WriteTex for VSCode

WriteTex is a tablet/ipad app that converts handwritten math to LaTeX. Checkout our website for more information.

This extension allows users to use WriteTex in Visual Studio Code. It seamlessly integrates OpenAI-compatible vision models to help you convert handwritten equations, diagrams, and text into LaTeX, all while maintaining context from your current document, resulting in better OCR accuracy.

✨ Features

🖼️ AI-Powered OCR

Context-Aware Translation: OCR results consider your document context for more accurate conversions
OpenAI-Compatible API: Works with any OpenAI-compatible vision model (GPT-4o, GPT-4 Vision, etc.)
Automatic Insertion: OCR results are automatically inserted at your cursor position
Streaming Support: Real-time streaming of OCR results

Intuitive Settings Panel: Configure all settings through a sidebar UI
Real-Time Server Control: Start and stop the OCR server with one click
Status Indicators: Visual feedback for server status

🔌 Network Discovery

mDNS Service Advertisement: Automatically discoverable on your local network
Cross-Platform Support: Works on Windows, macOS, and Linux
Multiple Network Interfaces: Binds to all available network interfaces

🎯 Supported File Types

LaTeX (.tex, .latex)
Markdown (.md, .markdown)
R Markdown (.rmd)
Quarto (.qmd)
Jupyter Notebooks (.ipynb)

📦 Installation

From VSCode Marketplace

Open VSCode
Go to Extensions (Ctrl+Shift+X or Cmd+Shift+X)
Search for "WriteTex"
Click Install

🚀 Getting Started

1. Configure API Settings

Click the WriteTex icon in the left sidebar to open the settings panel:

API Endpoint: Your OpenAI-compatible API endpoint (default: https://api.openai.com/v1)
Model: Vision model to use (e.g., gpt-4o, gpt-4-vision-preview)
API Key: Your API key for authentication

You can also configure these in VSCode settings:

{
  "writetex.apiEndpoint": "https://api.openai.com/v1",
  "writetex.apiModel": "gpt-4o",
  "writetex.apiKey": "your-api-key-here"
}

2. Adjust Context Settings (Optional)

Context Line Radius: Number of lines before/after cursor to include (default: 20)
Context Char Limit: Maximum characters of context (default: 2000)

3. Start the Server

Click the Start Server button in the sidebar. The status indicator will turn green when the server is running.

4. Use OCR

The WriteTex server runs on localhost:50905 and is advertised via mDNS as WriteTex VSCode @ <hostname>.

Connect your OCR client (mobile app, web app, etc.) to this endpoint and send image data. The OCR results will be automatically inserted at your cursor position in VSCode.

⚙️ Configuration

All settings can be configured through the sidebar UI or VSCode settings:

Setting	Default	Description
`writetex.apiEndpoint`	`https://api.openai.com/v1`	OpenAI-compatible API endpoint
`writetex.apiModel`	`gpt-4o`	Vision model for OCR
`writetex.apiKey`	`""`	API key (⚠️ stored in plaintext)
`writetex.contextLineRadius`	`20`	Lines of context around cursor
`writetex.contextCharLimit`	`2000`	Maximum context characters

🔧 Architecture

┌─────────────────┐
│   OCR Client    │ (Mobile/Desktop App)
│  (Send Image)   │
└────────┬────────┘
         │ HTTP POST
         │ /v1/chat/completions
         ▼
┌─────────────────┐
│  WriteTex       │
│  VSCode Ext     │
│  Port: 50905    │
└────────┬────────┘
         │
         ├─► mDNS Advertisement
         │   (Service Discovery)
         │
         ├─► Context Extraction
         │   (From active document)
         │
         ├─► OpenAI API Call
         │   (Vision model OCR)
         │
         └─► Insert at Cursor
             (In active document)

Request Flow

Image Capture: OCR client captures an image
API Request: Client sends image to WriteTex Vscode extension server (/v1/chat/completions)
Context Injection: WriteTex Vscode extension extracts context from the active document
Vision API Call: WriteTex Vscode extension forwards request to OpenAI-compatible API with context
Streaming Response: OCR results stream back to client
Auto-Insert: Results are automatically inserted at cursor position

Authentication

The server uses a hardcoded bearer token (writetex) for authentication. Clients must include:

Authorization: Bearer writetex

🌐 API Endpoints

Health Check

GET /health

Returns server status and port information.

OCR Endpoint

POST /v1/chat/completions
POST /chat/completions

OpenAI-compatible chat completions endpoint. Send image data as base64 in the message content.

🛠️ Development

Prerequisites

Node.js 16+
VSCode 1.104.0+

Clone

git clone https://github.com/jacklitstar/writetex-vscode.git
cd writetex-vscode

For Mac and Linux devices, change line 85 of package.json to

    "compile": "tsc -p ./ && cp -R src/locales out/locales",

Build

npm install
npm run compile

Watch Mode

npm run watch

Lint

npm run lint

Test

npm test

📝 Commands

Access commands via the Command Palette (Ctrl+Shift+P or Cmd+Shift+P):

WriteTex: Start OCR Server - Start the OCR proxy server
WriteTex: Stop OCR Server - Stop the OCR proxy server
WriteTex: Show Actions - Quick access to server controls

🐛 Troubleshooting

Server Won't Start

Check if port 50905 is already in use
Ensure your firewall allows connections on this port
Check the Debug Console for error messages

OCR Not Working

Verify API key is correctly configured
Check API endpoint is accessible
Ensure the model supports vision/image inputs
Check Debug Console for API errors

mDNS Issues on Windows

Ensure Bonjour Service is installed (comes with iTunes or standalone)
Check Windows Firewall settings
Verify network discovery is enabled

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Adding Translations

To add support for a new language or fix translations:

Create a new JSON file in src/locales/ (e.g., ru.json)
Copy the structure from src/locales/en.json
Translate all strings
Add language mapping in src/i18n.ts
Test with your language setting

📚 Resources

Documentation: https://www.writetex.com/docs/vscode.html

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

AI models for vibe coding
The VSCode extension development community
All contributors and translators

Vibe coded and LGTM by jacklitstar

WriteTex