AI rate limits in coding tools come in two kinds: plan quotas that run out over a window and then reset, and short request throttles that clear in seconds. The easiest way to keep the first kind in view across Claude Code, Codex, Cursor, GitHub Copilot and the rest is UsageAtlas, a free, open-source desktop app that shows remaining capacity and reset times side by side.
Two kinds of AI rate limits
Most "limit reached" moments are plan quotas. Your subscription includes a certain amount of work per window, and when that is used up you wait for the reset, buy more, or move to a bigger plan. The window can be a rolling few hours, a week or a calendar month, depending on the tool.
The other kind is a throttle. An API key, a cloud project or the provider's own servers can refuse requests that arrive too fast, usually with an HTTP 429 status. These clear on their own and do not use up your plan. Telling the two apart saves a lot of pointless waiting.
How the main tools measure usage
Providers change the size of their limits often, so the table below covers the shape of each limit, not the numbers. Check the linked docs for your plan's current figures.
| Tool | What is metered | Reset clock | Where to check |
|---|---|---|---|
| Claude Code (Pro, Max) | Session and weekly limits, shared with Claude chat | Rolling five hours, plus weekly | /usage |
| Codex | Messages per window, by model and task size | Five-hour window on some plans, weekly limits may apply | /status, Codex usage page |
| Cursor | Included usage, then credit or on-demand usage | Weekly pool on the newer subscription, monthly on legacy plans | Cursor dashboard |
| GitHub Copilot | AI credits for chat, agents and CLI | First of each month, 00:00 UTC | GitHub billing settings, IDE status icon |
A few details matter more than the numbers. Claude Code's session and weekly limits count at the same time, so a heavy burst can drain the weekly allowance early. Codex's Pro plans currently have no five-hour limit. Copilot code completions are not billed in credits on paid plans. Cursor's reset depends on which plan type you are on.
See every limit in one place with UsageAtlas
Each provider shows its own number in its own place. If you switch between two or three tools in a day, that means several commands and dashboards before you know whether a long agent run will finish.
UsageAtlas puts the supported windows from Codex, Claude Code, Cursor, GitHub Copilot, Warp, Kimi Code, Kilo Code, Factory / Droid, Amp and Qoder on one desktop screen, with remaining capacity and the reported reset time for each. Pick a remaining-capacity threshold per window and it sends one alert in that reset cycle, then stays quiet. It also keeps daily and historical views of tokens, requests, models and estimated cost.
Some sources, such as OpenCode, Antigravity CLI, Pi / OMP and Muse Code, add local activity only. They do not report quota windows, so there are no reset clocks or alerts for them.
UsageAtlas is free and open source under the MIT license, runs on Windows, macOS and Linux, and needs no account for local tracking. It does not read prompts or responses.
Limits of the view
Provider figures can lag or change, and cost values are estimates. Treat the dashboard as a planning view. If a deadline depends on the last few percent of a window, confirm it in the provider's own page too; that page stays the record for billing.
FAQ
Are AI rate limits the same as API rate limits? Not quite. Plan quotas cap how much you can do per window; API rate limits cap how fast requests can arrive. A 429 usually means the second kind.
Does UsageAtlas raise my limits? No. It shows the limits your providers report and warns you before you reach them.
Is UsageAtlas free? Yes. It is free and open source, and the source is on GitHub.