# Document Localizer

AI-powered desktop application for offline document localization and translation between 100+ locales. Entirely local - documents never leave the user's machine.

## What It Does

Transforms PDF and Markdown documents between locales (languages and regional variants) using local AI inference. Supports hundreds of locale pairs including English variants (en-US, en-GB, en-AU, en-CA, en-IN, en-NZ, en-IE, en-ZA, en-SG), Spanish variants (es-ES, es-MX, es-AR, es-CO), Portuguese (pt-PT, pt-BR), French (fr-FR, fr-CA), German (de-DE, de-AT, de-CH), and 90+ more. Translation quality depends on the local AI model installed.

## Core Features

- **Locale Selection**: 100+ locales covering English, Spanish, Portuguese, French, German, Italian, Chinese, Japanese, Korean, Arabic, Hindi, and many more
- **Privacy-First Local Translation**: All processing happens locally via Ollama, LM Studio, or llama.cpp - no cloud API calls, no telemetry, no data collection
- **PDF and Markdown Support**: Read PDFs and Markdown files, output to PDF or Markdown
- **Diff Review**: Side-by-side comparison view with paragraph-level approve or reject before export
- **Accuracy**: AI translations can contain errors. Output quality depends on the local model installed. Review every change in the diff view and approve it before relying on the output. This tool is not a substitute for a qualified human translator.
- **OpenAI-Compatible**: Works with any local model exposing an OpenAI-compatible API (Llama 3, Mistral, Phi, Gemma, Qwen, and more)
- **Cross-Platform**: macOS, Windows, and Linux

## Use Cases

1. **Content Localization**: Convert documentation, marketing copy, and product pages for international audiences
2. **Regional Adaptation**: Transform content between regional variants (en-US to en-GB, es-ES to es-MX, fr-FR to fr-CA)
3. **Privacy-Sensitive Translation**: Translate confidential documents (legal, medical, financial) without sending data to third-party APIs
4. **Offline Workflows**: Translate documents on air-gapped machines or in restricted-network environments
5. **Cost-Free Translation**: Avoid per-character cloud translation costs by running local models
6. **Publishing Preparation**: Adapt content for specific regional markets before publication

## Technical Specifications

- **Input formats**: PDF, Markdown (.md)
- **Output formats**: PDF, Markdown (.md), Word (.docx)
- **Export caveat**: PDF export flattens formatting - headings, lists, links, tables, and code blocks are stripped to plain text. Use Markdown or Word export to keep structure.
- **AI backends supported**: Ollama, LM Studio, llama.cpp (any OpenAI-compatible local server)
- **Recommended models**: Llama 3.2 (3B for low-end, 8B for quality), Mistral 7B, Phi-3, Gemma 2, Qwen 2.5
- **System requirements**: 4GB RAM minimum (8GB recommended for 7B+ models), ~5GB disk per model
- **Supported operating systems**: macOS, Windows, Linux
- **Application type**: Electron desktop app (offline)
- **Network access required**: No (except to download the app and models)
- **Telemetry**: None

## Comparison

Document Localizer vs. Cloud Translators (DeepL, Google Translate, Microsoft Translator):

- **Data Privacy**: All processing happens locally on your GPU. Your documents never leave your machine. Cloud translators send every document to their servers.
- **Cost**: Free and open source. Run on your own hardware. Cloud translators use subscription and metered per-character billing.
- **Formatting Retention**: Markdown structure carries through on Markdown and Word export. Code blocks and tables are dropped on PDF export. Cloud translators often break complex layouts and code formatting.
- **Custom Models**: Plug in any OpenAI-compatible local LLM (Llama 3, Mistral, Gemma, Qwen, etc.). Cloud translators offer black-box models with zero customization.
- **Diff & Review**: Built-in side-by-side diff editor lets you review and approve translations paragraph-by-paragraph before export. Cloud translators give you raw output with no review workflow.
- **Air-Gapped**: Works 100% offline with no internet connection required. Cloud translators require a constant connection.

Document Localizer is NOT:
- A CAT tool (Computer-Assisted Translation, like Trados or memoQ) - it is a fully automated AI translator
- A Translation Management System (TMS) - it is a single-user desktop app
- Cloud-based - it has no server component and no account system
- A general-purpose LLM frontend - it is purpose-built for document localization with diff review

Document Localizer IS:
- A privacy-first alternative to DeepL, Google Translate, and Microsoft Translator for users who cannot send documents to third-party APIs
- A free alternative to commercial CAT tools for one-off document translation
- A consumer of any OpenAI-compatible local model server (Ollama, LM Studio, llama.cpp, vLLM, etc.)

## Installation

1. Download the latest release for your platform: https://github.com/zimablue-io/document-localizer/releases/latest
2. Install a local AI backend:
   - Ollama (recommended): https://ollama.com
   - LM Studio: https://lmstudio.ai
   - llama.cpp: https://github.com/ggerganov/llama.cpp
3. Pull a model, for example: `ollama pull llama3.2:3b`
4. Launch Document Localizer, enter your API URL (e.g. `http://localhost:11434/v1`) and model name in Settings
5. Select source and target locales, then process your documents

## Workflow

1. **Upload**: Select PDF or Markdown files to add to your source library
2. **Configure**: Choose source locale and target locale
3. **Process**: AI localizes content in chunks locally via the selected backend
4. **Review**: Side-by-side diff view shows changes with paragraph-level accept or reject
5. **Export**: Export the approved document as Markdown or PDF

## Keywords and Topic Clusters

Primary topics: offline document translation, local LLM translation, privacy-first translation, AI document localizer, on-device translation, Ollama frontend, LM Studio GUI, llama.cpp translation app, PDF localizer, Markdown translator, English variant converter, British to American English, American to British English, regional language adaptation.

## Project Information

- **Website**: https://zimablue-io.github.io/document-localizer/
- **Repository**: https://github.com/zimablue-io/document-localizer
- **Issues**: https://github.com/zimablue-io/document-localizer/issues
- **Releases**: https://github.com/zimablue-io/document-localizer/releases
- **Changelog**: https://github.com/zimablue-io/document-localizer/blob/main/CHANGELOG.md
- **License**: https://github.com/zimablue-io/document-localizer/blob/main/LICENSE.md

## Licensing

Copyright (c) 2026 zimablue-io. Free for personal and non-commercial use. Commercial use requires written permission. Full license details in LICENSE.md at the repository root.