Your typical week
Two-hour blocks in local time, normalised to your busiest block, across every day the app has collected.

Know what your AI coding actually costs, how close you are to each limit, and where your usage is heading. Tokens, requests, models, and cost across your tools, in one running total.
Works withCodexClaudeCursorOpenCodeand more coming soon
No account required. Create a free account if you want to save and restore your usage history.


Sources
A source ships only once its sign-in path, its error handling, and its adapter tests are finished, so everything below is pulling real numbers right now. New coding tools join the same running total as they settle.
Goes through the signed-in Codex CLI’s own local app-server API.
Uses the Claude OAuth credentials already stored on this computer.
Reuses your signed-in Cursor desktop session from read-only local state.
Reads the local OpenCode database for per-step activity.
Today
Your token count sits against your own 28-day baseline rather than some average user’s. Below it you get the hours that produced those tokens, the tools that spent them, and the cost attached to both.

History
Every headline number carries its change against the period before it, so a heavy week reads as a heavy week instead of a figure with no past behind it.

Insights
All of it is built from session files already on your disk: the hours you reach for a model, the models you actually use, and how the load spreads across the week.

Two-hour blocks in local time, normalised to your busiest block, across every day the app has collected.

Share of tokens across every model your tools report, down to the ones you forgot you were still calling.

Which days carry the load, measured against your busiest weekday.

Limits and alerts
Set a remaining-capacity threshold per window, per tool. You get one desktop notification each reset cycle, and opening it drops you on the window that fired.

Every metered window, grouped by tool

Controls
Narrow to a single tool or open it back up to all of them, then move through days and ranges. Both menus behave the same way everywhere in the app.
Watch everything at once, or isolate one tool to see what it alone is costing you.

Today, yesterday, a specific date, or the last 7, 30, and 90 days, up to your full collected history.

Use UsageAtlas without an account. Create a free account to keep a copy of your usage history and restore it after changing or losing a computer. You control automatic saving in Settings.
Questions, answered
An AI coding usage tracker shows activity and capacity across your tools. UsageAtlas brings supported tokens, requests, models, estimated costs, limits, and reset times into one local desktop dashboard.
UsageAtlas currently supports Codex, Claude, Cursor, and OpenCode. Each source is added only after its sign-in path, error handling, and adapter tests are ready.
Yes. For supported provider windows, UsageAtlas shows remaining capacity and reset times, and can send one desktop alert per reset cycle at the threshold you choose.
Yes. UsageAtlas tracks supported Claude Code five-hour and weekly windows, Codex session and weekly windows, and Cursor plan, Auto + Composer, and API usage, plus local OpenCode activity.
No account is needed for local tracking. Create a free account with Google or GitHub to save usage history and restore it on another computer. Hourly saving is on by default after desktop sign-in while the app is running, unless you previously turned it off. You can also save and restore manually in Settings.
No. UsageAtlas processes bounded usage aggregates. Prompts and responses are never read, and a separate UsageAtlas account is not required.
Looking for a specific answer? Read the UsageAtlas guides, including Claude Code usage history, how to check Codex usage, and how to see OpenCode usage.
Download
UsageAtlas installs locally and picks up the supported tools already signed in on your computer. No account is required. Optional cloud save is free.
Squirrel installer with automatic updates for 64-bit Windows.
Signed and notarized disk images with automatic updates.
Runs on common 64-bit x86 and ARM Linux distributions.
Every release publishes SHA-256 checksums. macOS packages are signed and notarized by Apple. Windows installers are not code-signed yet, so Windows may show a SmartScreen prompt the first time you install.
Quick enough to check before you start a task, detailed enough for a proper look at the month once the invoice lands.