or give me a status in the command bar — you’ll get a per-agent rundown.

What you should see

The agent replies in its own window; when it finishes you get a toast (“Alex finished”) and the status dot settles. The task is also recorded in the agent’s persistent profile — Alex remembers what categories of work it has done across sessions.

Recap

Open (Ctrl+Shift+A) → task it (directly or “tell <name> to …”) → watch → stop → status. Everything you just typed also works by voice.

Next

Routing commands to agents · The Vibe Code workflow · Loops: refine until good