See the wallbefore it sees you.
cclimit sits in your Mac’s menu bar and watches your Claude Code limits: the 5-hour window, the weekly caps, every model. And it doesn’t just count. It tells you when you’ll hit them.
brew install --cask 1fc0nfig/tap/cclimitA percentage isn’t an answer.
Claude Code can tell you you’ve used 68%. It can’t answer the question you’re actually asking mid-session: can I keep going? cclimit watches your burn rate and answers in plain words.
What /usage tells you
68%
Is that fine? Is that a crisis? Depends entirely on your pace, which it doesn’t know.
cclimit
“At this pace, Fable hits the weekly cap Thursday afternoon.”
A verdict, not a number. Reschedule the heavy agent run, or relax. Either way, you knew before the wall did.
A forecast you can actually trust.
Prediction is easy; honest prediction is the product. Three rules keep cclimit’s forecast worth listening to:
It only speaks when you're burning.
The 5-hour forecast needs real movement, several rising samples in the last 15 minutes, before it says anything. Idle windows don't get extrapolated into panic.
Day-level honesty, never fake precision.
Claude Code usage is bursty, so weekly forecasts say “Thursday afternoon”, never “Thursday 14:07”. A wrong timestamp kills trust faster than no forecast at all.
Server truth always binds.
The gauges show exactly what api.anthropic.com reports, from the same endpoint /usage uses, correct across all your machines. The forecast just reads the slope.
And when it matters, it taps you on the shoulder:
One predictive notification per window, at least 15 minutes of lead time. Enough to land the change and let the agents coast instead of dying mid-refactor.
Your menu bar, your rules.
Nine icon styles, all drawn natively at every pixel density. Point the icon at whatever binds you: the 5-hour window, the weekly cap, or a single model. Try them:
The popover is yours too: drag sections into your order, hide the ones you don’t care about, keep the whole thing monochrome if that’s your taste.
Every model, automatically.
Anthropic ships per-model weekly windows now, and they change without notice. cclimit parses the limits feed generically, so when a new model gets its own cap, it simply appears in your popover. No update, no waiting on a release.
It also shows you which window is actually binding you right now. Usually not the one you’d guess.
Built like it sits next to your credentials. Because it does.
Read-only, by architecture
cclimit reads the OAuth token Claude Code already keeps in your Keychain and never writes anywhere. Tokens stay in memory; nothing is copied, cached, or re-stored.
Zero telemetry
The only network connection it ever opens is to api.anthropic.com, the same usage endpoint Claude Code's /usage command calls. No analytics, no crash reporters, no accounts.
Native and invisible
Swift + SwiftUI, under 5 MB, no dock icon, no Electron. Polling adapts to your activity (60 s while burning, minutes while idle) and pauses when your screen locks.
Open source, MIT
An app that sits next to your credentials should be readable to the last line. It is: the repo is public and the license lets you do anything with it.
Don’t take the page’s word for it. Read the source.
On the roadmap
Which agent ate your window?
If you run agents in parallel, the window doesn’t just vanish. Something spent it. The popover already breaks the current window down by project, so the answer is one click away: not just when, but who. And we’re building deeper analytics on top: per-project trends, history across windows, and more on the way.
Local files only, relative shares only. The server numbers stay the source of truth.
This 5-hour window, by project
FAQ
How does it read my usage?
From the same endpoint Claude Code's /usage command calls (api.anthropic.com/api/oauth/usage), using the OAuth token Claude Code already stores in your Keychain. Read-only: cclimit never writes to your Keychain or credentials file. The numbers are server-side, so they're correct across all your machines.
Is this affiliated with Anthropic?
No. cclimit is an independent open-source project. The cream, the coral, and the asterisk are an homage, nothing more. If Anthropic ever ships this natively, we'll happily retire.
How is this different from /usage?
/usage tells you where you are, when you think to ask. cclimit watches continuously, forecasts where you're heading, and warns you about 20 minutes before the 5-hour wall. You stop thinking about limits. That's the point.
Will it drain my battery?
No. It polls every 60–120 seconds only while you're actively burning, backs off to 5–10 minutes when idle, and pauses entirely when your screen locks. It's a native SwiftUI app under 5 MB; the energy cost rounds to zero.
Does anything leave my Mac?
One thing: the usage request to api.anthropic.com. That's the complete list. No telemetry, no analytics, no accounts, no update pings beyond checking GitHub releases when you ask it to.
What does it cost?
Nothing, ever. MIT-licensed. If it saves your afternoon once, star the repo. That's the entire business model.
Stop getting surprised by the wall.
One line in your terminal, one glance in your menu bar. Free forever.
brew install --cask 1fc0nfig/tap/cclimit