Free usage forecasts · Claude Code + OpenAI Codex · macOS
See your Claude Code and Codex limits coming.
A free usage tracker for Claude Code and OpenAI Codex that lives in your menu bar. It reads the logs already on your Mac and forecasts when each account runs dry, so you see the wall coming with hours to spare instead of finding out when a turn stops coming back. Watch one account or both, side by side, never mixed. No sign-in. No network. Nothing leaves your Mac.
You run dry about 1h 26m before the reset. Drop to Sonnet for the rest of this window and you land at 94%.
Scroll to see how it works
Two of your three sessions are on Opus at max effort: 71% of this window's burn.
Projected to hit 140% by reset; Sonnet lands it around 94%. Lowering effort alone doesn't get you there.
The forecast, one click from the menu bar.
Click the chip. The popover says which limit hits first, when, and what to change. It is rebuilt here from the app's own views, and the command chip really copies.
- A burn rate, not a percentage. 1.4× means tokens are going 40% faster than the clock. Above 1.0× you run dry before the reset; below it, you're leaving capacity unspent.
- The week, forecast. Learned from your own history, so it says when you run out, not just how much you've used.
- The fix, worked out. Every model and effort change is measured against the hours you have left. On Claude, the cheapest one that lands under the cap comes with its command, ready to paste. On Codex, the what-if panel compares Sol, Terra, and Luna using OpenAI's published token credit rates and labels the result as an estimate.
Your Opus week is projected to hit 105% by reset; Sonnet lands it around 83%.
Lowering effort alone doesn't get you there.
Two agents. Two accounts. One menu bar.
BurnCast watches Claude Code and OpenAI Codex as what they are: separate accounts. Each keeps its own limits, reset clocks, history, and notifications. Nothing is summed and nothing is averaged — a heavy Codex day never moves your Claude forecast an inch.
- One meter or two, your call. By default the menu bar shows whichever account is closer to its limit. Pin it to one account, or give each its own meter; click a meter and the forecast that opens is that account's alone.
- Codex figures are exact from the first minute. Its logs carry your account's own limit percentages on every request, so there is nothing to install and nothing to estimate.
- The right windows for each. Claude Code gets the five-hour session, the week, and the Opus week; Codex gets the five-hour session and the week. Neither is ever shown a meter its plan does not have.
- Separate everything. Two histories in two files, two reset clocks, two sets of alerts. Add or remove either account in Settings, and neither ever touches the other's numbers.
Where the week lands.
The session resets in five hours; the weekly limit is the one that ruins a Friday. BurnCast learns what each weekday costs you and says where the week ends, not just where it is.
The app's own week view, running in the page. Keep scrolling and the week runs hot, or switch it yourself and open the Models tab.
Change one thing. Watch the forecast move.
Anyone can print “use less.” BurnCast measures the moves you could actually make (drop a tier, lower effort, close a session) against the hours you have left, and names the cheapest one that still gets you to the reset. If nothing does, it says so.
If I change something
now · 58% opus, 37% fable
now · 56% x-high
Pick a model or an effort level and the forecast recomputes against the hours you have left. Only the remaining week is scaled. What is already spent is spent. A Codex account offers Sol, Terra, and Luna alongside an effort dial whose ladder runs one rung higher; model results are estimates because switching models can also change the token count.
- Your own baseline. Built from your history, not a generic token table.
- Early enough to act. Advisory, heads-up, critical. The first word arrives while a change still fixes it.
- Honest about uncertainty. A band, not a line. Where it's wide, the app says so.
Burning too fast is one failure. Burning too slow is another.
Most meters only tell you when you're in trouble. Half the value is the other direction: capacity you paid for and didn't use. Colour is the verdict; the number is your burn rate.
No uplink. No telemetry.
BurnCast reads the logs already on your Mac and keeps everything it computes there. The binary imports no networking framework, so there is no channel for anything to leave on.
- Free, and not the kind of free that sells you. No ads, no in-app purchases, no subscription. Nothing is collected that could be sold in the first place.
- No sign-in, in either direction. There is no BurnCast account to create, and the credentials for the accounts it measures are never read. It knows your usage from logs, not from access.
- No analytics, no crash SDK, no update ping. Not off by default. Absent.
- Your prompts and your code are never read. The parser takes token counts, timestamps and model names, and skips the message bodies entirely.
- Uninstalling is a folder you delete. No daemon, no login item left behind, no receipts.
Questions, answered.
How do I check my Claude Code usage?
Claude Code tells you what you have spent. BurnCast reads the same logs and forecasts where that lands: your menu bar carries a live burn rate, and above 1.0× the forecast has you running dry before the window resets. Below it, you have room to spare. There is no command to run.
Does it work with OpenAI Codex?
