Session, weekly and per-model quota — with burn rate, reset countdown and what it cost. In your menu bar.
macOS 14 or newer · Claude Code CLI · Claude Pro or Max
Five-hour sessions and the weekly quota run independently, and you find out about both at the moment Claude Code stops answering. What burned the quota, and what it cost, isn't visible at all.
The percentage sits in your menu bar and changes shape at 80 % and 95 %. Burn rate becomes a time: −5.2 %/h, empty at 19:40 — or “reset comes first”, when that's the truth. And the history shows which repository cost what.
The icon carries status by shape, not colour alone: normal, 80 %, 95 %, exhausted, stale, offline, no access, demo — eight states you can tell apart. Ring, ticks or dot, your choice, and it reads correctly on both a light and a dark menu bar.
Burn rate is turned into a time of exhaustion. When the reset arrives first, the app says so instead of inventing a frightening hour. Notifications fire at 80 % and 95 % of the session and 80 % of the week — each one once per limit window.
Cost is computed from your local logs against date-aware rates, then broken down by project, model and token type. Most people discover that three quarters of the money is cache reads.
Every number carries its source and its age — read from the API, or computed on this Mac. When the connection drops the number doesn't fall to zero, it is demoted: dashed outline, the time of the last reading, the countdown stopped. A separate banner appears when the problem is on Anthropic's side, so someone else's outage never looks like your quota running out.
Three layouts — instrument, dense, calm — four palettes, three glyph styles, a menu bar mode and desktop widgets. Switching is instant; it's one component set throughout.
A gauge that goes quiet is useless. A gauge that lies is worse.
Flip the switch — the connection to the API drops.
Still 33 %. It just can't confirm it right now.
It never had a reading. It shows zero anyway.
A gauge that goes quiet is useless. A gauge that lies is worse.
See what the app actually reads →It finds the credentials the Claude Code CLI already created. No password, no account.
Read access to ~/.claude turns on history and cost. Live monitoring works without it.
It warns you at 80 % and 95 %, and tells you when the limit resets.
Answers written the way the app behaves.
No. It never asks you to sign in. It reads the single Keychain item the Claude Code CLI already created, read-only, and sends that token only to Anthropic's official API. No other Keychain item is touched.
One Keychain item, and — only if you grant it — your local ~/.claude logs, to compute cost history. A built-in inspector shows exactly what was read and from which address, with the address taken from the running code rather than typed into a string.
Only the token, and only to api.anthropic.com. There is no server of ours in between, no analytics and no account. You can revoke access in Keychain Access at any time, and the app keeps no copy.
Yes. Demo mode explores the whole interface on sample data, and every screen is labelled as demo while you're in it — including the inspector.
Yes. There are no in-app purchases and no locked features. Everything described on this page is in the free app.
It indexes your existing logs once. On a real 2.5 GB log — 721 files, 48,860 records — that first pass took 310 seconds. Every scan after it takes about half a second, because 720 of the 721 files are skipped without being read.
macOS 14 or newer. It needs the Claude Code CLI and a Claude Pro or Max plan — it reads the limits of the plan you already have.
No. It's an independent, unofficial app, not created by, affiliated with, endorsed by or sponsored by Anthropic. “Claude” and “Claude Code” are trademarks of Anthropic, PBC, used here only to describe compatibility.
An unretouched screenshot: every item the app read, and the address it was read from, taken from the running code.

Free. No account. macOS 14 or newer.