Hey Product Hunt — I built OOCR because I kept seeing "free online OCR" tools where the free part meant your document was the product. OOCR is different: PP-OCRv6 (PaddleOCR) runs entirely in your browser via WebAssembly. There is no server. Open DevTools → Network during recognition — you'll see zero requests containing your image. The absence is auditable. How it works: 1. Paste a screenshot (Ctrl+V) — lands in the workspace, never transmitted 2. PP-OCRv6 draws bounding boxes over recognized text (WebGPU accelerated where available) 3. Copy the text or export a searchable PDF with an invisible, copyable text layer 4. All assembled client-side — no server, no API, no cloud Why build this? Privacy by architecture, not by policy. When a site says "we delete your files after 60 minutes," that's a promise — a setting someone can flip. When a site has no upload endpoint, there's nothing to flip. The Network tab is your proof. Technical stack: SvelteKit, ONNX Runtime Web (WebGPU preferred, WASM fallback), PP-OCRv6 (three tiers: Tiny ~4MB, Small ~10MB default, Medium ~22MB best accuracy), jsPDF for searchable PDF assembly. Self-hosted models on Cloudflare Pages. Entirely open source. Would love feedback on: accuracy expectations, mobile UX (it works on phones but the UI was designed desktop-first), and what file formats you'd want supported next.

oocr.app

OCR that never leaves your browser

OCR that never leaves your browser