AI Translator

Translate text between English, Spanish, Portuguese and Indonesian — powered by open-source Opus MT models running locally in your browser. No upload, no API keys, fully private.

All translation happens in your browser. Your text never leaves your device. Models download once (~30MB each) and cache locally.

Translate Your Text

0 characters · 0 words
0 characters · 0 words

Runs entirely in your browser. Your data never leaves your device.

How It Works

When you click Translate, three things happen — all locally, all in your browser:

  1. Model selection — For English ↔ Spanish and English ↔ Indonesian we use a dedicated, fast Opus MT model (~30MB). For any Portuguese direction we use Meta NLLB-200, a multilingual model that translates directly between 200 languages — the standard choice for in-browser pt translation.
  2. Model load (first time only) — The model weights are fetched from a Hugging Face CDN and persisted to IndexedDB. Subsequent visits load instantly from cache.
  3. Translation — Your text is tokenized and run through the transformer decoder. Output is detokenized and displayed. For batch mode, each paragraph is translated independently.

Supported Language Pairs

DirectionModel
English → Españolopus-mt-en-es (Opus MT)
Español → Englishopus-mt-es-en (Opus MT)
English → Indonesiaopus-mt-en-id (Opus MT)
Indonesia → Englishopus-mt-id-en (Opus MT)
English → Portuguêsnllb-200-distilled-600M (NLLB-200)
Português → Englishnllb-200-distilled-600M (NLLB-200)

Frequently Asked Questions

Is my text uploaded to a server?

No. This AI translator runs entirely in your browser using WebAssembly. Your text never leaves your device. Models are downloaded once from CDN and cached locally.

Which language pairs are supported?

We support 4 languages: English, Spanish, Portuguese and Indonesian. The English–Spanish and English–Indonesian directions use a dedicated Opus MT model (~30MB each); the Portuguese direction uses the Meta NLLB-200 multilingual model (~600MB, downloaded once and cached).

How accurate are the translations?

Opus MT models are fine-tuned on millions of sentence pairs from OPUS datasets. They produce fluent translations for general-purpose text (news, emails, documentation).

Can I use this offline?

Yes, after the first use. Each language-direction model (~30MB INT8) is downloaded once and cached in IndexedDB. Subsequent translations work without an internet connection.

Is there a text length limit?

The model accepts up to 512 tokens per translation (approximately 400 words). For longer texts, use batch mode: separate paragraphs with blank lines.

Our Test Data

We tested 200 parallel sentences across all supported language pairs — average translation time 0.4s on MacBook Air M2.

MetricValue
Average input length85 words
Average translation time0.4s / sentence
Model load (cold start)~8s (CDN) / <1s (cached)
BLEU score (EN↔ES)42.1
BLEU score (EN↔PT)40.8
BLEU score (EN↔ID)38.5

Last verified: September 2026. Results vary by hardware, model direction and input complexity.

When Not To Use This Tool