Auto-translate, display names, and a billing rework

This release brings auto-translation to incoming chat and specialist messages, a custom display name for how you appear in conversations and @-mentions, and a billing rework that introduces a Max tier and a smoother checkout. Settings now has its own page, and you can stop a streaming response right from the message it came from.

Incoming messages translate automatically

If you have auto-translate enabled, new messages from other people and from specialists are translated as they arrive in the timeline — no need to open each message to read it in your language. Messages you wrote yourself are skipped, messages already in your language are skipped, and translations switch over when you change your language preference. Specialist replies translate the same way as messages from people.

Set a display name for @-mentions

Personal Settings now has a Display Name section. The placeholder shows the name pulled from your sign-in provider, and leaving the field blank keeps that name in place. Once you save a name, it appears wherever you show up in conversations and is the name others use to tag you with @. The mention list in the message composer respects this name for you and for everyone else in the workspace.

Billing rework with a Max tier

The billing flow has been reshaped. There’s a new Max tier alongside the existing plans, credit top-ups apply correctly, new accounts receive an initial credit grant, and the checkout itself is cleaner end to end. Settings, Invite, and Billing Upgrade also moved out of a floating dialog and onto their own pages — Cmd/Ctrl+, and the billing deep link still open them, and they now load with their data ready to go.

Chat reliability and per-conversation browser panels

You can now stop a specialist directly on its own message card instead of from the composer, so cancelling one streaming response in a multi-agent channel no longer interrupts the others. Chat connections also recover their sign-in automatically when a reconnect comes back unauthorized instead of failing silently, and clearer credential diagnostics make it easier to tell whether a provider issue is a missing key, an expired token, or a system keychain that’s unavailable.

Browser panels in chat now have their own preview per conversation, and you can open developer tools on those panels in release builds via the Inspect right-click menu — useful when an embedded preview is misbehaving and you want to look under the hood.

Settings, suggestions, and translation defaults

Empty-state chat suggestion chips now route to a fixed specialist when one is mapped to that suggestion, so the same chip always sends to the same specialist. The Translation Model picker has a refreshed, curated list, with Claude Sonnet 4.5 as the new default. Local Models settings now lets you search Hugging Face directly, shows estimated download size before you start, and reports live transfer speed during the download.

Polish & fixes

A layout crash that could happen while a new pane was being added is fixed. The duplicate “workspace created” toast is gone. An internal model no longer appears in the model status list. Workspace keychain operations no longer hang under contention. The settings dialog now closes cleanly when you log out instead of floating above the login screen. The traffic-light position settles correctly when the system theme changes. The channel rename popover stops shifting when your mouse leaves the sidebar item. Dark-mode card, popover, and elevated-surface backgrounds now render at the correct color. Display-name handling in mentions falls back consistently when the value is empty. Knowledge Garden, Tasks, and unfinished Apps links are hidden in Manage and Development settings while those features come together. Switching accounts now fully clears cached data — workspace, sandbox, permissions, apps, models, GitHub, and MCP — so stale data from a previous session no longer lingers.