Carry stories across languages.
A desktop app for translating books and long-form writing, with the whole work in view.
Whole-book understanding · Consistent terminology · Evidence-based review
Quick start · Language support · Documentation
English | 简体中文
Thank you to PackyCode for sponsoring Wenyi.
|
PackyCode provides access to leading AI models through one API endpoint and one API key, with automatic failover and dedicated routes for Codex and Claude Code. |
| Typical approach | Wenyi |
|---|---|
| Segments translated in isolation, unaware of surrounding content | Whole-book prescan with chapter digests and rolling context |
| Glossary managed manually or as an afterthought | Real-time term extraction with conflict detection, fed back into subsequent batches |
| Single-pass translation, fragile to interruptions | Batch checkpoints and chapter status tracking: reopen the project and resume saved progress |
| Raw model output, no systematic quality process | Translate → polish → evidence-driven whole-book review |
Wenyi is designed for long-form texts — novels, social-science monographs, narrative nonfiction, and more.
- Local Desktop app — import books, translate, proofread, and save exports in one application. Projects stay in an independent local workspace; packaged releases include the Python engine, with no separate Python, PostgreSQL, Redis, or Docker setup. See the Desktop guide.
- Visual proofreading — English and Chinese interfaces, live translation progress, paragraph editing with revision history, and whole-book review with evidence and publication results. See the interface preview.
- Native file and credential support — drag files into a new project, choose export destinations with the system save dialog, and save API keys in the OS credential store when available.
- Whole-book understanding — prescans the source before translation, creating per-chapter digests and a book-level synopsis injected into every batch
- Real-time glossary — extracts proper names, terms, and recurring expressions as translation progresses; detects conflicting translations and surfaces them for resolution
- Multi-stage quality — optional polishing (strong model) and an evidence-driven whole-book AI review
- Resumability — completed batches and chapter progress are saved; reopen the application and continue from saved checkpoints
- Multiple LLM providers — DeepSeek, OpenAI, OpenRouter, OrcaRouter, Google Gemini, Ollama, vLLM, and generic OpenAI-compatible endpoints; keep three convenient tiers or select models per operation, mix connections, and share request limits. See model routing.
- Native EPUB preservation — writes translated text back into the original XHTML templates and attempts to preserve styles, images, TOC, and anchors
- Bilingual output — optional source-and-translation edition with visually subdued source text, including dark mode support.
Desktop and Web share these workspace pages. The screenshot below shows the translation overview in the Web Chinese interface; English is available in Settings.
Translation overview: usage by step, cache hit rates, and run durations.
With Python 3.10+ and uv installed, run from the repository root:
uv sync --locked
export DEEPSEEK_API_KEY="YOUR_DEEPSEEK_API_KEY"
uv run wenyi translate book.epubThis shell example uses the default DeepSeek provider. Supply your own key and review config.yaml before translating. For Windows commands, other providers, and resume/export options, see the CLI guide.
Prefer a graphical app? See the Desktop quick start for downloads and setup.
| Input | Output |
|---|---|
| EPUB, FB2, TXT, Markdown, HTML, PDF, DOCX | EPUB (monolingual / bilingual), TXT, HTML, Markdown, DOCX |
| SRT (movie / series subtitles) | SRT (monolingual / bilingual) |
- PDF input defaults to MinerU and requires its API key for the initial conversion; the resulting HTML is cached and reused. The BabelDOC bridge is optional for layout-preserving PDFs. See PDF configuration.
- EPUB output attempts to preserve the original book's styles, images, table of contents, and anchors. Vertical layout is converted to horizontal for Chinese reading.
- SRT projects use a lightweight subtitle workflow without a glossary, polishing, or whole-book review.
- DOCX uses the full book pipeline. Headings, simple tables, lists, and common run/paragraph styles are preserved where possible; translated Chinese uses Song (宋体).
Choose languages when creating a project, including directions such as Chinese → English or English → Japanese. Built-in profiles cover Chinese, English, Japanese, Korean, French, German, Spanish, Italian, Portuguese, Russian, Vietnamese, and selected variants. Multilingual translation remains experimental; real-model long-form quality needs further evaluation.
Wenyi combines whole-book understanding, batch translation, optional polishing, review, and export. See the translation pipeline for the flowchart and stage details.
- Desktop guide — installation/builds, API keys, native import and save, local data, and troubleshooting
- CLI guide — local installation, commands, translation, resume, and export
- Web deployment — optional self-hosted browser workspace with its own project data
- Configuration — providers, languages, pipeline switches, segmentation, paths
- Translation pipeline — whole-book analysis, terminology, context, polishing, review
- Architecture — module responsibilities, shared services, and platform boundaries
- Contributing — development, testing, and contribution guidelines
Public-domain translation examples are shared through wenyi-bookcase. Do not publish copyrighted text, private books, or workspace data containing sensitive information without permission.
- Multilingual translation is experimental. Prompt instructions use English, while generated descriptive metadata follows the translation target.
- Polishing and final review are the most expensive stages. Shadow fixing may trigger multiple full-book review passes and additional Fixer calls.
- PDF input defaults to MinerU and requires an API key for the initial conversion. The BabelDOC bridge is optional for layout-preserving PDFs.
- SRT translation does not include a glossary, polishing, or whole-book review.
- Translation quality is bounded by the capabilities of the chosen LLM model.
- Very long books may produce large workspaces; storage requirements grow with book length.
- Discord server
- QQ group: 1055065098
- GitHub Issues — bug reports and feature requests
- GitHub Discussions — ideas and questions
Wenyi is also hosted on AtomGit: https://atomgit.com/BigDawnGhost/wenyi