Usage limits
Your provider's own reported quota, fetched live from its usage API. The bars are real, not a local guess.
4-inch desk display
A 4-inch always-on desk display that shows your live Claude Code, Codex and Gemini usage — straight from the machine you already work on, with the logins you already have.
Ships worldwide · delivery expected before Q4 2026 · early-access price locked for the first 100
What's on screen
No invented metrics. Each number has a real source — and we're upfront about which is measured and which is estimated.
Your provider's own reported quota, fetched live from its usage API. The bars are real, not a local guess.
Totalled from the CLI logs on the computer running the broker. No API key, and reading them costs you nothing.
Token counts × a public price list. A useful estimate, not a bill — on a subscription plan it's notional, not money charged.
How it fits together
A small open-source server on your laptop reads your CLI usage; the device polls it over your Wi-Fi; a one-line marketplace plugin pairs the two from your AI.
The tokenmonitor-mcp server reads your Claude Code / Codex / Gemini logs and serves the numbers.
Polls the broker over your LAN and draws the dashboard. Holds no provider logins of its own.
Installed from the marketplace. You pair the device by asking your AI to run /tokenmonitor:configure.
Is it safe?
The specific version — not a "trust us" badge.
It isn't built to read or send your prompts or code. The broker parses usage and token counts out of your local logs — there's no path that ships their contents anywhere, and it's open source so you can verify that.
The server is open source (Apache-2.0). The part that runs on your laptop and touches your logs is yours to read, line by line.
No inbound exposure. The broker only listens on your LAN for the device (port 8765) — nothing to forward from the internet, no telemetry, no analytics. Its outbound calls go to your providers and a price list, nowhere else.
The device doesn't phone home to us. Its only direct internet calls are a public clock server and the weather — never an AI provider, never your usage data.
No account with us, no cloud lock-in. You use the provider logins you already have. There's no Fractal Manifold sign-up.
Signed requests & signed firmware. Device polls are HMAC-signed and replay-protected; updates are Ed25519-signed with automatic rollback.
Three providers · day, night & auto
Claude Code, Codex and Gemini each get their own brand-tinted palette in a day and a night flavour — six looks, drawn pixel-exact at 480×480. Auto follows your city's real sunrise and sunset; or lock one with a tap, or set it from your CLI. Try it:
Why a dedicated display
You can type a command or open a menubar widget — but the moment you're deep in a task is exactly when you won't. An always-on panel tells you the answer before you remember to ask.
It just sits there, lit. No command, no tab, no notch widget fighting for your menubar — you glance and you know.
Usage lives on the desk, not in your editor. Nothing to dismiss, nothing covering your code.
Standby after 20 min, day/night dimming, alert volume you set. The few alerts that fire — low battery, limit overshoot, session rollover — are debounced so they never spam you.
Not a CLI replacement. It's the at-a-glance layer on top of the tools you already run.
Before you ask
No. There's no new API key and no extra tokens spent. Token counts are read from logs already on your machine, and the limit percentages come from each provider's usage API — a read, not a request that bills you. The dollar figure is an estimate from public list prices, not a charge. More in the FAQ →
No flashing — units ship pre-flashed. You plug in USB-C, join the device's Wi-Fi once to enter your network, then pair it by asking your AI to run a skill. On your computer you install the tokenmonitor-mcp broker (one command) plus the marketplace plugin that runs it — nothing is flashed. See the 5-step setup →
Enable one or all three. It works with Claude Code, Codex CLI and Gemini CLI; a provider whose CLI you don't run simply doesn't appear on the device. The pairing plugin installs into whichever of those you use. Provider details →
The broker and plugin are open source (Apache-2.0), so the device keeps working whether or not we're around — you can run, audit and fork the server yourself. The limit, plainly: new hardware, repairs and future firmware signing depend on us continuing. Firmware & longevity →
The first run is a small, carefully assembled batch — each unit pre-flashed and tested, ready to pair out of the box. We're finalising production with a manufacturing partner now; join the waitlist to lock the early-access price and we'll email you the moment units open.
*The early-access price ($11099€) is locked for the first 100 people to join the waitlist. The standard price may change before launch; the early-access price for those first 100 won't. Shown in USD across the Americas and EUR elsewhere — switch above.