Share the browser with an agent
Every human and agent tab appears in the same window. A purple sparkle marks agent tabs; a magenta edge and task controls identify the page where the agent is working. Your own tabs remain usable.
Take over or stop work
Choose Take Over or press Shift–Command–Period to use an agent's page yourself. Browser actions from the agent are blocked until you explicitly return control. Choose Resume Agent to continue agent control, or dismiss the takeover message to leave the browser idle so a newly declared task can start. Clicking or typing in an agent-controlled tab does not grant control; use the explicit control first.
Stop interrupts pending browser work and keeps the tabs open for inspection. Public WebKit APIs cannot forcibly terminate JavaScript already executing in a page, so an evaluation timeout can report possible late effects. Observe the page before repeating an action that might have changed it.
Inspect an agent's work
Open the inspector with Option–Command–I or the toolbar button. Activity lists commands, Page shows the selected tab's owner and semantic snapshot, and Agent Connection shows task state, API status, and the selected login profile.
Navigate with the keyboard
| Shortcut | Action |
|---|---|
| Command–L | Select the full address; opens a human tab if the current page belongs to an agent |
| Command–T | Open a tab |
| Command–W | Close a tab |
| Shift–Command–[ / ] | Switch tabs |
| Command–[ / ] | Go back or forward |
Keep the browser window open and non-minimized for an agent's native mouse and keyboard input. It can remain behind other apps without moving your system pointer or bringing the browser forward.
What survives a restart
Tabs start empty after an app restart. Saved WebKit login data and imported profiles persist, but open-tab history does not. Agents must join again and open the pages they need. See login profiles for importing sessions from another browser.