Argos Translate

Offline neural machine translation library and CLI with installable language packs for over 35 languages.

Open SourceSelf HostedOffline Capable
0.0 (0)

About

Offline translation is the whole point of Argos Translate, a Python library and CLI that runs neural machine translation entirely on the local machine. It builds on OpenNMT with a CTranslate2 inference backend and distributes trained models as installable language packs covering more than 35 languages; when a direct pair is missing, it pivots automatically through an intermediate language, so Spanish to French works with only Spanish-English and English-French packs installed, at some cost in quality. Everything installs with pip install argostranslate, a separate argostranslategui package adds a desktop interface, and setting the ARGOS_DEVICE_TYPE environment variable to cuda or auto enables GPU acceleration, though CPU inference is the default and works fine. The project is dual licensed under MIT or CC0, has about 6,300 stars, and documents its API on Read the Docs. Its highest-profile deployment is LibreTranslate, which wraps it in a self-hosted web API, and privacy-sensitive applications choose it because no text ever leaves the machine.

Should you use Argos Translate?

Pick it when

Pick it when translation has to run inside your own Python app or script with no network calls, for example on a laptop, an air-gapped server, or a privacy-sensitive product, and CPU-only machines are all you have.

Look elsewhere when

Skip it if you want a ready HTTP service for other apps or teammates, where LibreTranslate wraps this engine. Pairs without a direct pack pivot through an intermediate language and lose quality, so check your pairs first.

Alternatives to Argos Translate

  • LibreTranslate

    Same engine behind a REST API, web UI and Docker images, easier to share across services, but AGPL-3.0 instead of MIT or CC0 and more to operate.

  • CTranslate2

    The inference engine Argos runs on; use it directly to serve your own translation models with more control, but no language packs or pivoting.

  • Transformers

    Runs larger multilingual models such as NLLB with more language pairs, but heavier PyTorch downloads, no packaged offline packs, and NLLB is non-commercial.

Reviews (0)

Leave a Review

No reviews yet. Be the first to review!

Details

Price
Free
Platform
Local/Desktop
Difficulty
Easy (2/5)
License
MIT OR CC0-1.0
Added
Jul 29, 2026

Tags