🔍 Extract text from images entirely in your browser — no upload, no API keys, no privacy concerns.

🎙️ Try it now → https://naklitechie.github.io/ScanLocal/

• Five OCR engines in one tool:

  • 📜 Default (Tesseract) — Most reliable browser path, broad language support
  • 🚀 Fast (PaddleOCR) — Printed text path via Paddle models
  • ✍️ Handwriting (TrOCR) — SOTA for handwritten text, ~500 MB
  • 🧠 Smart OCR — Transformer fallback for printed text, no boxes
  • 🔭 Qwen2-VL (Experimental) — Best-effort visual document reader, very large download

• 100% client-side — Your images never leave your device

• No dependencies — Single HTML file, no npm, no bundler

• Offline-ready — Models cached after first load

• Privacy-first — No tracking, no analytics, no data collection

• Receipts and invoices
• Scanned documents
• Business cards
• Screenshots with text
• Handwritten notes
• Historical documents

1. Open the app in your browser
2. Choose an engine:
   • Default for the safest first try in-browser
   • Fast for receipts, documents, screenshots
   • Handwriting for handwritten notes
   • Smart for an alternate transformer-based read on printed text
   • Qwen for experimental visual reading of complex documents
3. Pick a Tesseract language if you're using the default engine
4. Drop an image or click to upload
5. Wait for text extraction
6. Copy the extracted text

# Just open index.html in your browser
open ScanLocal/index.html

cd ScanLocal
python3 -m http.server 8000
# Open http://localhost:8000

The app is automatically deployed to:
https://naklitechie.github.io/ScanLocal/

• Images: PNG, JPG/JPEG, WebP
• Max size: 10 MB per image
• Languages: 100+ (varies by engine)
• Tesseract languages in UI: English, Spanish, French, German, Italian, Portuguese, Russian, Hindi, Arabic, Japanese, Korean, Chinese (Simplified)

Coloured with italy-02 · PERGAMENA — Renaissance manuscript parchment, Vatican library. Yellowed-cream body, ink-brown text, Vatican-cobalt action accent — the place where text is preserved and read, fitting for an OCR tool.

Palette pulled from Rangrez, the global colour-palette library that backs all NakliTechie projects.

ScanLocal is part of the NakliTechie series — browser-native tools that keep your data on your device.

• ❌ No server uploads
• ❌ No API keys
• ❌ No tracking
• ❌ No analytics
• ✅ All processing happens in your browser
• ✅ Models cached locally after first load

MIT License — see LICENSE file.

ScanLocal follows the same blueprint as:

• BabelLocal — Browser translator (200 languages)
• StripLocal — EXIF metadata stripper
• GambitLocal — Chess vs Stockfish
• VoiceVault — Audio transcription
• KingMe — Checkers vs AI
• SnipLocal — Background remover

All single-file, zero-dependency, browser-native apps.

ScanLocal/
├── index.html          # The entire app (shipped)
├── README.md           # This file (shipped)
├── LICENSE             # MIT license (shipped)
├── .gitignore          # Git ignore rules
└── PLAN.md             # Planning notes (not shipped)

None! Just edit index.html and refresh your browser.

# Enable GitHub Pages on the repository
# Settings → Pages → Source: main branch → /root
# Push to main branch
git push origin main

Issues and PRs welcome! Please read IDEAS.md for the series guidelines before contributing.

Chirag Patnaik
NakliTechie

See PLAN.md for detailed implementation plan.

Phase 1 (Current): Core infrastructure ✅

• Base HTML skeleton
• Engine selector UI (4 engines)
• Dropzone + preview
• PaddleOCR integration
• TrOCR integration
• Tesseract.js integration
• Smart OCR integration

Phase 2: Engine integration Phase 3: UI polish Phase 4: Testing Phase 5: Launch