Question 1

Is it free?

Accepted Answer

Yes. There is no backend, so there is no per-request cost on the host side. Static assets and model weights are served by public CDNs.

Question 2

What model is used?

Accepted Answer

PaddleOCR PP-OCRv5 mobile, via ppu-paddle-ocr and onnxruntime-web.

Question 3

What about large PDFs?

Accepted Answer

Pages are rasterised one at a time and processed in parallel by the worker pool. Files above 500 MB require explicit confirmation to avoid out-of-memory failures.

Question 4

Can I embed this?

Accepted Answer

Yes. A hidden iframe + postMessage SDK is published at /embed.html and /embed-client.js. See /embed-demo.html.

Question 5

Is the source code open?

Accepted Answer

github.com/rotkonetworks/ocr.rotko.net.

Question 6

Are documents uploaded anywhere?

Accepted Answer

No. The pipeline is entirely client-side. Verify in the browser's network tab during processing.