localize-fast: A CLI Tool for Bulk File Translation
Introduction
Translating multiple files while maintaining their structure is tedious. Most tools either require cloud services (raising privacy concerns) or manual copy-pasting (error-prone). I built localize-fast to automate this using local AI--no external servers, no data leaks.
How It Works
The tool maps directories and uses Ollama to run language models locally for translation:
- Local LLMs via Ollama: Models run offline using Ollama, with no data sent externally
- Fail-Safe: Stops and reports errors rather than producing broken files
Why Local LLMs?
LLMs have effectively renders many tools and business models obsolete. The most powerful open-source LLMs are fully capable of providing good translations of the most common language. Thus there is no reason to outsource this work to external services--local LLMs can do the work just fine!s
Trade-offs vs cloud services:
- No per-request costs
- Full data control
- Requires 8GB+ RAM
- Less nuanced than GPT-4
Get Started
Source available on GitHub. Basic usage:
# With Ollama running
translateall ./docs en fr ./output --chunk-size 3000
Current Limits
- Less powerful models sometimes fail to complete the task
- Model prompt engineering required
A simple tool for privacy-focused batch translation--not perfect, but openly improvable.