Yes, as a full account alongside Claude Code, not a bolt-on. BurnCast reads the session logs the Codex CLI writes to ~/.codex/sessions, watches both of Codex's windows — the five-hour session and the week — and forecasts them the same way it does Claude's. Codex's logs carry your account's own limit percentages on every request, so its figures are exact with nothing to install. Only the sessions folder is read; never the file that holds your sign-in.
Can I watch Claude Code and Codex at the same time?
Yes. Each account keeps its own limits, reset clocks, history, and notifications; nothing is summed. The menu bar shows whichever account is closer to its limit, or pin it to one, or give each account its own meter — clicking a meter opens that account's own forecast.
How is this different from ccusage and other Claude Code usage trackers?
Most trackers add up what you have already spent. BurnCast runs that spending forward: it predicts which limit you will hit, the day and hour you will hit it, and names the change that moves the date. And it sits in your menu bar rather than behind a command, so you see it coming rather than going looking. It also watches OpenAI Codex alongside Claude Code, as two separate accounts.
Why is it called BurnCast?
Burn rate and forecast. It does two things: measures how fast you are burning tokens right now, and predicts where that rate lands you by the time the window resets. The first number lives in your menu bar. The second is one click away.
Does it need my API key or an account?
No, neither. There is no BurnCast account to create, and it never asks for — or ever sees — your Anthropic or OpenAI credentials. It measures your accounts from the logs Claude Code and Codex already write on your Mac, without ever connecting to them. For Codex, folder access is scoped to the session logs alone — never the file that holds your sign-in.
Is it really free?
Yes, and there is nothing to buy later. No ads, no in-app purchases, no subscription, and no paid tier holding features back. What you download is the whole app. It collects no data either, so nothing about you is being sold instead.
Which plans does it work with?
Any Claude or ChatGPT plan with the rolling five-hour session and weekly limits. BurnCast works out which windows apply to each account and tells you which one you will hit first.
How accurate is it?
Out of the box it estimates from your logs, and the estimate improves as your history grows. For exact Claude Code figures, switch on the usage helper — setup offers it the first time you open the app. Codex needs no helper: its logs carry your account's own percentages on every request, so those figures are exact from the start.
I mostly use Claude on the web. Does BurnCast see that?
Usually, yes. Your limits are shared across everything you use Claude for, so BurnCast only needs one source on your Mac to read them from. The Claude desktop app is one: leave its menu-bar icon on, and the version you download from this site reads your real figures from what it keeps on disk. Claude Code is the other: sign in with the same account and switch on the usage helper. Either way, the percentages you see are your whole account's, and setup walks you through it.
What's the difference between the App Store and direct builds?
It is the same app either way. The App Store version asks once for your ~/.claude folder, because apps from the store start with access to nothing until you say otherwise, and it updates itself through the store. The version from this site does not ask, and it can also read your figures from the Claude desktop app. Everything else, including the usage helper, is identical.
Will the helper break my status line?
No. It chains ahead of whatever command you already have, backs up settings.json first, and Remove helper restores it verbatim.
Does it slow anything down?
No. There is no background service and nothing being polled. The first read of a large history takes a couple of seconds, and after that it only looks at what changed, which is too quick to notice.
Does it send anything anywhere?
No. The binary contains no networking code at all. Not off by default, absent. There is no update check, no telemetry, and no sign-in, because there is nothing that could reach a server.
Does it read my prompts or my code?
No. The parser takes token counts, timestamps and model names and skips message bodies entirely.
What do I need to run it?
macOS 14 Sonoma or later, on Apple silicon or Intel. The build is universal and about 4 MB. You also need Claude Code or the OpenAI Codex CLI on the same Mac, since those write the logs BurnCast reads. Nothing else to install.
Is it signed and notarized?
Yes. It is signed with a Developer ID certificate and notarized by Apple, so it opens normally rather than behind the “unidentified developer” warning. The download page publishes the checksum if you want to confirm your copy is the one that was published.
Windows or Linux?
Not for now. BurnCast is macOS only.
How do I uninstall it?
Remove the helper if you installed it (Settings → Official usage gauge → Remove helper), drag the app to the Trash, and delete ~/Library/Application Support/BurnCast. Nothing else is left behind.
Get BurnCast.
Free: no ads, no in-app purchases, no sign-in. Nothing leaves your Mac.
Version 1.0.254 · 7.2 MB · macOS 14 Sonoma or later · Universal
- Developer ID signed · notarized · stapled
- Reads your local Claude Code and Codex logs, nothing else
- No third-party SDKs · no network connections
- Downloadthe 4 MB disk image
- Dragto Applications
- Doneyour first forecast lands within seconds of your next turn
From the Mac App Store it is one click instead, and the app asks once for the folder it reads. Same app, and it updates itself from there.