v2.4.3: Browser Automation That Works Across Every Machine, Plus a One-Click Off Switch

If you have ever signed the Chrome extension into the same account from two computers and watched your browser actions bounce back as "Invalid secret code," v2.4.3 is the release you have been waiting for. This one is about making browser automation behave like a grown-up feature: reliable across multiple machines, controllable with a single switch, and honest about what it is doing. There is also a cleaner dashboard, a wider kanban board, and a quiet but important fix to usage accounting.

Browser automation now works across every machine

The old behavior was a genuine footgun. The browser gateway routed every action to the first connected session for an account, no matter which machine it actually came from. Connect the extension on your laptop and your desktop at the same time, and actions meant for one would land on the other and get rejected as "Invalid secret code." If you ran more than one computer, browser control was effectively unreliable.

Actions are now routed to the exact extension that holds the matching secret. That single change unlocks the way people actually work:

• Connect the extension from several computers at once.
• Have different employees driven by different machines, in parallel.
• Reassign an employee to a new machine, and the previous session is cleanly replaced rather than left to fight over the same actions.

The security model holds up under all of this. The secret never leaves your machine. The extension advertises only a one-way hash of it, so routing knows where to send an action without anything sensitive traveling over the wire.

A single switch to turn browser control off

Giving an AI employee control of your browser is powerful, and power you cannot quickly revoke is power you end up not using. So the extension Settings page now has a Browser automation toggle.

Flip it off and every browser action from your employees is refused immediately. The extensions stay connected, your employees keep their pairing, the capability is simply suspended until you flip it back on. It applies to all of your employees at once, so when you want hands off the wheel, it is one switch, not a cleanup project. This replaces the old per-chat unassign button, which only handled one conversation at a time and was easy to forget.

Tables that are actually tables

A small but satisfying fix. When an agent returned a markdown table in the extension chat, say, a list of your open tabs, it rendered as a wall of [object Object],[object Object]…undefined. Useless. Tables now display as proper, styled tables, so the information you asked for is actually readable.

The dashboard stops leaking its own plumbing

The hover preview on each employee card was showing the raw last prompt and reply. That was fine when the trigger was a person typing, but when it was a system message you saw the internal machinery: a speaker tag, a [KANBAN_CARD_UPDATE:{…}] token, a NEW EMAIL TASK … REQUIRED WORKFLOW block. It was noisy and a little unprofessional.

Those are now translated into short, human-readable labels. An email trigger reads "New email received." Delegations show who they came from or went to. Messages from Slack, Teams, Gmail, Outlook, WhatsApp, Telegram, or a webhook show a clean channel label and the actual message. Scheduled tasks show the task. Kanban updates, memory reminders, context summaries, and skill updates all get friendly labels instead of raw tokens. Replies are tidied too, so markdown symbols and table pipes no longer leak into the one-line preview.

A wider kanban board

The kanban board, whether you open it from the chat sidebar or the dashboard, was capped at a fixed width and left a lot of empty space on a big monitor. It now uses the full available width. Columns stretch to fill the screen, and still scroll horizontally when the window is too narrow to show them all. More of your work visible at a glance, less wasted space.

Claude and Codex usage is back in the books

This one matters if you watch your costs. When Claude and Codex moved to their new streaming engines, their per-turn token usage started being written to a new on-disk location that the usage collector was not reading. The result: Claude and Codex sessions quietly stopped appearing in the Usage Overview, while custom LLMs were unaffected. If your numbers looked suspiciously low, that is why.

The collector now reads the correct source for all three engine types, and the missing usage is back-filled automatically on the next turn. Your Usage Overview is complete and trustworthy again.

Versions and rollout

The Chrome extension is bumped to v1.1.2, and the desktop app is bundled at v1.0.0. The extension changes ship in v1.1.2, so update the extension to get the multi-machine routing, the kill-switch, and the table fix. The server-side changes, the switch enforcement, the dashboard previews, the wider kanban, and the usage accounting, are already live.

None of this is flashy. It is the difference between a browser-automation feature you trust on your real setup and one you only ever tried once on a single laptop. v2.4.3 makes it dependable.

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