v4.0.17: See Which Employee Is Using Your Browser and What It Is Doing, and Stop Agent Finally Stops

If an AI employee is driving your browser, you should be able to tell which one, and what it is doing, without asking anybody. From v4.0.17 you can.

The browser tells you who is using it

With extension 1.4.4, when an employee acts on a page a banner appears naming it, the site, and the action: "Alex is using this browser on example.com: Clicking on the page". It shows on the employee's own tab and on whichever tab you happen to be looking at, so you find out even when you are somewhere else.

While it works, its tab joins a coloured group in the tab bar carrying its name. Your pinned tabs and your own groups are left alone.

Three details matter more than the feature itself.

The name comes from the server, not from the employee, so one employee cannot present itself as another. If an employee's browser skill is out of date it shows as "An employee" rather than borrowing a name it has no right to.

What you type is never shown or logged. Only the site being opened is recorded. The extension's settings keep a list of recent actions, and that list contains sites and actions, not content.

The banner gets out of your way. Drag it where you want and it stays there, and it leaves ten seconds after the last action. One honest limitation: Chrome does not allow an extension to draw on its own pages or on error pages, so the banner cannot appear there.

Stop Agent now stops

The Stop Agent button on the page did nothing. That is a bad thing for a button to do, and worse for that particular button.

It now blocks every employee on that browser and clears the banner from all tabs, and stays that way until you turn browser automation back on in the extension settings. An action already running is allowed to finish rather than being cut off halfway through, which is the right trade: a half-completed click is usually worse than a completed one.

Files arrive as links you can click

Employees now hand over files as proper Markdown links, with the file name as the text and the full path as the target. Existing employees pick up the rule when the instance restarts and apply it from their next session.

Three related annoyances went with it. A link inside a plain message, one with no list or heading, used to come out broken as [file.md]( file.md ) and could not be opened. A closing parenthesis after a path or a web address used to get swallowed into the link, while addresses containing their own parentheses now stay whole. And a relative link like [notes.md](notes.md) opened the app in a new tab instead of the file, which now opens properly in the file explorer.

Small things individually. Together they are the difference between an employee that tells you where a file is and one that hands it to you.

Two smaller improvements

You can sign in on more devices before being paused. The limit on emailed login codes goes from 3 to 25 per user per 15 minutes. Anyone signing in across several browsers and extensions in one session was hitting the old ceiling in normal use. The limit on wrong code attempts is unchanged, since that one is doing a different job.

The LLM providers list is readable again. Names, addresses and models were being cut off or squeezed into narrow columns by the buttons sitting next to them. They now get the full width with the buttons on their own row underneath, and anything still too long shows in full on hover.

Want to test the most advanced AI employees? Try it here: https://Geta.Team