The translation engine from Firefox — running in your tab
The documents people most need translated — contracts, medical records, ID scans, official letters — are exactly the ones that should never be pasted into a translation website. So here the engine does not sit on a server: it is the same neural translation engine Mozilla builds into Firefox — running right here, in this browser tab. The model for your language pair downloads once (roughly 20–40 MB) and is cached — after that you can pull the network cable and it still translates. 50+ languages work in both directions; pairs that do not involve English are bridged through it automatically.
Drop a PDF and its headings, lists and tables keep their structure through the translation, ready to save as .txt, .md or an editable .docx — or just paste a paragraph and translate that. The honest framing: this is machine translation for understanding a document, not a certified translation for a court or embassy. For everyday business text the quality is close to the big online translators — Mozilla only ships a language pack when its quality is nearly on par with Google Translate. The exact visual layout of the PDF is not preserved (the text flow and structure are), and scanned pages have no text layer to translate — those need OCR first.