contextweave 0.2.2

ContextWeave: context-aware document translation with glossary management

English | 中文

ContextWeave

Context-aware document translation / 语境感知文档翻译

ContextWeave (语络译) is a fully automatic desktop translation app for long novels, books, PDFs, scanned documents, manga, and subtitles. It aims to preserve source formatting while keeping terminology and translation style consistent across the whole work.

Advanced documentation covers glossary memory, previous-context injection, format preservation, CLI automation, and advanced use cases.

Who ContextWeave Is For

• Novel, web novel, and light novel translation
• Long books and documents that need consistent naming and terminology
• Scanned books, PDFs, and manga that need OCR before translation
• People who want a desktop workflow instead of managing prompts by hand

Why ContextWeave

• Builds a glossary from your source material
• Carries context forward across chapters and pages, with useful summaries injected alongside glossary context
• Preserves original formatting for text-native files
• Handles text, EPUB, PDF, scanned pages, manga, and subtitles in one app

Install

Current desktop builds are unsigned, so the first launch may show an OS security warning.

macOS

• Download the latest .dmg
• Open it and drag ContextWeave.app into Applications
• Launch ContextWeave.app from Applications
• If macOS blocks it because the developer cannot be verified, go to System Settings -> Privacy & Security
• In the Security section, click Open Anyway for ContextWeave.app, then confirm Open

Windows

• Download the latest .zip
• Unzip it anywhere
• Run ContextWeave.exe
• If Windows SmartScreen warns that the app is unrecognized, click More info -> Run anyway

Setup

1. Open the projects screen and click Setup Wizard

This is the home screen for projects. Use Setup Wizard for the quickest first-time setup.

2. Choose providers and paste API keys

The wizard collects the providers it needs up front. For most users, DeepSeek + Gemini is the most practical starting point.

3. Review the workflow profile

The review step shows which connection and model ContextWeave will use for each workflow step.

Quality spends more for better reasoning and can get very, very expensive unless you are using only DeepSeek. Balanced is the safest default. Budget is the cheapest option when you want to minimize cost.

Translation

1. Create a project

Pick a project name, target language, and workflow profile.

2. Open the project work page

Import files in reading order so terminology and context stay consistent across the whole book, then click Translate and Export to start. Double-click a file if you want to inspect each step manually or retouch images.

3. Optional: import existing term translations

Open Terms, then use Import Terms if you already have a terminology list you want ContextWeave to reuse. A simple JSON object like {"original": "translated"} is enough.

Demo EPUBs

This sample EPUB was generated with Translate and Export directly from the French Project Gutenberg EPUB for The Count of Monte Cristo, Tome I, using DeepSeek.

Quality can be dramatically better with Gemini or GPT, but the cost is also significantly higher.

• The Count of Monte Cristo.epub - English output. Cost: under $2.5.

CLI

ContextWeave also includes a small CLI for config-driven one-shot translation and basic book management. From a source checkout, use uv run contextweave-cli; from an installed package, use contextweave-cli.

contextweave-cli config path
contextweave-cli config init
contextweave-cli config validate

contextweave-cli run ./book.epub --output ./translated/book.epub
contextweave-cli run ./chapter.txt --output ./translated/chapter.txt --json
contextweave-cli run ./episode.srt --output ./translated/episode.srt --no-polish

contextweave-cli books list
contextweave-cli books show BOOK_ID
contextweave-cli books delete BOOK_ID --yes

The CLI resolves config from --config, then CONTEXTWEAVE_CONFIG, then the nearest contextweave.yaml/.contextweave.yaml walking upward, then the platform default shown by contextweave-cli config path. The config mirrors the setup UI: connections define provider endpoints and workflow_profiles route each translation step. Prefer api_key_env so API keys stay in environment variables instead of config files or task snapshots. Use --no-polish when a one-shot run should skip the polish pass, which can be useful for timing-sensitive subtitle output.

A commented starting point is available at docs/examples/contextweave-cli.yaml.

What To Know Before Using ContextWeave

• The setup wizard path is mainly tested with DeepSeek + Gemini. Claude and GPT should also work well, but I do not recommend going below DeepSeek-class models.
• Image editing is expensive, and hallucinations are still common. For image reembedding, GPT Image 2 is recommended when available.
• OCR does not preserve original layout for PDFs and scanned books. It rebuilds from content instead. Manga is the exception.
• Import in reading order if you want the glossary and context to build correctly.
• Samples are still limited because testing across formats is expensive. Bug reports are very welcome.

Supported Formats

Type	Import	Export	OCR needed before translation?
Text	.txt, .md	txt	No
PDF	.pdf	epub, md	Yes
Scanned book	image files or folders	epub, md	Yes
Manga	.cbz, image folders	cbz	Yes
EPUB	.epub	epub, md, docx, html	No, but image OCR is supported
Subtitle	.srt, .vtt, .ass, .ssa	srt, vtt, ass, ssa	No