BlogDevelopers2 October 20266 min read

See your Claude Code and Codex usage limits on your Mac

If you use Claude Code or Codex every day, you've hit this: you're halfway through a refactor and the agent stops because you've reached your 5-hour limit. Or worse, your weekly one. The numbers exist, but you have to go looking for them. Here's how to keep them in view all the time, in the MacBook notch.

How the limits work, in brief

Claude and ChatGPT subscription plans meter agentic coding in rolling windows. There's a short 5-hour session window and a longer weekly window. Each fills up as you use the agent and resets on its own schedule. Running out of either pauses your work until it resets, unless you switch models or pay for extra usage.

The tricky part is that heavy tasks, like long contexts, large repos, and many tool calls, can burn through a window much faster than you'd guess. Planning around the reset time beats being surprised by it.

Checking limits the built-in way

• Claude Code: run /usage in a session to see your current 5-hour and weekly usage.
• Codex: the CLI shows your rate-limit status, and records it in its local session logs.

Both work, but only when you remember to check, and only in the terminal you're already busy in.

Seeing them in the notch

NotchDude has an AI Usage widget for this. (Disclosure: I make it, and built this widget because I kept hitting the limit myself.) Hover the notch and you see, for each tool:

• 5-hour and weekly limits as bars or rings, depending on the card size, with a countdown to each reset.
• Token totals for today and the last 7 days, and a daily chart.
• Cached context re-read this week, kept separate so it doesn't inflate your fresh token count.
• A warning colour as you approach a limit: orange at 70% and red at 90%.
• Agent sessions: while Claude Code or Codex is working, its card names the project and shows whether it's busy or waiting for you. The closed notch shows the agent's icon and its 5-hour limit, so you can keep an eye on it without opening anything.
• Cursor and Antigravity cards that show recent activity. Those tools keep their limits on their servers.

Where the numbers come from (and what's never read)

Claude Code and Codex both keep local logs on your Mac, in ~/.claude and ~/.codex. NotchDude reads those folders read-only, and takes only token counts, model names, and limit percentages, plus, for agent sessions, the project folder name, timing, and tool names. Your prompts, conversations, and code are skipped.

• Codex writes its plan limits into its session logs, so NotchDude reads them directly.
• Claude Code caches its last-known usage locally. If you allow it, NotchDude can also ask Anthropic for the same numbers /usage shows, using the sign-in Claude Code already saved in your Mac's Keychain. macOS asks you first, and if you decline, NotchDude uses the cached numbers instead.

No account and no API key to paste, and none of your code ever leaves your Mac.

Tips to make your limits last

• Watch the reset, not just the percentage. 80% used with 20 minutes until reset is fine. 80% with four hours to go means it's time to slow down.
• Start big tasks right after a reset. Kick off long refactors or multi-file changes early in a window.
• Keep context lean. Clear or compact long sessions, and point the agent at specific files rather than the whole repo.
• Use the right model for the job. Save the heaviest model for hard problems. Routine edits rarely need it.
• Check the weekly trend. If the 7-day chart climbs every Monday, plan your heavy days around it.

Other ways to track AI usage on a Mac

Command-line tools that parse Claude Code's logs, and small menu-bar apps, do part of this. Vibe Notch, a free open-source app, takes a different approach: it shows Claude Code sessions in the notch so you can approve or deny prompts. NotchDude focuses on limits, reset times, and what each agent is working on across both Claude Code and Codex, next to the other tools you already reach for in the notch: clipboard history, a focus timer, music, and notes.

FAQ