Never hit a limit
you didn't
see coming.
AI-Cockpit puts the limits and the spend of eight AI services on one panel — Claude, ChatGPT & Codex, the Anthropic API, the OpenAI API, Kimi K3, OpenRouter, Grok and GitHub Copilot. It only reads. Nothing passes through a server of mine.
Both free · optional tip in the app
- What
- The limits and spend of eight AI services on one panel — Claude, ChatGPT and Codex, the Anthropic and OpenAI APIs, Kimi, OpenRouter, Grok, GitHub Copilot.
- Where
- Mac menu bar, plus iPhone, iPad, Apple Watch — and the Windows notification area. The phone fetches everything itself, no Mac has to be awake.
- Cost
- Free. No account with me, no subscription, no pro tier. An optional tip in the app that unlocks nothing.
- Trust
- It only reads. Keys stay in the device's keychain — on Windows, in the Windows Credential Manager. Nothing passes through a server of mine, no telemetry. No API key is required — the subscription cards run on OAuth.
The sentences this panel was built to prevent
Try again at 2:00 PM.
The limit isn’t the problem.
The surprise is.
It never arrives at a good moment. You are three files deep in a refactor, the context is warm, and the window was at 90% forty minutes ago — you just had no way of knowing that.
AI-Cockpit keeps the numbers where you already look: the Mac’s menu bar, your Home Screen, your wrist. Eight services on one panel — you watch a window fill up, and it turns amber before the window runs out. The phone fetches its own figures; no Mac has to be awake for it.
Look before you sign in
Demo mode fills every card with sample figures — no account, no key, nothing to enter. It is in the app on all three platforms, and it is the fastest way to see whether the cockpit shows what you need.
The full cockpit
Two columns, ordered by provider: the subscription sits above the API bill of the same house. Below, your active Claude Code sessions with context usage.
dashboard — full view with forecast and sparklines
Quick start
Nothing to read up on: a four-step assistant opens on first launch, Skip works on every step, and everything can be done later in the settings. One thing to know beforehand — the cockpit reads your accounts. The subscriptions and API credit it shows are ones you already hold and pay for at the providers themselves.
What it does with those numbers
The services are listed above. This is what the cockpit makes of them.
A forecast, not just a numberMAC + IOS + WIN
Every window says when it resets and, from your pace, when you will run into it — the figure you actually plan around.
Alerts before the wallMAC + IOS + WIN
A notification when a window crosses your own threshold. Two levels, both off until you switch them on.
Widgets everywhereMAC + IOS + WATCH
Desktop widgets on the Mac, Home Screen, Lock Screen and StandBy on the phone, a complication in four shapes on the wrist.
The menu bar, your wayMAC
Both values, only the most critical one, a ring or the time remaining — plus a key-figures bar you compose yourself.
Live Claude Code sessionsMAC + WIN
What is running right now, with model, subagents, token shares and context window — and a heads-up when a session waits for you.
Outages, namedMAC + IOS + WIN
When a provider reports an incident, its card says so — instead of leaving you to wonder whether the fault is yours.
Demo modeMAC + IOS + WIN
Every card filled with sample figures, without a single account. Look around first, sign in later.
Ask it out loudIOS
«How full is my Claude window?» — Siri and Shortcuts answer from the figures already on the device.
Eight services. Up to ten cards. One cockpit.
Every subscription and API you pay for, side by side — each with the one figure that decides whether you can keep working. Add only what you use; every card is optional. Claude counts twice if you want it to: a second account gets its own card.
All eight services run on all three platforms: OpenRouter and Grok arrived with version 6.0 for the Mac and are on iPhone, iPad and Apple Watch as of version 2.3; GitHub Copilot has been on iPhone, iPad and Apple Watch since version 2.5 and is on the Mac since 6.5. Only Claude Code sessions stays off the phone — it reads sessions running on the machine itself, so there would be nothing to show. Mac and Windows both have it.
One cockpit, five devices
The Mac app watches while you work. The iPhone app — which brings iPad and Apple Watch with it — answers the question you have everywhere else: is there anything left? And on Windows the same cockpit sits in the notification area of the taskbar.
the same number on the phone and on the wrist — sample figures from demo mode
AI-Cockpit for Mac
The full instrument panel, one hotkey away.
- All ten cards — including a second Claude account and your live Claude Code sessions
- Drag the cards into the order you want them in
- History, sparklines and limit forecasts
- Four menu bar styles — and you choose which figures they show
- Alerts before a window runs out
Free
AI Cockpit Mobile
The same numbers, wherever you are.
