Wenyi

Overview

Wenyi is a local desktop app for translating books and long-form texts such as novels, social-science monographs, and narrative nonfiction. It prescans the whole book, tracks terminology as translation proceeds, and supports polishing and evidence-based review before export.

Key Features

• Desktop workspace for importing books, translating, proofreading, and exporting in one application; packaged releases include the Python engine without separate PostgreSQL, Redis, or Docker setup.
• Whole-book prescan creates per-chapter digests and a book-level synopsis that are injected into every translation batch.
• Real-time glossary extracts proper names, terms, and recurring expressions, detects conflicting translations, and feeds terms into later batches.
• Multi-stage workflow includes translation, optional polishing with a strong model, and evidence-driven whole-book AI review.
• Batch checkpoints and chapter status tracking let users reopen a project and resume saved progress.
• EPUB output writes translated text back into original XHTML templates and attempts to preserve styles, images, table of contents, and anchors; bilingual output is optional.

Use Cases

• Translating novels, social-science monographs, and narrative nonfiction across languages such as Chinese to English or English to Japanese.
• Producing monolingual or bilingual EPUB editions while attempting to preserve original layout elements.
• Translating SRT subtitles through a lightweight subtitle workflow without glossary, polishing, or whole-book review.
• Converting PDF input with MinerU before translation, with an optional BabelDOC bridge for layout-preserving PDFs.

Getting Started

• With Python 3.10+ and uv installed, run uv sync --locked from the repository root.
• Set DEEPSEEK_API_KEY to your own DeepSeek key and review config.yaml before translating.
• Run uv run wenyi translate book.epub, or use the Desktop quick start for a graphical app.

Deployment & Requirements

• CLI use requires Python 3.10+ and uv.
• Packaged desktop releases include the Python engine and do not require separate Python, PostgreSQL, Redis, or Docker setup.
• PDF input defaults to MinerU and requires its API key for initial conversion; the BabelDOC bridge is optional for layout-preserving PDFs.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• Multilingual translation remains experimental, with English prompt instructions and target-language metadata.
• Polishing and final review are the most expensive stages.
• PDF input depends on MinerU and its API key for initial conversion.
• Translation quality is bounded by the capabilities of the chosen LLM model.
• Very long books can create large workspaces and increase storage requirements.