OpenTrans

PyPI - Python Version PyPI - Status

Documentation: https://rainstorm108.github.io/OpenTrans/

PyPI: https://pypi.org/project/opentrans/

Source Code: https://github.com/RainStorm108/OpenTrans

OpenTrans is a privacy-first, local-LLM batch translation utility powered by Ollama. It mirrors a source directory into a target language while preserving the exact file hierarchy, making it ideal for documenting software, translating blogs, or localizing static sites.

Key Features

Local-First (Ollama): Private, cost-free translation using models like Gemma, Llama 3, or DeepSeek.
Syntax Shielding: Automatically protects code blocks (``), inline code (), LaTeX math ($), and Markdown links from being corrupted by the LLM.
Directory Mirroring: Recursively replicates your source folder structure in the output destination.
Parallel Processing: Uses ThreadPoolExecutor for high-speed batch handling of large file sets.
Smart Caching: Uses hashing to track file changes. Only files that have been modified since the last run are sent to the LLM, saving significant time and compute resources.
Resilient File Handling: Automatically pulls required models from Ollama with a real-time progress bar if they are missing.

Installation

Prerequisites: You must have Ollama installed and running.

Option 1: Install via PyPI (Recommended)

pip install opentrans

Option 2: Install via uv

uv tool install opentrans

Option 3: From Source

git clone https://github.com/rainstorm108/OpenTrans.git
cd OpenTrans
uv tool install .

Usage

Basic Command

The simplest way to run OpenTrans is by specifying the input and output directories:

opentrans ./docs ./translated_docs --config ./config.yaml

Example Output:

Target Language: Chinese
Input:  /home/user/Projects/OpenTrans/docs
Output: /home/user/Projects/OpenTrans/translated_docs
Using Model: translategemma:4b

Translating Files: 100%|████████████████████████████████| 12/12 [00:45<00:00, 3.7s/file]

Complete.

Configuration (config.yaml)

OpenTrans relies on a YAML configuration file to define the model and target language. See example.

Running Examples

opentrans ./Example/Docusaurus/docs ./Example/Docusaurus/i18n/zh-hans/docusaurus-plugin-content-blog/current ./config.yaml

Local Development

Clone the repository

git clone [https://github.com/rainstorm108/OpenTrans.git](https://github.com/rainstorm108/OpenTrans.git)
cd OpenTrans

Set up the environment (using uv)

uv sync
source .venv/bin/activate

Install in editable mode

uv pip install -e .

Run tests
```
hatch test
```

Project Structure

OpenTrans/
├── src/
│   └── opentrans/
│       ├── __init__.py        
│       ├── main.py            # CLI interface using Click
│       ├── settings.py        # Configuration management
│       ├── translator.py      # Core translation logic
│       ├── cache_manager.py   # Caching system
│       └── hasher.py          # File content hashing
├── tests/
├── config.yaml                
├── pyproject.toml
└── README.md

Workflow

graph TD
    %% Setup Phase
    Start((Start)) --> CheckRunning{Ollama Running?}
    CheckRunning -- No --> Err[Error: Start Ollama]
    CheckRunning -- Yes --> CheckModel{Model Exists?}

    CheckModel -- No --> Pull[Download Model with Progress Bar]
    Pull --> Init
    CheckModel -- Yes --> Init[Load Paths & Config]

    %% The Loop
    Init --> NextFile{Next File?}
    NextFile -- No --> Exit([Exit])
    NextFile -- Yes --> IsTranslatable{File Type Supported?}

    %% Branching Logic
    IsTranslatable -- No --> Mirror[Copy Original File]
    IsTranslatable -- Yes --> Shield[Apply Syntax Placeholders]

    Shield --> Trans[Translate via Ollama API]
    Trans --> Restore[Restore Protected Code/Math]
    Restore --> Save[Create Dir & Save File]
    Save --> NextFile

Todo

[x] Click-based CLI interface
[x] Placeholder-based syntax protection (Code/LaTeX)
[x] Multi-threaded parallel processing
[x] Implement hash caching to skip unchanged files
[ ] Finish the Docusaurus translate script
[ ] User Tree-sitter to replace the code blocks before translation instead of regex
[ ] Add support for more file types
[ ] ...

License

This project is licensed under the MIT License.