getomni-ai/zerox ? reverse-engineered prompt

Build me a simple OCR tool that takes a PDF, DOCX, or image file, turns each page into images, sends them to a vision model, and returns clean Markdown that I can use for AI ingestion.

I want it to work from both Node and Python, with a small API that accepts a local file path or a file URL, plus my model credentials. It should handle weird layouts like tables, charts, invoices, and multi page documents as well as it can, and let me choose things like which pages to process, whether to keep the formatting consistent across pages, and whether to return structured data from a schema. If possible, add a way to save the final result to a file too.

Please make the defaults sensible, handle errors without crashing the whole run unless I ask for that, and include a couple of examples I can run right away. If you need to check the latest docs for any model provider, feel free to look them up online.