What it does

Feed it an EPUB, TXT, PDF, or SRT file and it spits out a bilingual version—original text paired with AI-translated prose. It handles the boring orchestration: chunking the book, calling translation APIs, stitching results back into a readable format, and resuming if your connection drops mid-chapter.

The interesting bit

The project doesn’t lock you into OpenAI. It supports a sprawling roster of backends—Google Translate (free), DeepL, Claude, Gemini, Qwen, xAI, Ollama, Groq, even custom OpenAI-compatible endpoints via a JSON config file. The --use_context flag is a nice touch: it feeds the model surrounding paragraphs so pronouns and references don’t turn into gibberish.

Key highlights

• Supports 15+ translation backends including self-hosted Ollama and free options (Google, Tencent TranSmart, DeepL Free)
• Custom provider config via bbm_providers.json means no source edits to add DeepSeek or SiliconFlow
• Resume interrupted translations with --resume; auto-saves temp files on Ctrl+C
• Granular HTML tag control: translate only <p> tags or include/exclude specific elements
• PDF inputs get a TXT fallback automatically if EPUB generation fails

Caveats

• The README warns this is “exclusively designed for public domain works” with a disclaimer to review; the legal framing is present but enforcement is on you
• PDF support is partial: outputs bilingual TXT reliably, EPUB creation is best-effort
• --book_from mentions Kobo support but the README cuts off mid-sentence, so actual e-reader compatibility is unclear

Verdict

Worth a look if you regularly read foreign-language ebooks and want side-by-side translations without manual copy-paste. Skip it if you need polished, publisher-grade output—this is automation for personal use, not professional localization.