Hey everyone! I built chunks.md to make OCR fast, private, and hassle-free. Drop a PDF or image into your browser and get clean markdown back on desktop or mobile. Under the hood it runs 7 different OCR models via ONNX Runtime Web — from PaddleOCR v5 for general documents to manga-ocr for Japanese text. You pick the model, drop your files, and get structured text ready to paste into ChatGPT, Claude, or feed into your RAG pipeline. A few things that make it different: Fully on-device; all inference runs in a Web Worker, your files never leave your machine. Works on mobile; same full OCR experience on your phone or tablet. No account, no limits; just open the site and use it. Multiple OCR engines; PaddleOCR v5/v3, SmolDocling, PaddleOCR-VL, and manga-ocr so you can pick the best model for your content. Fully customizable; every OCR model exposes its settings (detection thresholds, crop padding, token limits, prompts, and more) so you can tune results for your specific documents. AI-native output; clean markdown chunks designed for LLM consumption. I built this for my own workflow and figured others might find it useful too. Would love to hear what you think!

chunks.md

Turn your files into clean markdown. Processed locally.

Turn your files into clean markdown. Processed locally.