It is probably not worth the complexity currently but considering they are using small local CPU models for OCR like tesseract, if they add the support of reading files on the web then I wouldn't be so sure of the CPU bound aspect.