BlogGuide8 October 20268 min read

Track your Claude Code and Codex usage at a glance, right in the notch

You're deep in a task, Claude Code is halfway through a change across ten files, and then it stops: you've hit your limit. The numbers were there all along, you just had no reason to go and look. AI Usage in NotchDude puts them where you already look a hundred times a day, the top of your screen, so you can plan around your limits instead of running into them.

Why checking usage is easy to forget

Claude and ChatGPT plans meter coding agents in two rolling windows: a 5-hour window and a weekly one. You can check them, with /usage in Claude Code or /status in Codex, but only from inside the session you're busy in, and only when you think to. Nobody stops mid-flow to type a command just in case.

So the limit becomes a surprise. And a surprise at 4 p.m. on a Thursday, with the weekly window nearly gone, can cost you the rest of the week. The fix isn't more discipline. It's having the number somewhere you'd see it anyway.

Three ways to glance at your usage

(Disclosure: I make NotchDude, and built AI Usage because I kept hitting the limit myself.) It gives you three levels of detail, depending on how much you want to know:

1. Without opening anything: the closed notch

While Claude Code or Codex is working, the closed notch shows the agent's icon, gently pulsing, and a ring of the limit it's spending. You see it fill as you work, without hovering, clicking, or switching windows. (When music or the focus timer is showing there, they take that spot.)

2. A hover away: every limit and reset

Hover the notch and open AI Usage. Each tool gets a card with:

• Your 5-hour and weekly limits, as bars or rings, with a countdown to each reset.
• The project each agent is working in, and whether it's busy or waiting for you, so you know when to switch back.
• Today's tokens, and on larger cards, a chart of the last 7 days.

3. On Home, next to everything else

Add AI Usage to Home and your limits sit beside your calendar, music, and focus timer, so a quick look at the notch covers your whole day.

What the colours tell you

The percentage alone doesn't tell the whole story. 80% used with 20 minutes until the reset is fine; 80% with four hours to go means it's time to slow down. That's why every limit shows its reset time right beside it.

Example: red at 95%

Here's what to do the moment you see it:

1. Let the current step finish, then don't start a new multi-file task. Long contexts and many tool calls spend the last few percent fastest.
2. Check the weekly bar too. If it's also high, a reset in 1 h 40 m only gives you a fresh 5-hour window, not a fresh week.
3. Switch tools or tasks. If Codex has room, hand it the routine work. Or use the time to review the agent's changes, write tests, or plan the next step yourself.
4. Pick up after the reset. The countdown tells you exactly when; start the big task then, with a full window.

The point of seeing red at 95% is choosing what happens next, rather than having the limit, or a bill, choose for you.

Don't let the limit turn into a bill

Hitting a limit used to mean waiting. Now both Claude and ChatGPT can keep you going past your plan's limits on paid credits, billed at usage rates. That's handy when a deadline can't wait, but easy to run up without noticing if it's switched on and you're not watching. Check these settings once:

Then let the notch do the watching. Orange at 70% and red at 90% give you time to decide before your plan's limit is reached and credits start being used. NotchDude only shows your usage; it never buys credits or changes your plan, so these settings stay the real off switch. Settings change over time, so check Anthropic's and OpenAI's help pages for the current options, and look again after app updates.

A day with AI Usage

• 9:00, planning. The 5-hour window has just reset and the weekly bar sits at 35%. A good morning for the big multi-file change.
• 11:30, mid-task. The ring on the closed notch turns orange. You let the agent finish the current change and save the next one for after lunch, when the window resets.
• 14:00, waiting for you. The card says Claude Code is waiting in your project. You'd stepped into email; now you know it's your turn.
• 16:20, red at 95%. One task too many. You stop the agent at a good point, review its changes yourself, and pick up after the reset instead of paying for extra usage.
• Thursday, the weekly view. The weekly bar is at 80% with two days to go. You use Codex for the routine work and keep Claude Code for the hard problem.

None of these moments needed a command, a dashboard, or a browser tab. For more ways to stretch a window, see tips to make your limits last.

Who it's for

• Claude Pro and Max subscribers who use Claude Code every day and want to stop guessing.
• ChatGPT Plus and Pro subscribers using Codex, with the same 5-hour and weekly view.
• People who use both, who can see which one has room left and pick accordingly.
• Cursor and Antigravity users: those tools keep their limits on their servers, so their cards show when you last used them and whether their agent is working.

Set it up in two minutes

1. Download NotchDude and open it. Every Pro feature, AI Usage included, is unlocked for your first 48 hours.
2. Hover the notch and choose the AI Usage tab. If you don't see it, turn it on in Settings › Widgets.
3. For live Claude numbers, choose Always Allow when macOS asks to let NotchDude read Claude Code's sign-in. Without it, you see the numbers Claude Code last saved.
4. Drag cards into the order you like; hold one to resize or remove it.

Numbers update every minute while the widget is open, and you can click refresh any time. There's no account, no API key to paste, and nothing to configure in Claude Code or Codex.

Private by design

Claude Code and Codex already keep logs on your Mac, in ~/.claude and ~/.codex. NotchDude reads them read-only and takes only token counts, model names, limit percentages, project folder names, timing, and tool names. Your prompts, conversations, and code are never read, and never leave your Mac. The full breakdown explains where each number comes from.

FAQ