AI Autocomplete for VS Code

AI-powered inline code completions for JavaScript, TypeScript, and Dart (Bring your own API key).

Features

• Multiple AI Providers: Claude (Anthropic), Gemini (Google), or Ollama (Local Models)
• Inline Completions: Get AI-powered code suggestions as you type
• Language Support: JavaScript, TypeScript, JSX, TSX, and Dart
• Local AI Support: Run models offline with Ollama - no API costs!
• Smart Caching: Reduces API calls with LRU cache
• Throttling: Prevents API spam with configurable debounce
• Error Handling: Graceful degradation with informative messages
• Language-Aware: Context-aware completions for each language

Supported Languages

• JavaScript (.js, .jsx, .mjs)
• TypeScript (.ts, .tsx)
• Dart (.dart) - Great for Flutter development!

Setup

Choose your preferred AI provider:

Option 1: Local Models with Ollama (Recommended for Privacy & Cost)

Why Ollama?

• ✅ 100% Free - No API costs
• ✅ Privacy - Your code never leaves your machine
• ✅ Offline - Works without internet
• ✅ Fast - 2-5 second response times on modern hardware

Setup Steps:

1. Install Ollama

   • Download from ollama.ai
   • Windows: Run the installer
   • Mac: brew install ollama
   • Linux: curl -fsSL https://ollama.ai/install.sh | sh

2. Pull a Model

   # Lightweight model (recommended for autocomplete)
   ollama pull gemma3:1b
   
   # OR a more powerful model
   ollama pull deepseek-r1:8b
   

3. Verify Ollama is Running

   ollama list
   

   You should see your downloaded model(s)

4. Configure Extension

   • Open VS Code Command Palette (Ctrl+Shift+P)
   • Run AI Autocomplete: Select Model
   • Choose Ollama (Local)
   • Run AI Autocomplete: Set Ollama Model
   • Enter your model name (e.g., gemma3:1b)

Recommended Models:

• gemma3:1b - Fast, lightweight (1GB RAM)
• qwen2.5-coder:1.5b - Optimized for code (2GB RAM)
• deepseek-r1:8b - More accurate, slower (8GB RAM)

Option 2: Claude API (Anthropic)

1. Go to Anthropic Console
2. Sign up or log in
3. Navigate to "API Keys" section
4. Create a new API key
5. Run AI Autocomplete: Set API Key in VS Code
6. Run AI Autocomplete: Select Model and choose a Claude model

Option 3: Gemini API (Google)

1. Go to Google AI Studio
2. Create an API key
3. Run AI Autocomplete: Set API Key in VS Code
4. Run AI Autocomplete: Select Model and choose a Gemini model

Usage

Automatic Completions

Completions appear automatically as you type in supported languages. The extension:

• Extracts context around your cursor
• Sends it to API
• Displays suggestions inline
• Caches results to minimize API calls

Manual Trigger

Press Ctrl+Shift+Space (or Cmd+Shift+Space on Mac) to manually request a completion.

Commands

Available commands in the Command Palette:

• AI Autocomplete: Set API Key - Update your Claude/Gemini API key
• AI Autocomplete: Select Model - Choose AI provider (Claude, Gemini, or Ollama)
• AI Autocomplete: Set Ollama Base URL - Configure Ollama server URL (default: http://localhost:11434)
• AI Autocomplete: Set Ollama Model - Choose which local model to use
• AI Autocomplete: Toggle Autocomplete - Enable/disable the extension
• AI Autocomplete: Set Temperature - Adjust creativity (0=deterministic, 1=creative)
• AI Autocomplete: Set Debounce Delay - Change wait time before requesting (100-2000ms)
• AI Autocomplete: Clear Completion Cache - Clear cached completions
• AI Autocomplete: Trigger Completion Manually - Request a completion manually

Configuration

Open Settings and search for "AI Autocomplete" to customize:

Cost Estimation

Based on Claude Sonnet 4 pricing:

• Input: $3 / million tokens
• Output: $15 / million tokens

Typical usage: ~$0.10-0.50 per hour of active coding

Example:

• 50 completions/hour
• 500 input tokens average = 25,000 total input tokens
• 100 output tokens average = 5,000 total output tokens
• Cost: (25,000 × $3/M) + (5,000 × $15/M) ≈ $0.15

Performance Notes

• Request Throttling: Max 1 request per 300ms to prevent API spam
• Smart Caching: 100-entry LRU cache with 5-minute TTL
• Aggressive Filtering: Only completes in code (not comments/strings)
• Timeouts: 5-second timeout per request to avoid blocking
• Memory: Uses <50MB memory typical usage

Limitations

• Only works in supported languages (JS/TS/Dart)
• No completions in comments or string literals
• Requires network connection
• API key is stored in VS Code settings

Troubleshooting

"API key not configured"

• Run AIAutocomplete: Set API Key command
• Or add your key to VS Code settings

"Rate limited by API"

• Reduce debounceDelay setting to decrease request frequency
• Wait a moment before continuing typing

"Request timeout"

For Ollama:

• Ensure Ollama is running: ollama list
• Check if model is loaded: ollama run <model-name> "test"
• Verify base URL in settings (should be http://localhost:11434 or http://127.0.0.1:11434)

For Claude/Gemini:

• The API took too long
• Check your internet connection
• Try again

No completions appearing

• Ensure language is supported (JS/TS/Dart)
• Check you're not in a comment or string
• Verify API key is valid
• Check VS Code output channel: AI Autocomplete

Privacy & Security

• Your API key is stored securely in VS Code's encrypted settings
• Code context is sent to Anthropic API for processing
• Results are cached locally (not stored externally)
• No telemetry or usage tracking

License

MIT

Support

For issues or questions:

1. Check the troubleshooting section
2. View logs in VS Code output channel: AI Autocomplete
3. File an issue here

Made with ❤️ for developers who love AI-assisted coding