Ratica

Nothing is uploaded

Read any book
in your language.

Ratica translates whole PDF books on your own computer with a local AI model, and keeps every page looking like the original. Free and open source.

See it

The same page. Another language.

Drag the handle. Headings, the list, the chart labels and the note box stay exactly where they were.

Original page in English
The same page translated by Ratica
OriginalTürkçe

Demo page written for this site. Translated by Ratica 0.1 with the default model.

How it works

Three steps. No account.

  1. 1

    Open a PDF

    Choose a book or drop it on the window. Ratica recognises the language it is written in.

  2. 2

    Pick a language

    Choose one of 23 languages and press Translate. Pause and resume whenever you like; progress is saved.

  3. 3

    Read the translated book

    You get a PDF in the original layout and an EPUB for e-readers, saved next to your PDF.

Ratica window with a book chosen, Spanish selected and the estimated time shown
Choose a book and a language
Ratica window after a finished translation with buttons to open the PDF and the EPUB
Open the PDF or the EPUB

On first start Ratica downloads its translation engine and AI model once (about 5–6 GB) and measures your computer's speed. After that it works offline.

Why Ratica

Made for long books, not snippets.

Keeps the layout

Every page stays where it was: headings, tables of contents, lists, figures, colours and page numbers.

Private by design

The AI model runs on your computer. Your books and text are never uploaded.

No cost, no API key

No account, no subscription and no per-page fee. Free and open source.

Code & formulas stay intact

Code, URLs and formulas are left exactly as they are, so technical books stay correct.

PDF and EPUB

Read the translation as the original pages, or as an EPUB that reflows on phones and e-readers.

Uses your GPU

Picks CUDA, Metal or Vulkan by itself and leaves room for your other work.

What's new

Latest release

Full release notes

Loading the notes of the latest release… If nothing appears, read them on GitHub.

Languages

23 languages, in any direction.

  • Azerbaijani
  • Bulgarian
  • Chinese (Simplified)
  • Czech
  • Dutch
  • English
  • French
  • German
  • Greek
  • Hindi
  • Indonesian
  • Italian
  • Japanese
  • Korean
  • Polish
  • Portuguese
  • Romanian
  • Russian
  • Spanish
  • Swedish
  • Turkish
  • Ukrainian
  • Vietnamese

Requirements

Runs on the computer you have.

NVIDIA GPU, 6 GB or moreRecommended · CUDA
Apple Silicon Mac (M1 or newer)Supported · Metal
AMD or Intel GPUShould work · Vulkan, not yet tested
No GPUWorks, but slow

Windows 10/11, macOS 12 or newer, or Linux, with about 6 GB of free disk space. On an entry-level 6 GB GPU a 200-page book takes about 30–45 minutes.

FAQ

Questions

Is Ratica free?

Yes. Ratica is free and open source under the AGPL-3.0 license. There is no account, subscription or API key.

Does Ratica upload my books?

No. Translation runs on your own computer. Ratica only goes online on first start to download its translation engine and AI model (about 5–6 GB). It never uploads your files or text.

Which AI model does it use?

Gemma-4-E4B-it (Apache 2.0) through llama.cpp. It was chosen after comparing several open models on translation quality, protection of code and URLs, speed and memory use.

How long does a book take?

On an entry-level GPU with 6 GB of memory, a 200-page book takes about 30–45 minutes. Faster GPUs finish sooner. Without a GPU it takes many hours.

Can it translate scanned PDFs or comics?

Not yet. Ratica translates PDFs that contain real text. Scanned pages need OCR, which is planned. Text inside images and speech bubbles stays in the original language for now.

Why does Windows show a warning?

The Windows installer is not code-signed yet, so SmartScreen may warn the first time. Choose More info → Run anyway. The macOS app is signed and notarized by Apple.

How do I uninstall it?

The engine and model live outside the app. Delete them from the bottom of Ratica's window, or let the Windows uninstaller remove them. Then remove the app as usual.

Open source, built in the open.

AGPL-3.0. Built with llama.cpp, Gemma, PyMuPDF and PySide6. Hardware reports and translation feedback in any language are very welcome.