A translator that reads the whole book before translating it

What it does

Wenyi translates entire books — EPUB, PDF, DOCX, FB2, TXT, Markdown, HTML — using an LLM provider of your choice (DeepSeek, OpenAI, Gemini, Ollama, vLLM, or anything OpenAI-compatible). It runs as a desktop app, a CLI, or a self-hosted web workspace, and exports back to most of the same formats, including bilingual editions. Crucially, it treats translation as a book-length project: progress is checkpointed, so you can close and reopen a project mid-book.

The interesting bit

The boring part is where the value sits: chapter digests, a book-level synopsis, and rolling context keep each translation batch aware of the rest of the work, so chapter 40 doesn’t forget what chapter 3 established. Names and recurring terms are extracted as it goes, conflicts are surfaced for you to resolve, and those decisions feed back into later batches.

Key highlights

• Whole-book context via chapter digests, synopsis, and rolling context windows
• Terminology extraction with conflict resolution that propagates to subsequent batches
• Paragraph-level editing with revision history, plus optional polishing and evidence-driven whole-book review
• Resumable checkpoints — completed batches and chapters are saved
• EPUB translation writes back into the original XHTML templates, preserving styles, images, and TOC; vertical layout converts to horizontal for Chinese readers
• Model routing lets you pick different models per operation and share request limits

Caveats

• Quality is bounded by whatever LLM you point it at; polishing and whole-book review can significantly inflate token usage (shadow fixing may trigger multiple full-book review passes)
• PDF is the roughest edge: MinerU is the default backend and needs its own API key; PDF output only works for saved BabelDOC projects, which is a separately licensed AGPL service
• Workspaces grow with book length — very long books can eat serious disk space

Verdict

Worth a look if you translate long fiction or nonfiction and are tired of chapter-by-chapter terminology drift. If you just need a quick page or paragraph translated, it’s a lot of machinery for the job.