💎 Gemstone AI

🌟 Universal AI Chat System with On-Device Intelligence

This is a open-source multi-platform AI Chat System written with Kotlin Compose Multiplatform and Python.

---

✨ Key Features

• 🎯 Universal Platform Support - Run on Web, Android, iOS, Windows, Linux, and macOS
• 🧠 On-Device AI - Privacy-first local AI processing with Python Multiplatform
• 🎨 Modern UI - Beautiful, responsive interface built with Compose Multiplatform
• 💾 Offline Capability - Full functionality without internet connection
• 🔄 Real-time Chat - WebSocket-based real-time messaging
• 📱 Cross-Platform Sync - Seamless experience across all devices

🚀 Platform Support

🌐 Web	📱 Android	🍎 iOS
🪟 Windows	🐧 Linux	🍎 macOS

Note: Developed with Python Multiplatform

🛠️ Technology Stacks

🔧 Model Serving API

⚡ Note: Planning to support OpenAI-like API for AI model serving.

• Transformers/BitsAndBytes & PyTorch (INT4/INT8)
• Transformers/AutoGPTQ & PyTorch (SAFETENSORS)
• Llama-cpp-python & Numpy (GGUF)

📱 Mobile/Web Client APP

• Kotlin Multiplatform Mobile
• Compose Multiplatform Mobile

🔮 Note: Scheduled to adapt Python Multiplatform for On-Device AI model serving.

🤖 Supported Models

🖥️ Server Models

• Qwen 3 14b 4bitQ (BitsAndBytes)
• To be done...

📱 On-Device Models

• Llama 3.1 8B (llama.cpp)
• Qwen 3 14b 4bitQ (llama.cpp)
• To be done...

📖 API Reference

chat_history = [
{
"role": "user",
"content": "오늘 대전 날씨가 어때?"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_20250614125444",
"function": {
"name": "get_weather",
"arguments": {
"location": "Daejeon, South Korea",
"unit": "celsius"
}
}
}
]
},
{
"role": "tool",
"tool_call_id": "call_20250614125444",
"content": "<cached_result:call_20250614125444>"
},
{
"role": "assistant",
"content": "오늘 대전의 날씨는 다음과 같습니다:\n\n- 🌡️ 온도: 26.3°C (체감 온도: 30.6°C)\n- ☁️ 날씨: 주로 맑음\n- 💧 습도: 81%\n- 💨 바람: 13.8 m/s (북동)\n- 📊 기압: 994 hPa\n\n이 정보는 Open-Meteo에서 제공합니다."
}
]

🏗️ APP Architecture

Gemstone AI follows Clean Architecture principles with clear separation of concerns:

📁 app/
├── 🎯 domain/ # Business logic and entities
├── 🔄 adapter/ # Controllers and presenters
└── 🛠️ framework/ # UI, database, network, and AI infrastructure

📁 Detailed Project Structure (Click to expand)

