An AI usage tracker collects the tokens, requests, models, limits and reset times from the coding tools you use and puts them in one view, so you know what you used and what is left. UsageAtlas is a free, open-source AI usage tracker that does this on your own computer for Claude Code, Codex, Cursor, OpenCode, GitHub Copilot and other coding tools.
Why one tool's own screen is not enough
Each coding tool reports its own numbers in its own place. Claude Code has /usage, which shows session cost, plan usage limits and activity stats. The Codex CLI has /status, which shows context usage and rate limits. Cursor and Copilot keep their numbers in account pages.
That works while you use one tool. Once your day moves between two or three, nobody adds the numbers up for you, and the screens show the current moment rather than last month. An AI usage tracker fills both gaps: it combines the tools and it keeps the history.
What a good AI usage tracker should show
Two kinds of numbers matter, and most built-in screens show only one of them.
| Number | What it answers | Where UsageAtlas gets it |
|---|---|---|
| Tokens and requests | What did I use, and with which model? | Local session history and signed-in sessions |
| Estimated cost | Roughly what was that worth at list prices? | Calculated from tokens where model pricing is known |
| Usage windows | How much is left before I get stopped? | Limits the provider reports, such as Claude's five-hour and weekly windows |
| Reset times | When does capacity come back? | The provider's reported reset, shown in local time |
| History | Is today normal for me? | Daily totals kept over 7, 30 or 90 days |
Tokens tell you about the past. Remaining capacity tells you whether the next long agent run will fit. You want both on the same screen.
How UsageAtlas covers the tools
Coverage depends on what each tool exposes:
- Claude Code, Codex and Cursor: usage windows and reset times where the provider reports them, plus token history. Claude uses the sign-in already on your computer, Codex goes through the signed-in Codex CLI, and Cursor reads your signed-in desktop session.
- OpenCode, Antigravity CLI, Pi / OMP and Muse Code: tokens, requests, models and sessions from supported local files. These sources do not report subscription limits.
- GitHub Copilot, Warp, Kimi Code, Kilo Code, Factory / Droid, Amp and Qoder: current allowances, balances and resets from a connected account. They do not give per-request token history.
The dashboard has a Today view, a per-tool filter, model mix, peak hours and a weekly heatmap. You can set an alert at a threshold you pick, and the app notifies you once per reset cycle before a window runs out.
Set it up in a few steps
- Download UsageAtlas for Windows, macOS or Linux from usageatlas.com.
- Open it. Detected tools appear on their own; connect the account-based ones under Settings, Sources.
- Pick an alert threshold for the windows you hit most.
- Use it for a week, then compare days against your own baseline.
No account is needed for local tracking. If you want history on a second computer, a free Google or GitHub sign-in saves and restores it.
What it does not do
UsageAtlas does not read prompts or responses; it works from usage totals. Its cost figures are estimates unless a provider reports the number directly, so the provider's invoice stays the record for billing. It is not a hosted analytics product and does not replace a provider's own dashboard.
If you only use Claude Code and like reports in the terminal, a CLI reporter such as ccusage may be all you need. Many people run both.