Track AI usage across tools with one desktop monitor

The way to track AI usage across tools is to let one app read every coding tool you use and show the results side by side, with alerts before a limit runs out. UsageAtlas is the free, open-source app built for that: it watches Claude Code, Codex, Cursor, OpenCode, GitHub Copilot and more from your desktop, and you set it up once.

Why per-tool checking breaks down

Checking each tool by hand works for a week. Then you forget to look at Codex before a long run, or you find out about the Claude weekly limit when it stops you. The problem is not a lack of data. Each provider shows its own numbers, in its own place, only when you ask.

Monitoring fixes that the same way it fixes it for web pages: one place that watches for you and speaks up when something needs attention.

Step 1: Install one tracker

Download UsageAtlas for Windows, macOS or Linux. It runs in the tray, needs no account for local tracking, and reads usage totals rather than your prompts or responses.

Step 2: Let it find your tools

On first run it picks up what is already on your computer:

Claude Code, through the sign-in already stored on the machine
Codex, through the signed-in Codex CLI
Cursor, from your signed-in desktop session
OpenCode, Antigravity CLI, Pi / OMP and Muse Code, from supported local session files

For GitHub Copilot, Warp, Kimi Code, Kilo Code, Factory / Droid, Amp and Qoder, connect the account under Settings, Sources. Those connections report allowances and balances, not token history.

Step 3: Read the combined view

Once the tools are in, the Today view shows tokens, requests and estimated cost across all of them, plus a capacity gauge for each usage window a provider reports. Use the per-tool filter when one service looks off. A tool that reports only a balance, such as Warp or Amp, shows what is left rather than a token count, so read those rows as a fuel gauge, not a log.

What you want to know	Where to look in UsageAtlas
Total use today across tools	Today view
Which tool or model did most of the work	Per-tool filter and model mix
How close each plan is to its limit	Capacity gauges and reset times
Whether this week is unusual	7, 30 and 90 day history
When you tend to work hardest	Peak hours and weekly heatmap

Step 4: Set alerts per window

Pick a remaining-capacity threshold for the windows you care about, such as Claude's weekly limit or a Codex session window. UsageAtlas sends one desktop notification per reset cycle when a window reaches it, which is enough time to switch tools or save the long task for later. Start with a conservative threshold for the first week, then move it once you know how fast a normal day drains each window.

Step 5: Keep history if you change machines

Local history stays on your computer. If you want it to survive a new laptop, sign in with a free Google or GitHub account and the app saves usage summaries to restore later. Project names, file paths, prompts and credentials are left out of those saves.

Limits to know about

Provider-reported windows and resets can be delayed or changed by the provider, and some tools do not report limits at all. UsageAtlas labels cost figures as estimates, so check billing with the provider. It is a personal desktop tracker, not a team analytics service.

If you only use one tool, its built-in command, /usage in Claude Code or /status in the Codex CLI, may be enough. Once you track AI usage across tools, the desktop view saves the daily round of checks.

FAQ

Can one app track Claude Code, Codex and Cursor together?

Yes. UsageAtlas reads all three, plus OpenCode and other tools, and shows them in one desktop dashboard with a per-tool filter.

Will I get a warning before I hit a usage limit?

If the provider reports the window, yes. You pick a threshold per window and UsageAtlas sends a desktop notification once per reset cycle.

Does tracking need an account?

No. Local tracking works without one. A free account is only needed to save history and restore it on another computer.