Detect
rewr.it/detect scores a draft and shows you what it found. It does not rewrite anything.
The words never leave the tab
The scoring runs in your browser. POST /api/detect is called afterwards and it
receives the result — a score, a word count, a handful of counts. Not the text.
That is enforced by the endpoint, not promised by a comment: the route accepts a body of at most 4KB containing numbers and enumerated verdicts. There is no field a draft could arrive in.
Limits
Detection costs nothing to run, so the caps exist to stop a script pointing at a free oracle, not to ration anything.
| Plan | Runs per rolling 24h |
|---|---|
| Anonymous | 3 |
| Free | 20 |
| Pro | 100 |
There is also a per-IP ceiling of 100 a day, which a cookie-cycling script hits and a household does not.
Reading the result
Same two dimensions and same bands as the tells report. The AI number is the one about machine authorship; the voice number is about sounding like the target voice.
A high score means the text contains patterns models produce often. It does not mean a model wrote it, and no threshold makes it mean that. People who write carefully in a formal register score higher than people who do not. Used on a student's essay or a colleague's memo, this produces a number with an accusation-shaped hole around it, and the number cannot fill it.
It is built for looking at your own draft before you send it.
No CORS
/api/detect has no CORS headers. The only client is the same-origin page. If
you want scoring in your own tool, the checker is client-side code in the repo —
bundle it and run it locally, which is what the extension does.