Launching today
Burnstate
Claude Code, Codex & Cursor limits, paced in your menu bar
100 followers
Claude Code, Codex & Cursor limits, paced in your menu bar
100 followers
Burnstate is a native Mac menu bar app for people who code with AI. It shows your Claude Code, Codex, Cursor and Vercel AI Gateway limits, marks how much of each window has passed, and tells you when you'd run out at your current pace. Answer Claude Code permission prompts and questions from a card, see every session's tokens and cost, take breaks that count only time away, and share a year of usage as a link. No account, no analytics.
Payment Required
Launch Team / Built With
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hey Product Hunt 👋
We're Serhii and Yozhef. We code with Claude Code and Codex all day, and kept hitting the same wall: the 5-hour limit runs out in the middle of a task, and you only find out when it's already gone.
The usage page didn't help. "60% used" sounds fine, unless only an hour of the window has passed. What we wanted to know was simple: at this pace, do I make it to the reset?
So we built Burnstate, a native Mac menu bar app that measures every AI coding limit against the clock.
What it does
- ⏱ Pace, not just percent. Every meter has a white tick for how much of the window has passed. Fill ahead of the tick means you're burning faster than the clock, and Burnstate tells you when you'd run dry.
- 🧰 All your tools in one place. Claude Code, Codex, Cursor and Vercel AI Gateway, side by side. It uses the logins already on your Mac.
- 💬 Answer Claude Code from a card. Permission prompts and multiple-choice questions land in a card over whatever you're doing. Allow, deny or pick an answer without switching to the terminal.
- 🔔 A tap on the shoulder. Cards when a turn finishes, a session needs you or a limit is about to run out. They never steal focus.
- 📊 Every session, live. What's working, what's waiting on you, and what each session cost, from your local logs.
- 🌙 Cooldown. Reminders to take a break, eat and stop for the day. A break counts only time you're actually away, and your agents keep working while you're gone.
- 🔥 Burn report. Your year of AI coding on one page, shared as a link. The numbers live inside the link itself, so nothing is uploaded.
Privacy
No account, no analytics, no Burnstate server. The app talks to each provider directly with your own logins, and your transcripts stay on your Mac.
Availability
Burnstate needs macOS 26.5 Tahoe or later. It's available as a direct download with a one-time license from $12, updates included. It's coming to Setapp and the Mac App Store soon.
We'd love your feedback:
- Which provider should we add next?
- What does your limit look like on a bad day? Do you pace yourself or push until the wall?
We'll be here all day answering questions. Thanks for checking it out! 🙏
👉 https://burnstate.dev
@yozhef Hey guys congrats! I hit the Claude Code wall in the middle of a task way too often. Does it also ping you when a session is just sitting there waiting for you or only for the limits?
Can I see a different warning for each tool? I would like to get more notice before hitting a weekly limit than a five-hour one.
@samulskaya Good one! Right now the thresholds are the same for every tool, but they're based on pace, not just the percentage. Burnstate warns you about each limit separately, the 5-hour and the weekly one, when there's less than 15% left or when you're burning more than 20% ahead of the clock.
That second part is what helps with weekly limits: if you've used 70% by Wednesday, you'll get a heads-up days before you hit the wall, not hours.
Custom thresholds per tool and per limit aren't there yet, but it's a really useful idea and we've added it to our list. Thanks!
the pace tick instead of just % is so right, 60% used means nothing if its only been an hour. does it handle 2 claude code accounts on the same mac or just one login
@bacelj Thanks, that's exactly why we built it 🙌
Right now it's one Claude Code account: Burnstate shows the limits for the login Claude Code is currently using on that Mac. Sessions and token stats from your local logs are counted across your Claude config folders, but they aren't split per account yet.
Good news: we're already working on the next release, and support for multiple Claude Code accounts is coming in it. We'll ping you here when it's out!
push till the wall every time. does it track the weekly limit too?
@leoheo Ha, same here 😅 Yes, it tracks the weekly limit too. For Claude Code and Codex you get both the 5-hour and the weekly window, each with its own pace tick, and for Claude also the per-model weekly limit when your plan has one. So you can push the 5-hour window to the wall and still see if the week will hold.
Congrats on the launch! Does the run-out estimate adjust quickly if one big task burns through a lot of usage?
@marianna_tymchuk Thanks so much! 🙏
Yes, it picks it up quickly. Burnstate refreshes your limits about every 5 minutes, so a big task shows up on the next refresh: the meter jumps past the pace tick and the run-out time moves earlier.
One detail: the estimate uses your average burn rate since the window started, not just the last few minutes. In a 5-hour window one heavy task shifts it a lot. In the weekly window a single burst is smoothed out, so the forecast won't panic over one big afternoon, but you'll still see the meter run ahead of the tick right away.
How does Burnstate calculate the usage? As usage differs when models are changed for the respective LLMs.
@rajat_gupta51 Great question! There are two parts to it:
1. Limits come straight from the provider. The percentages and reset times are what Anthropic, OpenAI and Cursor report for your account, so they already reflect how each model counts against your plan. Where a provider has a separate limit for a model (like Claude's weekly limit for one model next to the all-models one), Burnstate shows it as its own meter.
2. Tokens and cost come from your local logs. Burnstate reads Claude Code and Codex logs on your Mac and counts tokens per model, including cache reads and writes. Each model is priced at its own published API rate, so switching from Sonnet to Opus shows up in the cost right away. Think of it as what that work would cost at API prices, not your actual bill. Cursor reports its own costs, so we use those as is.
The pace forecast uses how fast your limit is actually moving, so if a heavier model burns it faster, the forecast catches that too.
The pace tick is what sold me — '60% used' never tells me if I'm actually in trouble, but fill vs. time passed does at a glance. I hit the Claude Code wall mid-task too often, usually right in the middle of a big refactor, so having that in the menu bar before I start something long is exactly the check I'd want. Congrats on the launch — the no-account, local-logs approach is a nice touch too.