Chat answers questions. Agent mode gets things done. This walkthrough runs a real multi-step task end to end so you can see how the agent plans, asks permission, and acts.
Before you start
- A provider is connected and enabled (Configuring Providers).
- The Web tool is on (it's enabled by default) and the File System tool is enabled in Settings → Tools and Plugins if you want the agent to write files.
Step 1 — Open the launcher in agent mode
Press ⌘ Return to open the launcher. Enter agent mode by typing /agent at the start of your prompt, or press ⌘1. You'll see the mode indicator switch to Agent.
Step 2 — Give it a goal, not a step list
Describe the outcome you want and let the agent figure out the steps:
/agent Research the top 3 note-taking apps for macOS in 2026,
compare pricing and key features, and write the summary to
~/Desktop/notes-apps.md as a markdown table.
Step 3 — Watch it plan
The agent breaks the goal into steps — search the web, read a few pages, synthesize, write a file. It streams its reasoning as it goes.
Step 4 — Approve tool actions
When the agent wants to take an action that touches your system or the network, it pauses and shows you exactly what it will do:
- Web search / fetch — the query or URL
- Write file — the path and a preview of the content
- Run shell / code — the exact command
Click Approve to let a step run, or Deny to stop it. You can let safe, read-only actions through and scrutinize the writes.
Tip: If you trust a task, you can approve steps quickly. If you're exploring, deny anything you didn't expect and rephrase your request.
Step 5 — Review the result
When the agent finishes, it summarizes what it did and where the output went. Open ~/Desktop/notes-apps.md to see the file it created. If something's off, just reply in the same conversation — the agent keeps the full context and can revise.
Make it repeatable
Once a task works, you can:
- Save it as a schedule so it runs every morning or on a cron — see Task Scheduler.
- Turn the know-how into a skill so the agent handles similar requests automatically — see Skills.
- Hand it to a specialized agent with its own model and pinned skills — see Agents.
What's next
- Launcher Modes — every mode and how to trigger it
- Tools & Integrations — give the agent more capabilities
- Commands & Tags — steer the agent precisely