app/
├── domain/
│ ├── entity/
│ │ ├── Message.kt
│ │ ├── ChatSession.kt
│ │ ├── User.kt
│ │ ├── AIModel.kt
│ │ └── value/
│ │ ├── MessageId.kt
│ │ ├── SessionId.kt
│ │ └── Timestamp.kt
│ ├── usecase/
│ │ ├── chat/
│ │ │ ├── SendMessageUseCase.kt
│ │ │ ├── GetChatHistoryUseCase.kt
│ │ │ ├── CreateSessionUseCase.kt
│ │ │ └── DeleteSessionUseCase.kt
│ │ ├── ai/
│ │ │ ├── SwitchAIModelUseCase.kt
│ │ │ ├── DownloadModelUseCase.kt
│ │ │ └── GetAvailableModelsUseCase.kt
│ │ └── user/
│ │ ├── SaveUserPreferencesUseCase.kt
│ │ └── GetUserPreferencesUseCase.kt
│ ├── repository/
│ │ ├── ChatRepository.kt
│ │ ├── AIModelRepository.kt
│ │ └── UserRepository.kt
│ └── service/
│ ├── AIService.kt
│ ├── ValidationService.kt
│ └── NotificationService.kt
├── adapter/
│ ├── controller/
│ │ ├── chat/
│ │ │ ├── ChatController.kt
│ │ │ ├── SessionController.kt
│ │ │ └── dto/
│ │ │ ├── SendMessageRequest.kt
│ │ │ └── CreateSessionRequest.kt
│ │ ├── ai/
│ │ │ ├── AIModelController.kt
│ │ │ └── dto/
│ │ │ └── SwitchModelRequest.kt
│ │ └── settings/
│ │ ├── SettingsController.kt
│ │ └── dto/
│ │ └── UpdatePreferencesRequest.kt
│ ├── presenter/
│ │ ├── chat/
│ │ │ ├── ChatPresenter.kt
│ │ │ ├── SessionPresenter.kt
│ │ │ └── model/
│ │ │ ├── ChatUiState.kt
│ │ │ ├── MessageUiModel.kt
│ │ │ └── SessionUiModel.kt
│ │ ├── ai/
│ │ │ ├── AIModelPresenter.kt
│ │ │ └── model/
│ │ │ ├── AIModelUiState.kt
│ │ │ └── ModelDownloadUiModel.kt
│ │ └── settings/
│ │ ├── SettingsPresenter.kt
│ │ └── model/
│ │ └── SettingsUiState.kt
└── framework/
├── ui/
│ ├── compose/
│ │ ├── screen/
│ │ │ ├── chat/
│ │ │ │ ├── MainScreen.kt
│ │ │ │ ├── SideScreen.kt
│ │ │ │ ├── ChatScreen.kt
│ │ │ │ └── component/
│ │ │ │ ├── MessageItem.kt
│ │ │ │ ├── MessageInput.kt
│ │ │ │ └── SubjectCard.kt
│ │ │ ├── model/
│ │ │ │ ├── ModelSelectionScreen.kt
│ │ │ │ └── component/
│ │ │ │ ├── ModelItem.kt
│ │ │ │ └── DownloadProgress.kt
│ │ │ └── settings/
│ │ │ ├── SettingsScreen.kt
│ │ │ └── component/
│ │ │ └── PreferenceItem.kt
│ │ ├── navigation/
│ │ │ ├── AppNavigation.kt
│ │ │ └── NavigationRoute.kt
│ │ └── theme/
│ │ ├── Color.kt
│ │ ├── Dimen.kt
│ │ ├── Shape.kt
│ │ ├── Theme.kt
│ │ └── Type.kt
│ └── viewmodel/
│ ├── ChatViewModel.kt
│ ├── AIModelViewModel.kt
│ └── SettingsViewModel.kt
├── database/
│ └── ChatDatabase.kt
├── network/
│ ├── http/
│ │ ├── HttpClientFactory.kt
│ │ └── api/
│ │ ├── ChatApiService.kt
│ │ └── AIModelApiService.kt
│ └── websocket/
│ └── ChatWebSocketClient.kt
├── ai/
│ ├── local/
│ │ └── LocalModelManager.kt
│ ├── remote/
│ │ ├── OpenAIClient.kt
│ │ ├── AnthropicClient.kt
│ │ └── HuggingFaceClient.kt
│ └── common/
│ └── ModelManager.kt
└── storage/
└── FileManager.kt

🚀 Installation

📋 Prerequisites

• Git
• Python 3.12+
• Java 21+ (for Kotlin Multiplatform)
• Xcode 14+ - For iOS development (macOS only)

🔧 Setup

1️⃣ Clone this repository

git clone https://github.com/thisisthepy/Gemstone.git
cd Gemstone

2️⃣ Install Python dependencies

uv sync

3️⃣ Install Llama-cpp-python with CUDA support (Optional)

For GPU acceleration with NVIDIA cards

CMAKE_ARGS="-DGGML_CUDA=on -DLLAVA_BUILD=off -DCMAKE_CUDA_ARCHITECTURES=native" \
FORCE_CMAKE=1 uv pip install llama-cpp-python --no-cache-dir --force-reinstall --upgrade

4️⃣ Run the server

python -m api run server

🎯 Usage

🖥️ Desktop Application

./gradlew :app:run

📱 Android

./gradlew :app:installDebug

🌐 Web

./gradlew :app:wasmJsBrowserDevelopmentRun

🍎 iOS

./gradlew :app:iosSimulatorArm64Test

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

🐛 Bug Reports

• Use the issue tracker
• Provide detailed reproduction steps
• Include platform and version information

💡 Feature Requests

• Check existing feature requests
• Propose new ideas with clear use cases
• Consider implementation complexity

📄 License

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

🙏 Acknowledgments

• JetBrains for Kotlin Multiplatform and Compose
• Hugging Face for the Transformers library
• Llama.cpp for efficient local model serving
• Llama-cpp-python for Python bindings
• Meta for Llama models
• Alibaba for Qwen models
• Google for Gemma models

---

🎉 Ready to chat with AI across all your devices!

🐛 Report Bug | 💬 Discuss on Discord | ⭐ Star on GitHub