Spark - Local LLM Chat App for Android

Spark turns your Android phone into a powerful AI chat companion by running Large Language Models directly on your device. Chat privately with no cloud dependencies, and even use it as a local API server for your applications.

⚠️ Early Development Notice: Spark is in active development and things are moving fast! If you encounter any bugs or issues, please help us improve by:

• 🐛 Opening an issue
• 💬 Reporting it in our Discord community

Your feedback is incredibly valuable in making Spark better for everyone!

✨ Features

• 💬 Private Chat: Have conversations with AI models running 100% on your device
• 🤖 Multiple Models: Choose from small (0.5GB) to powerful (4GB+) models
• 🌐 Local API: Use as a drop-in replacement for OpenAI's API in your apps
• 🎨 Beautiful UI: Modern Material You design with dark mode
• 🔄 Easy Updates: One-click model downloads from HuggingFace
• 🔒 Full Privacy: No data leaves your device

📱 Requirements

• Android 8.1 or newer
• 64-bit processor (ARMv8)
• Storage: 2GB+ free space
• RAM:
  • Minimum: 3GB free RAM
  • Recommended: 6GB+ for larger models
  • High-end: 8GB+ for multimodal models

🤖 Available Models

Multimodal Models

• Gemma-3n-E2B-it (3.0GB)
  • Vision + Text capabilities
  • 4096 token context
  • Needs 8GB RAM
• Gemma-3n-E4B-it (4.2GB)
  • Enhanced multimodal model
  • 4096 token context
  • Needs 9GB RAM

Text Models

• Gemma3-1B-IT (0.5GB) ⭐
  • Fast & lightweight
  • 2048 token context
  • Only needs 4GB RAM
• Qwen2.5-1.5B (1.5GB)
  • Strong multilingual support
  • 1280 token context
  • Needs 6GB RAM
• DeepSeek R1 (1.5GB)
  • Best for reasoning tasks
  • 1280 token context
  • Needs 6GB RAM

🚀 Getting Started

1. Install Spark from Latest Release
2. Open app and go to Models tab
3. Download a model (start with Gemma3-1B-IT for best compatibility)
4. Create a new chat and start talking!

🌐 Using the API

The API server lets you use Spark models in your own apps as a local alternative to OpenAI's API.

1. In Spark app:

   • Load your chosen model
   • Go to Server tab
   • Start server and note your phone's IP address

2. In your application:

   • Replace localhost with your phone's IP
   • Use standard OpenAI API format:

# List available models
GET http://<phone-ip>:8080/v1/models

# Chat completion
POST http://<phone-ip>:8080/v1/chat/completions
{
  "model": "gemma3-1b-it",
  "messages": [{"role": "user", "content": "Hello!"}]
}

🔗 Links

• Discord Community
• Report Issues
• Latest Release

📄 License

MIT License - see LICENSE

🙏 Built With

• MediaPipe - LLM inference
• Ktor - API server
• Jetpack Compose - Modern UI