https://github.com/amitxm/AIDocPrep is the one I built that is true local OCR. I made sure there is NO internet use, even accidentally. Also, added redaction.
I built a free tool(https://mdpdf.net), based on the fast PDF conversion library (https://github.com/firecrawl/pdf-inspector). It runs entirely in the user’s browser. In my view, the biggest pain points when converting PDF to Markdown are whether it can properly handle content like tables and code blocks, as well as OCR. Current OCR approaches don’t seem very effective unless you use an LLM.
Also, it seems disingenuous to have “locally” in the title when your default is cloud inference providers and I don't see you naming a recommended LLM that could be hosted locally and is proven to work well for this task. You list “Your files” under “Stays on your machine” and then require an API key for a cloud provider.
Worth noting there’s also textsnap if you want something that’s actually local end-to-end via ONNX on CPU, no GPU, no API key, nothing leaves your machine. https://github.com/kouhxp/textsnap
https://github.com/amitxm/AIDocPrep is the one I built that is true local OCR. I made sure there is NO internet use, even accidentally. Also, added redaction.
Seems like it sends your documents to the LLM of your choice. In what world is that “locally”??
LLMs have been self-hostable for a while now.
I am aware, but unless it includes a model and model runner, it’s not offering local it’s allowing local.
Why would I just not route my docs to my local vlm/llm directly? which I already currently do.
That is an entirely different question. If you like your current solution, there's no reason to switch.
I built a free tool(https://mdpdf.net), based on the fast PDF conversion library (https://github.com/firecrawl/pdf-inspector). It runs entirely in the user’s browser. In my view, the biggest pain points when converting PDF to Markdown are whether it can properly handle content like tables and code blocks, as well as OCR. Current OCR approaches don’t seem very effective unless you use an LLM.
How does this differ from open source cli tools such as Marker (https://github.com/datalab-to/marker)?
Also, it seems disingenuous to have “locally” in the title when your default is cloud inference providers and I don't see you naming a recommended LLM that could be hosted locally and is proven to work well for this task. You list “Your files” under “Stays on your machine” and then require an API key for a cloud provider.
Worth noting there’s also textsnap if you want something that’s actually local end-to-end via ONNX on CPU, no GPU, no API key, nothing leaves your machine. https://github.com/kouhxp/textsnap
Shoutout to marker-2. By far the best OCR system, IME. The amount of annoying little edge cases they know about and cover for you is dizzying
As a service: https://into.md/
Docling is also a tool to consider, purely local, even though it's now AI-friendly too: https://docling.ai/
Have someone actually used it? Any feedback? There are a lot of alternatives at the moment, all of them break on complex excel documents
I am also building an app that converts web pages to Markdown. How do you handle messy scans and tables?
i really need this tool like these concepts