Arn Manga LLM Translator — Privacy Policy

Detection, OCR and drawing run entirely on your device. The only data that leaves
your browser is what translation requires: the page image and/or recognized text,
sent to the AI provider or cloud endpoint YOU configure, using YOUR API key.

• On-device: text detection, panel ordering, OCR (Baberu/Tesseract), redrawing,
  the translation cache and character book. Model weights (~50 MB) download once
  from public mirrors and are cached in IndexedDB.
• Sent out, and only when you translate: page image + text → your chosen LLM
  provider (OpenAI-compatible / OpenAI Responses / Anthropic / Gemini, or a custom
  base URL); page image (+ OCR text if that mode is on) → the cloud endpoint you
  pasted (optional, off by default); model/font/OCR-data files → public mirrors
  (huggingface.co, cdn.jsdelivr.net, raw.githubusercontent.com); chapter ID only →
  api.mangadex.org (MangaDex only, to resolve the manga title).
• API keys are stored only in your browser (chrome.storage.local) and sent only to
  the endpoint above over HTTPS. No developer server, no accounts, no analytics,
  no ads, no selling or model training by us — we never see your data.
• You can clear the cache/character book anytime (popup → Clear, or options →
  Advanced). Uninstalling removes everything stored locally.