Usage & limits

Coding subscriptions come with limits: a five-hour session window, a weekly window, sometimes separate weekly limits per model. PaneScale shows how much of each is left, when it resets and whether your current pace will last — and, from your local session logs, what your agents would have cost at API list prices.

You see it in two places:

• the usage meter at the bottom of the left sidebar, always visible,
• the Usage page — click the meter, use the command palette, or choose Usage & limits in the account menu.

#The usage meter

One line per limit window, grouped by provider:

Below the windows: what your agents would have cost today and in the last 30 days. Lines from a source older than 15 minutes are dimmed.

Hover the meter for a detail card: every window with x % left, the exact reset time, the usage of the last hour (“+6 %/h”) and a one-line forecast:

• Pace is fine — you’ll have used about x % at reset,
• Pace is tight — it lasts about until the reset,
• Pace too high — used up at a given time,
• Limit reached — blocked until the reset.

#The Usage page

#Spend

The top card shows what your coding agents cost, computed locally from the session logs of Claude Code and Codex — including additional Claude Code configuration folders you use for other accounts.

• Switch between Today, Yesterday and 30 days.
• The donut splits the total by provider. Click its centre to switch between dollars and tokens. Hover a slice or a legend row to highlight it.
• Models lists the seven biggest models of the period with their share; the rest is summed up.
• Last 30 days shows one stacked bar per day with the daily average and the peak.

Prices come from public price lists (LiteLLM, models.dev and the OpenUsage supplement) and are refreshed daily. PaneScale stores token counts, not dollars, so new prices apply to your whole history without reading the logs again. Models without a known price are flagged — their tokens count, their dollars are missing.

The first scan reads all logs once and shows its progress (“312 of 840 files”); afterwards only new entries are read and the page is up to date within seconds.

#Subscription cards

One card per provider, with your plan (for example Max 5x) and account, and where the numbers come from: live from your Claude sessions, queried 2 min ago, or another source. A chip warns when something is off: Limit reached, Query throttled, Sign-in expired or Not reachable.

For every window:

• a big x % left and how much is used,
• Reset in … with the exact time on hover,
• a bar with a mark for the even pace,
• a hint when it matters: Limit in 1 h 20 (with a flame), at this pace empty in 3 h, tight · lasts about until the reset, or Limit reached · free at 19:09.

Below that, if your plan has them: extra usage, cloud credits and their expiry, a 30-day usage trend, links to the provider’s Status page and usage Dashboard, and the facts for today, yesterday and the last 30 days with the most used model.

If neither CLI is signed in with a subscription, the page tells you how: sign in with claude or codex login.

#Where the limits come from

For each window, PaneScale uses the most recent of these observations:

Providers allow only about one usage query per minute and account. When a query is throttled, PaneScale waits as long as the provider asks — or longer with each retry — and keeps showing the last values. Refresh on the Usage page asks again sooner.

The last state and a history are saved locally, so the meter is filled the moment PaneScale starts.

#Alerts

When a window passes 90 %, PaneScale shows a notice with the reset time and a link to the Usage page. When a window that was nearly full resets, you get a short free again notice. Each alert appears once per window. Turn them off with desktop notifications in Settings → Notifications.

#Usage, budgets and the Guardian

The Usage page shows your provider’s limits. Token Guardian enforces your own budgets on runs that PaneScale starts. Budgets & Guardian in the page header takes you there.