New: the Usage window. See where your quota went – by project, model and hour, up to 365 days back. Raw tokens and the equivalent at API list prices (not an invoice), the estimated share of the 5-hour and 7-day windows, and a weekday × hour heatmap. "Create analysis" saves an HTML report, plus CSV export. AI Cockpit reads back what Claude Code still keeps, then saves the totals itself – numbers only, stored locally. Also new: a monthly budget with warnings, and an optional daily summary. Fixed: Anthropic API costs were a hundred times too high, session tokens were counted about 2.7 times, Kimi shows yuan on the China platform, Copilot labels AI credits correctly.

Fixed: the ChatGPT card showed your Codex credit balance as a dollar amount. It is not one. OpenAI keeps usage credits and your dollar wallet apart, and states that credits have no cash value. The line now says what it is — credits — and so does the amount spent since the start of the month. The App Store listing was reworked at the same time: new screenshots taken from the running app, showing all eight services instead of five, each with a line saying what it is good for.

Copilot counts again. GitHub moved its billing from premium requests to AI credits in June — if your plan had switched, the card showed nothing. It now asks both and shows the one that carries figures. Fixed: the model-scoped weekly window still fell back to Opus, although it has been Fable since the start of September. The App Store listing has been rewritten as well: the description now says what the app actually reads, including the live ChatGPT fetch that replaced reading local Codex logs.

The desktop widgets used to freeze over time: they asked to be redrawn far too often and ran into the system limit. Fixed — the tile stays current. When the ChatGPT sign-in breaks or Kimi rejects a key, the card now takes you where it can actually be fixed. Until now it offered only Retry, which in those cases could never help. The key-figures bar keeps a missing value in place, greyed out, instead of dropping the tile and rearranging the whole bar. The Anthropic card has a new colour. The old one sat too close to the secondary text and, in dark mode, too close to Claude — two cards you need to tell apart at a glance. Also: the connections list under Settings → Privacy is complete, demo mode now shows sample figures throughout instead of partly real ones, ChatGPT also appears under the Costs segment because of its credit line, and a cancelled request no longer counts as a service outage. New: a tip jar under Settings → About — a voluntary thank-you that unlocks nothing and changes nothing in the app.

One line that answers a question no dashboard does: what does AI cost you this month, all in? Enter your subscription prices once (Settings → Display → Subscription costs) and the cockpit adds your live API spend on top. Kimi and Grok stay out of the sum — they only report a balance, and an honest number beats a complete-looking one. Two equal links in Settings → About: send feedback, or rate the app. And the app may now ask for a rating — rarely, only after days of successful use, never after an error. Also in this update, from the unreleased 6.2: the plan badge on the Claude card is back and correct for both accounts (read per sign-in from Anthropic's profile endpoint), and the setup assistant now appears in English for English users — its strings were missing from the catalogue. Several more missing catalogue entries fixed on both languages.

A second Claude account. If you keep personal and work subscriptions apart, you could only store one until now — signing in a second time overwrote the first. Settings › Accounts now offers a second one, with its own card, its own mark and its own colour. Both accounts can be named, for instance «Personal» and «Work»; the name appears on the card, in the menu bar and in notifications. Nothing changes without a second account — the card only appears once you sign in. A demo mode. It fills every card with sample figures, without a sign-in and without keys. While it runs nothing is fetched and nothing is overwritten, and a band above the cards says so. A setup assistant on first launch. Until now, opening the app for the first time showed nothing at all — no dock icon, no window, just a small mark in the menu bar. Four steps now walk you through it, starting with where the app actually lives. Skip works on every step. Reset everything. Clears both Claude sign-ins and all keys from the keychain, resets every setting, signs out of «launch at login» and the hotkey, and deletes the recorded history. Nothing changes at the services themselves. Also: the Anthropic API card has its own colour now — it used to share Claude's orange, and with a second Claude account that would have been three cards barely told apart. The Grok card looked disabled because its colour sat at the lightness used for secondary text; it is now a cool silver.

Two new cards: OpenRouter and Grok. OpenRouter shows daily, weekly and monthly spend, the credit left on your account and — if the key carries a spending cap — how much of that cap remains. An ordinary key from openrouter.ai is enough; no organisation, no admin access. Grok shows the credit balance of an xAI account and its automatic spending cap. It needs the management key from the xAI console, not the ordinary API key. Cards can be moved. Drag a card to another position in the window, or into the other column, and it stays there. Settings has a reset that restores the built-in arrangement. The Anthropic API cost card works again. It asked for a report path that does not exist and stayed empty. Alongside the fix, a key that Anthropic will not accept for cost data now says so instead of showing nothing: cost figures need an admin key from an organisation, which individual accounts do not have. A key that gets turned down is a state, not a failure. Where a provider rejects a key, the card now carries a quiet note that names the kind of key required and offers a way straight into the settings, instead of retrying and reporting a fault. The light appearance is legible. Cards were darker than the ground they sat on — that, not the text colours, was what made the light theme look murky. Cards are now lighter than the background, no colour falls below a 4.5:1 contrast, and the settings sidebar no longer stays dark when the system is set the other way. Switching themes now redraws every card at once instead of leaving some in the old colours. ChatGPT is now called ChatGPT/Codex, on the card and in the settings.

Every card now carries its own mark. A small coloured square with the service's initial sits in front of each card name — C for Claude, G for ChatGPT, O for the OpenAI API, A for the Anthropic API, K for Kimi, S for your sessions. Until now colour alone told the cards apart, and the thin coloured edge at the left margin is out of view the moment you read the title. A cancelled request is no longer reported as a failure. Any fetch that came back cancelled — because a window went away, or a deadline expired — used to be treated like a broken service, and the raw system wording was shown on the card. Cancellation now has its own handling: whatever figures a card already had, it keeps. Requests now have a real ceiling. The previous timeout only measured the pause between two signs of life, not the total duration, so a service answering one byte at a time could hold a connection open for days. A request is now capped at two minutes. No system error text on a card. Messages written for developers, in whatever language the system happens to use, have been replaced with wording of our own: no internet connection, service unreachable, secure connection failed.

A fix reported by a customer, and one small addition. • Settings: the seven pages now sit in a sidebar on the left instead of a row of tabs across the top. On newer system releases that row could collapse into a small button in the title bar, leaving no visible way to switch pages at all. A sidebar cannot collapse, whatever the window size. • About now names the developer.

Fixes reported by a customer. • Settings window: it could appear squeezed, with rows pressed together and the lower part cut off. This happened on smaller displays, with a larger system text size, and on newer system releases that lay out controls with different spacing. The settings are now their own resizable window that adapts to your screen; anything that no longer fits scrolls instead of being compressed. • Main window: the bottom row with the sign-out, usage and quit buttons could be cut off by a few points. • Clearer labels: the buttons at the bottom now say which service they belong to — "Sign out of Claude" and "Claude usage page" instead of just "Sign out" and "Usage page".