huanghaitck/bookflow-scholar

Bookflow Scholar is a desktop application that translates scholarly PDFs, books, and monographs into other languages while preserving the structure of the original document. It is designed for researchers, translators, archivists, and publishers who need translations that remain faithful to the layout and citation structure of the source material. The project addresses a specific problem with PDF translation tools. Most tools extract text and translate it page by page, which breaks sentences at page boundaries, mixes headers and footers into the main text, and detaches figures from their captions. Bookflow Scholar takes a different approach. It first recovers complete logical units that span across pages, translates them with full context, then reinserts original page markers at the correct physical boundaries so readers can still cite and verify against the source edition. Users bring their own API keys for text and vision models, meaning documents go directly to the providers the user configures rather than to a hosted translation service. The app supports six languages: Simplified Chinese, English, French, German, Japanese, and Spanish, covering all 30 directed language pairs between them. Keys are stored through Windows Credential Manager. Users can choose text-only processing for simpler materials or enable multimodal analysis for complex layouts, maps, and difficult pages. The tool distinguishes body text, headings, headers, footers, footnotes, endnotes, plates, maps, tables, and captions, translating each category separately and returning them to their proper positions. It produces three editions from each source: the original, a target-language version, and a bilingual version. Long translation jobs can be paused, resumed, restarted, or cancelled, with retry support for failed steps. LibreOffice is an optional but recommended dependency for rendering office documents. To get started, users install a Windows setup package or portable ZIP, create a project, configure their API providers, import a PDF, select source and target languages, and start the process. The project is open source under the Apache License 2.0 and welcomes contributors and maintainers.