- All eight services, fetched by the phone itself — no Mac needed
- Home Screen widgets: one line per provider, in its colour
- Lock Screen and StandBy tiles with the age of every figure
- Demo mode — look around before signing in to anything
- Apple Watch included: a one-line verdict on your wrist and a complication in four shapes — no separate app, no extra download
Free · with an optional tip
AI-Cockpit for Windows
The same core, rebuilt for the taskbar.
- All eight services and up to ten cards, as on the Mac
- A ring in the notification area, showing the figure you choose
- Sign in to Claude and ChatGPT from the app itself
- Outage notices per provider, straight from the status pages
- Demo mode — look around before signing in to anything
Claude Code sessions run on Mac and Windows, not on the phone — they read files on the machine. Widgets are on Mac and iPhone; the tip jar and the voice query exist only in the App Store editions. Everything else is on all three.
About the Windows build: it is a snapshot, not a twin — 1.3 matches macOS 6.5 and will not follow every Mac release. It is in no store, so there are no automatic updates, and the file is unsigned: Windows will call it an unrecognised app, and you get past that with More info → Run anyway. Version 1.3 is built and tested, but a full round on real hardware is still running — if you would rather wait, 1.2 is still there.
What runs where
Windows is a snapshot, not a twin — and it is a test build: no store, no sandbox, no signature, no automatic updates.
| Mac | iPhone · iPad · Watch | Windows | |
|---|---|---|---|
| Version today | 6.5 | 2.5 | 1.3 · test build |
| Requires | macOS 14+ | iOS 26 | Windows 11 · x64 + ARM |
| Comes from | Mac App Store | App Store | GitHub release |
| All eight services | yes | yes | yes |
| Claude Code sessions | yes | no | yes |
| Widgets | yes | yes | no |
| Forecast and alerts | yes | yes | yes |
| Outage notices | yes | yes | yes |
| Demo mode | yes | yes | yes |
| Voice query (Siri) | no | yes | no |
| Tip jar | yes | yes | no |
| Sandboxed and signed | yes | yes | no |
| Updates | automatic | automatic | by hand |
| Credentials live in | Keychain | Keychain | Credential Manager |
Free, and nothing sold on the side
Both apps are free on their App Stores — no trial that expires, no feature held back for a paid tier. No subscription, no account with me, no usage limit. The subscriptions and API credit the cockpit reads are ones you already hold and pay for at the providers themselves.
It is paid for by a tip jar in the app, under Settings → About. It unlocks nothing and changes nothing — there is no paid tier to buy into. The long answer is in the FAQ.
- Credentials live only in the device's keychain — on Windows, in the Windows Credential Manager
- Direct connection to the official APIs — no server of mine in between, no telemetry, no ads
- The app only ever reads with a key: it creates, changes and deletes nothing
- Built by an indie developer in Switzerland — no VC, no data business (legal notice · Contact)
✓ Security-reviewed — four documented passes, the last against version 5.0 (10 August 2026). Checked against, not certified by:
A documented model-assisted review with Claude — not an external audit, and it says so plainly. The record names every host the app ever connects to, line by line: read the whole record, or check the privacy policy and the FAQ.
What's new
Mac 6.5, iPhone, iPad and Watch 2.5 and the Windows 1.3 test build are out — GitHub Copilot as an eighth provider, an outage indicator per provider, and ChatGPT figures from a live fetch. Every release in plain language →
How the numbers are read
Three pages go into the detail — each one also says what the app cannot know, because a number you cannot trust is worse than none.
Claude usage limits
The 5-hour and 7-day windows, model-scoped weekly limits, how the forecast is calculated and when it warns.
Codex quotas and reset times
Two Codex windows, never three: how quotas are matched by window length, where the live figures come from, what credits are.
API costs at a glance
Subscription quota is not API cost. Which services are read, why admin keys are needed, and what is only ever read.
Feedback & feature requests
Missing a provider? Want a metric the cockpit doesn't show yet? Tell me — short and direct is perfect.
Ready for the cockpit?
Mac, iPhone and Watch come from their App Stores, Windows as a test build — free, no subscription, no account with me: just the provider accounts you already have.
Both apps: free · no subscription · an optional tip inside the app
The subscriptions and API credit the cockpit reads are yours and stay with your providers — it adds no cost of its own. If you do leave a tip and change your mind, that goes through Apple, at reportaproblem.apple.com.