Meet Chloe, your operator dock, plus usage you can see

This release introduces Chloe, an operator dock you talk to in plain language to get things done across the app. You also get clear usage and cost graphs in settings, a faster-starting app, and a refreshed specialists experience.

Tell Chloe what you need

You can now open Chloe from the rail or with cmd-K and describe what you want in plain language. Chloe figures out where your request belongs and hands it off to the specialist best suited to the work, so you spend less time hunting for the right place to start and more time getting the answer. Chloe is available to everyone in this release.

Token usage and cost you can see

Settings now show graphs of your token usage and spend over time. You can watch consumption trends, spot what’s driving cost, and make decisions with the numbers in front of you instead of guessing.

A faster, calmer app

The app starts up noticeably faster, and reopening it on macOS no longer forces a minimized window back open or stacks up a second dock badge. These are quiet changes, but you should feel the difference every time you launch.

A refreshed specialists experience

The marketplace, specialist detail, and builder pages have been redesigned for a cleaner, more consistent look. When you set up a new specialist, you now see similar specialists that already exist, so you can build on what’s there instead of starting from scratch. Resuming a saved draft shows an overview banner so you know exactly where you left off, and the local Codex variant now streams its output as it works. Specialists can also read the channel description for better context, and code-writing specialists run with clearer, tighter permissions.

Stay in the loop on the threads that matter

You can now subscribe to or unsubscribe from a thread right from the thread view, so you only get pinged on the conversations you care about. Notifications are also more reliable: you won’t be sent to channels that no longer apply, and duplicate alerts for the same event have been collapsed into one.

A tidier chat and projects experience

Project pages have been overhauled and the old grid is gone, with list items now sized and spaced consistently across the app. Adding channels is smoother, mentions render correctly while you edit a message and are scoped to people actually in the conversation, and your loaded thread replies stick around after a refresh instead of disappearing. The chat panel also stays visible even when every sandbox panel is open.

Settings that make more sense

Configuration now lives inside your workspace settings, where you’d expect to find it. When you finish in settings, you’re returned to wherever you came from rather than always being dropped back into chat. Provider setup is clearer too: hosted Hugging Face is its own option, Bedrock inference profile labels read plainly, and if a GitHub token is rejected you now see the reason why. The personal translation engine controls have been removed.

Getting back in when access fails

If your workspace access hits a snag, recovery is clearer: the “Try again” action now actually retries and tells you what’s happening instead of leaving you stuck.

Polish & fixes

Workflows are steadier: the node palette no longer swallows clicks meant for canvas nodes, icons and colors survive a credential update, the editor header no longer overlaps in narrow windows, and the chat model picker hides raw errors. Connection tests for external tools now give clearer messages, and external tool cards are easier to read. Web pages opened in the in-app browser no longer flash a white background. The “Include archived” toggle now actually reveals archived tasks, and the tasks summary card has a tidier empty state. Dialog buttons, toast close buttons, and other small spacing and empty-state details have been cleaned up across the app, and a handful of stale errors and stray identifiers in chat and channels no longer surface.