LLM Code Autocomplete for VSCode

AI-powered code completion using any OpenAI-compatible API endpoint. Get intelligent code suggestions with support for OpenAI, Groq, Anthropic, and other providers that support the OpenAI API format.

Features

• Smart Context Detection: Automatically detects where to complete code

  • Completion markers (// TODO:, // COMPLETE:, etc.)
  • Empty function bodies
  • Empty code blocks
  • Current cursor position

• Multi-file Context: Analyzes all open tabs to understand your project structure

• Fast Streaming: See code being generated in real-time

• Multiple Providers: Works with any OpenAI-compatible API endpoint

• Flexible Configuration: Customize API endpoints, models, and parameters

Quick Start

Prerequisites

1. API Key: Get one from your preferred LLM provider:
   • OpenAI: platform.openai.com
   • Groq: console.groq.com
   • Or any other OpenAI-compatible provider
2. Node.js: Version 16.x or higher
3. VSCode: Version 1.85.0 or higher

Installation

1. Clone this repository:

   git clone <repository-url>
   cd llm-code-autocomplete
   

2. Install dependencies:

   npm install
   

3. Compile the extension:

   npm run compile
   

4. Open in VSCode:

   code .
   

5. Press F5 to launch a new Extension Development Host window

6. Configure your settings in VSCode:

   • Open Settings (Ctrl/Cmd + ,)
   • Search for "LLM Code Autocomplete"
   • Enter your API key and endpoint

Usage

Method 1: Using Markers

Add a marker where you want AI to complete code:

function calculateTotal(items) {
    // TODO: implement calculation logic
}

Then press Ctrl+Shift+G (Cmd+Shift+G on Mac)

Method 2: Empty Functions

def process_data(data):
    # Place cursor here and press Ctrl+Shift+G

Method 3: Empty Blocks

if (user.isAuthenticated) {
    // Cursor here + Ctrl+Shift+G
}

Method 4: Anywhere

Just place your cursor and press the hotkey!

Configuration

Access settings through VSCode Settings (search for "LLM Code Autocomplete"):

Example Endpoint Configurations

OpenAI:

• Endpoint: https://api.openai.com/v1
• Models: gpt-4, gpt-3.5-turbo, etc.

Groq:

• Endpoint: https://api.groq.com/openai/v1
• Models: mixtral-8x7b-32768, llama3-70b-8192, etc.

Local/Custom (e.g., vLLM, Ollama, etc.):

• Endpoint: http://localhost:8000/v1 (default)
• Models: Depends on your setup (e.g., Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8)
• API Key: Not required for vLLM (use "dummy" if needed)

Tips for Best Results

1. Use descriptive markers:

   // TODO: implement binary search with error handling
   

2. Keep related files open: The AI uses all open tabs for context

3. Define function signatures first:

   def merge_sort(arr: List[int]) -> List[int]:
       # AI will implement based on signature
   

4. Adjust temperature:

   • Low (0.1-0.3): Predictable, conventional code
   • High (0.5-0.7): More creative solutions

Development

Building from Source

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Watch mode (auto-compile)
npm run watch

# Run linter
npm run lint

Packaging

To create a .vsix file for distribution:

npm install -g vsce
vsce package

Publishing

vsce publish

Troubleshooting

"No completion generated"

• Try a different model suitable for your provider
• Increase maxTokens in settings
• Provide more context with comments

"API key not set"

• Check VSCode settings (LLM Code Autocomplete section)
• Or set environment variable: export LLM_API_KEY=your_key

API endpoint errors

• Verify your endpoint URL is correct
• Ensure it includes /v1 suffix for OpenAI compatibility
• Check that your API key is valid for the endpoint

Slow responses

• Try a smaller/faster model
• Reduce contextFileLimit in settings
• Disable streaming for single-shot completion

Incorrect completions

• Add comments explaining what you need
• Open related files for better context
• Use more specific markers

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

MIT License - see LICENSE file for details

Acknowledgments

• Built with the OpenAI SDK
• Inspired by GitHub Copilot and similar tools
• Thanks to the VSCode extension API documentation