Linux source beta · Runs on your computer

Your books.
Your local workflow.

Your Kindle books, PDFs and page images → local OCR and Gemma review → Markdown, HTML, PDF and EPUB.

Local-first · Source language preserved

Download source ZIPInstallation guide

This is a distribution page, not an online OCR service. Nothing uploaded here is processed. Install the application locally with an NVIDIA GPU and Ollama; the standard setup downloads Gemma 12B locally. C1 OCR review is included; 26B is optional.

What it looks like

Click any image to enlarge it.

Start screen with PDF / Images and Kindle capture tabs, page range, GPU mode and the Start conversion button
1 · Start. Upload a PDF or page images, or switch to the Kindle capture tab.
Review screen: source page image beside Original OCR, Gemma suggestions and Changes tabs, with Approve and Restore buttons
2 · Review against the page image. Compare the page image with the suggested text; the original OCR is kept until you approve a page.On this page Gemma proposed "cannot" for "can neither". The image check did not support it, so the suggestion was deferred and the original reading kept — see the note above the Approve / Restore buttons.
Exported Markdown with source-page markers, and a question answered from it by a local model
3 · Use the Markdown. The export keeps a source-page marker per screen, so an LLM's answer can be checked against the page.

Image 2 shows a public-domain scan (Ruskin, The Stones of Venice, 1890 edition, page 50, via Wikimedia Commons); images 1 and 3 use a short synthetic text written for this documentation. The answer in image 3 came from the local gemma4:12b through Ollama. The images illustrate the interface, not an accuracy level.

Import

Use PDFs or page images, or connect a supported Kindle reader in Chrome. Source material stays local in the default Gemma workflow.

Review

Gemma 12B is the default; 26B is optional. Möbius Custom C1 for OCR is an application review profile, not new model weights.

Export

Download Markdown for model context, or create HTML, PDF and EPUB separately. Original OCR and unapproved suggestions remain distinct.

Start locally

Extract the ZIP, read the installation guide, then run bash install.sh and bash run.sh. Install Ollama first. Python 3.10/3.11 and compatible NVIDIA drivers are required. The installer prepares 12B; see the guide for Chrome and optional 26B setup.

12B / 26B comparison and limitations

Optional vision models

Gemma 4 remains recommended. An opt-in OpenAI-compatible connector lets you choose another vision-capable model. Enabling it sends images and OCR text to your chosen endpoint. API compatibility does not guarantee completion or accuracy. Read the connector setup and privacy notes.

Know the limits

OCR and model suggestions can be wrong. Check important names, numbers and quotations against source images. GPU compatibility, languages and browser layouts are not universally validated. Read the validation scope before relying on this beta.