Rosetta Mark

AI-powered Markdown translation VS Code extension with context awareness and incremental updates.

Features

• Multi-Provider Support: Works with OpenAI, Google Gemini, Anthropic Claude, and Ollama (local)
• BYOK (Bring Your Own Key): Your API keys stay secure in VS Code's SecretStorage
• Context-Aware Translation: Preserves code blocks, frontmatter, and formatting
• Incremental Translation: Hash-based caching to avoid re-translating unchanged documents
• Split View Preview: Automatically opens translation side-by-side with source
• Streaming Support: Real-time progress feedback during translation

Setup

1. Install the Extension

Install from VS Code Marketplace or download the .vsix file. Marketplace: seewhyme.rosetta-mark

2. Set Your API Key

Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P) and run:

Rosetta Mark: Set API Key

Enter your API key. It will be securely stored in VS Code's SecretStorage.

3. Configure Global Settings

You can configure Rosetta Mark globally (for all projects) in two ways:

Option A: Settings UI (Recommended)

1. Open VS Code Settings:

   • Mac: Cmd+,
   • Windows/Linux: Ctrl+,
   • Or: Cmd+Shift+P → "Preferences: Open Settings (UI)"

2. Search for "Rosetta Mark" in the search bar

3. Configure your preferences:

   • Provider: Choose your AI provider (OpenAI, Google, Anthropic, Ollama)
   • Model: Set the model name
   • Target Language: Set your default translation language
   • Preview Mode: Choose how to display translations
   • Max Concurrency: Control parallel translation speed
   • Glossary: Add custom terminology mappings

Option B: settings.json (Advanced)

1. Open settings.json:

   • Cmd+Shift+P → "Preferences: Open User Settings (JSON)"

2. Add your configuration:

{
  "rosettaMark.provider": "openai",
  "rosettaMark.model": "gpt-4o-mini",
  "rosettaMark.targetLanguage": "zh-CN",
  "rosettaMark.previewMode": "preview",
  "rosettaMark.maxConcurrency": 3,
  "rosettaMark.glossary": [
    {
      "source": "API",
      "target": "应用程序接口",
      "caseSensitive": false
    }
  ]
}

💡 Tip: For project-specific settings, create .vscode/settings.json in your project root. Project settings override global settings.

Configuration

Available Settings

Glossary Configuration

Add custom terminology to ensure consistent translations:

{
  "rosettaMark.glossary": [
    {
      "source": "frontend",
      "target": "前端",
      "caseSensitive": false
    },
    {
      "source": "API",
      "target": "应用程序接口",
      "caseSensitive": true
    }
  ]
}

Usage

Quick Start

1. Open a Markdown file
2. Click the globe icon 🌐 in the editor toolbar, or
3. Use keyboard shortcut: Cmd+Shift+T (Mac) / Ctrl+Shift+T (Windows/Linux)
4. Or open Command Palette and run Rosetta Mark: Translate Markdown

Translation Process

The extension will:

• ✅ Check if translation is needed (using file hash)
• 🔄 Translate the content while preserving formatting
• 💾 Save to VS Code's workspace storage
• 👀 Open the translation in a split view

Advanced Features

Batch Translation: Right-click on a folder in Explorer and select "Batch Translate" to translate multiple files at once.

Selection Translation: Select any text and use Cmd+Alt+T (Mac) / Ctrl+Alt+T (Windows/Linux) to translate only the selected portion.

Translation Cache

Translations are stored in VS Code's workspace storage, not in your project directory, so projects do not need a .gitignore entry for Rosetta Mark output. The cache still mirrors the source file structure and is isolated by workspace and translation configuration.

Rosetta Mark automatically cleans the current workspace cache on extension startup and after saving a translation. By default it removes cached translations that have not been used for 30 days and keeps the workspace cache under 500 MB. Run Rosetta Mark: Clean Translation Cache from the Command Palette to clean expired entries or clear all cached translations for the current workspace.

If an old project-local .rosetta-mark/ cache exists, Rosetta Mark copies it into VS Code storage the first time the workspace is used. It does not delete the old project-local folder automatically.

Supported Providers

OpenAI

{
  "rosettaMark.provider": "openai",
  "rosettaMark.model": "gpt-4o-mini"
}

Google Gemini

{
  "rosettaMark.provider": "google",
  "rosettaMark.model": "gemini-2.0-flash-exp"
}

Anthropic Claude

{
  "rosettaMark.provider": "anthropic",
  "rosettaMark.model": "claude-3-5-sonnet-20241022"
}

Ollama (Local)

{
  "rosettaMark.provider": "ollama",
  "rosettaMark.model": "llama3.2",
  "rosettaMark.baseUrl": "http://localhost:11434/v1"
}

Performance Tuning

The defaults (maxConcurrency=4, maxBatchTokens=4000) are a safe middle ground that works well across all providers and document sizes. If you translate large documents frequently and want to squeeze out more speed, tune these two settings based on your provider and model.

How they interact

• maxBatchTokens controls how many paragraphs are merged into a single request. Larger = fewer requests, but each request takes longer to generate.
• maxConcurrency controls how many requests run in parallel. More workers help only when there are enough batches to feed them.

The two settings interact: if maxBatchTokens is large, the whole document may fit in just 1–2 batches, and extra workers will sit idle. If maxBatchTokens is too small, prompt overhead is repeated many times and the batch-level latency floor adds up.

Per-provider recommendations

Real-world benchmark

A 54 KB / 190-paragraph README translated end-to-end (155 paragraphs needing translation, 35 reused as code/frontmatter):

Quick rules

• If translations feel slow on a fast cloud model: try lowering maxBatchTokens to 2500 first, then raising maxConcurrency to 8.
• If you hit 429 rate-limit errors: lower maxConcurrency to 2–3, keep maxBatchTokens at 4000 or higher to compensate.
• Incremental re-translation is mostly limited by the number of changed paragraphs, not these settings — small edits to a long document complete in seconds regardless.

Troubleshooting

Common Issues

"API Key not set" error

• Run Rosetta Mark: Set API Key from Command Palette
• Make sure you're using the correct API key for your selected provider

Translation fails with timeout

• Try increasing the timeout in settings (coming soon)
• For large files, use batch translation with lower concurrency

Formatting lost after translation

• Make sure your markdown is valid
• Code blocks should have language identifiers
• Report formatting issues on GitHub

Rate limit errors

• Reduce maxConcurrency setting to 1 or 2
• Wait a few minutes before retrying
• Consider using a different provider

Development

# Install dependencies
pnpm install

# Compile
pnpm run compile

# Watch mode
pnpm run watch

# Run tests
pnpm test

# Lint
pnpm run lint

# Format code
pnpm run format

Contributing

Contributions are welcome! Please read our Contributing Guide first.

License

MIT - see LICENSE file for details

Made with ❤️ by seewhyme

If this extension helps you, please consider ⭐ starring the repo!