How to track Claude Code usage limits: the 5-hour window and the week

If you use Claude Code on a Pro or Max plan, you have probably met the message that says you have hit your limit, usually in the middle of something. The limits are not hidden, but they are easy to forget about until they stop you. This guide explains how they work, where to check them, and how to keep an eye on them without breaking your flow.

A note before we start: Anthropic changes plan details from time to time, and the exact amounts depend on your plan, the models you use and how long your conversations are. For the current numbers, Anthropic's help centre is the source of truth. What follows is how the limits are structured, which has been stable.

Two limits, two clocks

Claude's subscription plans limit usage in two overlapping ways.

The 5-hour window

Usage is counted in sessions that last five hours from your first message. Within that window you have a budget. When it is used up, you wait until the window resets. The window starts when you start, so if you begin at 9:10, it resets at 14:10, not on the hour.

The weekly limit

On top of the 5-hour windows, Anthropic introduced weekly limits in 2025. These cap the total over seven days, so you cannot simply use the maximum in every 5-hour window, day and night. Depending on your plan there can be more than one weekly limit, for example one across all models and a separate one for particular models.

The practical effect: the 5-hour window is the one you hit on a heavy afternoon. The weekly limit is the one you hit on a Thursday after a big week, and it takes longer to come back.

What uses it up

Usage is not counted in messages. It depends on how much the model reads and writes. A few things make Claude Code sessions heavier than chatting:

• Context. Every turn, the model reads the conversation so far, plus files it has opened. A long session costs more per turn than a fresh one.
• Model choice. Larger models use your allowance faster than smaller ones.
• Agentic loops. One instruction can lead to many model calls as Claude Code reads files, runs tests and fixes errors.
• Shared allowance. Your plan's usage is shared between Claude Code and Claude on the web and in the apps.

If you use Claude Code with an API key instead of a plan, none of this applies in the same way. You pay per token, and the limits you meet are API rate limits.

Where to check your usage

Inside Claude Code

Run /usage in a session. It shows how much of the current 5-hour window and of your weekly limits you have used, and when each one resets. This is the exact figure, from Anthropic, and the first place to look.

In your account settings

Claude's settings on the web show your plan's usage too. Useful when you are away from the terminal.

From your local transcripts

Claude Code keeps a transcript of every session on your Mac, as JSON Lines files under ~/.claude/projects/. Each assistant message records the tokens it used. Open source tools such as ccusage read these files and add them up by day, session or 5-hour block. This gives you history and detail that /usage does not, like which project used the most. It is an estimate of your plan usage, not the official number, because the plan's accounting is Anthropic's.

The problem with checking

All of these share a flaw: you have to remember to look. And the time you most need to know is exactly when you are deep in a task and not thinking about it. Running /usage every half hour is its own small interruption.

What helps is a number that is always in view, and a nudge before the limit, not at it.

Keeping it in view

We added this to Crowny because we kept running into it ourselves. Crowny shows your Claude usage in the MacBook notch: the 5-hour window as a ring and the week as a bar, each with the time it resets. The panel also shows today's usage by model.

How it gets the numbers, since it matters:

• By default, from your transcripts. Crowny reads the same local files as the tools above, reading only what has been added since it last looked, so it stays light. Where Claude Code has saved its own recent /usage figures for the current window, Crowny uses those. It says on screen where the numbers come from.
• Exact, if you choose. Connect it in Crowny's setup and it asks Anthropic for the same figures /usage shows, using the Claude Code sign-in already on your Mac.

Past 80% of a window, the notch lets you know once, so you can decide to wrap up, switch to a smaller model or save the big refactor for after the reset. When the window resets, it tells you that too. Nothing is sent to us. The transcripts never leave your Mac.

Habits that stretch your limits

Whatever you use to track it, these habits make the allowance go further:

1. Start fresh sessions for new tasks. A long conversation carries its whole history into every turn. Use /clear between unrelated tasks, and /compact when a long task needs to keep going.
2. Match the model to the job. Use the largest model for planning and hard problems, and a smaller one for routine edits.
3. Be specific. Pointing Claude Code at the right files saves it from reading half the repository to find them.
4. Keep a CLAUDE.md. Project notes it reads at the start save it from rediscovering your conventions every session.
5. Plan around the reset. If the window resets in 40 minutes, that is a good time for review, tests and writing, not a large new task.

In short

There is a 5-hour window that starts with your first message, and a weekly limit on top./usage gives the exact picture. Your local transcripts give history. A number in view gives you warning before you hit the wall.

If you would like that number in your notch, alongside every running Claude Code session, download Crowny or see what a licence includes. And if the other thing that slows you down is sessions waiting quietly for approval, see our guide to Claude Code notifications on a Mac.