[jisbuidling]

Thanks! Part of the reason I built it was to feed a clean, token-efficient markdown to models. I used pdf.js as the foundation and added my own rules on top for layout, images, formulas and tables. For OCR I integrated PaddleOCR. I'm also looking at whether Chrome's built-in model could help on the vision side. For now it's purely browser-side, and a desktop version is on the list. If you'd want a desktop version, what would make it worth downloading over the site?

[charltonraven]

I would like the desktop version for privacy and automation(if capable). But either way this is nice!

[jisbuidling]

got it, for automation part think you need something like cli?

[charltonraven]

Yes, integrating that into your production features would be great. Make it a skill that can be called for documentation and reporting. I believe that would save tokens. For example, when monitoring Claude and its task when asked "Put this into a PDF", it will first create it in .md, then build a python script to convert the .md to .pdf. When really the script that it creates is "useless" after running it. Your tool can be used automatically by the AI to keep it from doing extra jobs for this task.