AI usage New

See how much of your Claude and Codex plan limits you've used: the 5-hour session and the week, set against the time left until each one resets. Poni reads the numbers on your PC, shows them in a tab of their own and as a small ring in the closed Poni, and can warn you before a session runs out.

Settings → AI usage · The AI usage tab in the open Poni

Turning it on

AI usage is for Claude Code and Codex CLI used on this PC with a plan: Claude Code signed in with your Claude account, Codex signed in with ChatGPT. Both are off from the start.

1. Open Settings → AI usage.
2. Turn on Claude Code, Codex or both. Under each, Poni says whether it found it on this PC, for example ● Found: Max 20x plan or ● Found on this PC.
3. Open Poni. The AI usage tab, with the gauge icon, is there now.

The tab shows only while at least one service is on. Like the other tabs, you can move it or turn it off in Settings → Modules (see Settings).

Options

Turning a service off also forgets its numbers.

The tab

The tab shows one service per page. At the top are its logo and name, the plan (like Max 20x, Pro or Plus) and, at the right, when the numbers were read, like Updated 14:05.

• With both services on, the other one's name sits at the top right, like Codex › or ‹ Claude. Click it to go there.
• The mouse wheel or a two-finger swipe over the tab does the same, one service at a time.
• Next time, the tab opens on the service you looked at last.

Reading a row

Each limit is a row with its name, a bar, and what the numbers mean for you.

• The bar fills with how much of the limit you've used.
• The white tick across the bar marks how much of the window's time has gone: the 5 hours of a session, or the 7 days of a week. A bar that stays left of the tick is fine.
• At the right of the name, both as numbers: 42% used · 61% of the time gone.
• Next to the name, what that means: Plenty left when you've used less than half of what the time gone allows, On track when you use it no faster than the time goes, or Runs out around 16:40 at this pace when you use it faster, so at this speed the limit runs out before it resets (with a day, like Fri 09:20, when that isn't today).
• Under the bar: when the window started at the left, and when it resets at the right, like resets in 3h 50m, or resets Fri 09:00 when that's more than a day off.
• The row turns amber from 70 % used, or when it would run out before it resets. It turns red from 90 %.

Very early in a window (under 3 % of its time, about 9 minutes of a session) there's no verdict yet. When a window's reset time has passed, its row starts again from 0 % until fresh numbers come in. With four limits or more the rows sit closer together: the line under the bar goes, and the reset time sits next to the percentage.

Claude's rows

Which rows you see depends on what Anthropic reports for your plan.

Codex's rows

Codex shows a 5-hour session and This week. Codex CLI writes its limits into its session logs each time it answers, so the numbers are as fresh as your last Codex turn, and Updated shows when that was. They only show once you've used Codex signed in with ChatGPT.

The ring in the closed Poni

With Show in the closed Poni on, a small ring sits at the left edge of the closed Poni, next to the Caps Lock mark when that shows. It fills with the busiest 5-hour session of the services that are on: green, amber from 70 %, red from 90 %. When that session resets, the ring empties again. It doesn't show while Poni is open.

The warning

When a 5-hour session reaches the Warn at line (90 % from the start), the closed Poni says so with a chime, for example Claude session 90% used with Resets in 41 min, at 14:30 under it. Click it to open the AI usage tab. Each session warns only once. If Poni is open at that moment, a short message shows in its top bar instead. Set Warn at to Off for no warning.

What Poni reads, and what it sends

• Claude: Poni reads Claude Code's sign-in on this PC, the .credentials.json file in %USERPROFILE%\.claude (or in CLAUDE_CONFIG_DIR when you set one). With it, Poni asks Anthropic for your usage, the same request Claude Code's /usage makes. It asks every 3 minutes, and less often after an error.
• Poni only reads the sign-in. It never renews or changes it, since that could sign Claude Code out. The sign-in goes nowhere but that one request to Anthropic, and it's never written to poni.log.
• Codex: Poni reads the logs Codex CLI keeps in %USERPROFILE%\.codex\sessions (or CODEX_HOME), on your PC and offline. It looks every 30 seconds and reads again only when the newest log has changed.
• Nothing else is sent anywhere. The last numbers are kept on your PC, so the tab has them at once after a restart.
• Poni only reads while a service is on and something shows the numbers: the tab, the ring or the warning. With both services off, nothing runs.

Open Claude Code to refresh

Claude Code's sign-in expires now and then, and Claude Code renews it the next time it runs. Until then the tab keeps the last numbers, and its top right says Open Claude Code to refresh. Start Claude Code. Once it has renewed its sign-in, Poni notices within about 30 seconds and the numbers come back.

Messages in the tab