v2.3.7: Reconnecting Claude Works Again, Billing Gets Clearer, and Every Employee Ships With More Skills

v2.3.7 is a big one. The headline fix: reconnecting your Claude account works again, after Anthropic quietly moved things around and broke the flow for everyone. Alongside it, SaaS billing gets a lot friendlier (a new cheapest tier, a live credit bar, and clearer trial limits), every new employee now ships with two handy skills built in, and a batch of small annoyances got cleaned up.

Reconnecting Claude works again

If you bring your own Claude account, the "Reconnect Claude" button had stopped working. You would click it, the screen would say "Waiting for Claude session to start", and then nothing.

This one was not our doing. Anthropic changed three things at once without warning: they moved the web address where you log in, and they changed a status message in Claude Code so that a logged-out session now looks, to software reading it, exactly like a logged-in one. Our reconnect flow was reading those old signals, so it got stuck waiting for something that would never arrive.

It is fixed now, on both places where you can reconnect: the button in the dashboard and the terminal's own login screen. Click reconnect, complete the login, and you are back in. No waiting, no hang.

SaaS billing got a lot clearer

A handful of changes for anyone on a hosted plan, all aimed at removing surprises.

There is a new Mini tier, the cheapest option, sitting below Standard. It is the lightest and most affordable way to run an employee, and it leans on the new built-in vision skill (more on that below) so it can still handle images even though it is a lean model.

Free trials are clearer about their limits. During a trial, the top two tiers (Max and Medium) are locked, new employees start on Mini, and the lock is real rather than cosmetic: trying to switch to a premium tier during the trial is genuinely blocked, not just greyed out. When you activate a paid plan, everything unlocks.

And there is now a live credit bar in the sidebar, showing how much of your allowance you have used and how much is left, green while you have plenty, amber and then red as you get close. A gentle prompt appears at the halfway and three-quarter marks to suggest a tier that fits your usage. No more discovering you are out of credits mid-conversation.

We also fixed something quieter: on hosted plans, the usage dashboard was not counting any of the actual usage. It now does, so the numbers you see are real.

See your Claude and Codex subscription usage (bring-your-own-key)

If you connect your own Claude or Codex subscription, admins can now see how much of that subscription's quota is already used, both the rolling five-hour window and the weekly one, right in the usage panel. The information always existed inside the tools themselves; now it is in one place where you actually look. It only shows when a relevant account is connected, and it is hidden entirely on hosted plans where it does not apply.

Every new employee now arrives with two skills built in

New employees are now created with two useful skills already installed, so you do not have to add them by hand.

The first is a vision fallback: it lets an employee work with images even when its underlying model has no built-in image understanding. This is what makes the lean Mini tier still able to look at a screenshot or a photo you send it.

The second is the document processor, for creating and reading PDFs, Word files, spreadsheets, and presentations. It was already standard on most employees; now it is guaranteed on every single one, however you create them.

Existing employees can still add either skill in one step from the skills menu.

Codex history is back

A recent update to the Codex tooling (the version we picked up around June 4) quietly stopped saving conversation history in the format we read, which showed up as an empty context gauge for Codex-based employees. We have rolled that tooling back to the previous stable version so history is written properly again, and fixed a related issue where clearing a session could wipe more history than it should. Codex employees keep their history through a session clear now.

Smaller fixes worth mentioning

• The compact button asks first. The one-click compact button (the one that frees up space in a full conversation) used to act the instant you clicked it. It now asks for a quick confirmation, showing how full the conversation currently is, so an accidental click does not compress your chat.
• Close buttons stay where you can reach them. In longer settings panels, the close X and the header used to scroll away as you went down the page. They stay pinned at the top now.
• The white-label toggle saves itself. The white-label mode switch sat far above its Save button, so people flipped it and left without saving. It now saves the moment you click it, with a confirmation.
• No more double messages. A message occasionally appeared twice in a row until you refreshed. That duplicate is gone.

Picking it up

Hard refresh (Ctrl+Shift+R) to load v2.3.7. A few of the employee-level changes apply the next time an employee's session is cleared.

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