LinguaHaru

About this project

The slow part of translating a document isn't the translating — it's rebuilding the formatting afterwards. Hand over a forty-page contract and you get back correct sentences wrapped in a broken document: tables misaligned, headers gone, the table of contents flattened into plain text. And if the API runs out of credit on page 30, the first 29 pages were for nothing.

LinguaHaru hands you back the same file, in another language. Word tables, styles, headers and footers, footnotes, the TOC, SmartArt and comments are written back into the original structure. Excel keeps merged cells, number formats and cross-sheet formulas working — translate a sheet name and every formula pointing at it is rewritten too. PowerPoint shrinks text that grew in translation to fit its box instead of letting it pile up. Scanned manga gets its bubbles detected, the original lettering erased, and the translation typeset vertically. In bilingual mode the translation is bolded and colored, so you can tell the two apart at a glance.

Nothing is lost if a run dies halfway: every segment is flushed to disk as it finishes, a dead key and an empty balance are told apart and explained in your own language, and the job lands in history so one click resumes exactly where it stopped.

Where it is today: the Windows portable build runs straight from an unzipped folder — no Python to install. 17 document and subtitle formats work out of the box (Word / PowerPoint / Excel / PDF / EPUB / Markdown / SRT / ASS…), while image OCR, manga, speech transcription and live captions install as plugins on demand, with GPU runtimes set up for you if you have an NVIDIA card. 13 fully localized UI languages; the desktop app and the browser console have the same features, and the browser one can be opened up to your LAN for colleagues. Point it at any OpenAI-compatible endpoint, or at a local LM Studio / Ollama to run fully offline. GPL, free, no in-app purchases.

One honest note

it doesn't claim to understand language better than the online translator you already use — it can run on the very same model. It is thorough about exactly one thing: the file you get back is ready to send, with no hour spent fixing formatting.