Silicon

Silicon is a native macOS/iOS app to chat with local large language models, optimized for Apple silicon and works on iPhone, iPad, Mac, and Vision Pro. Your chat history is saved locally with SwiftData, and you can customize the appearance of the app.

Features

• 14+ Local LLM Models - Run models entirely on-device
• Chat Management - Organize conversations with folders
• Customizable UI - Theme, font, and appearance settings
• Apple Silicon Optimized - Hardware-accelerated inference via MLX
• Privacy-First - All data stays on your device
• Siri Integration - Use Shortcuts for voice interactions
• Reasoning Models - Support for DeepSeek-R1 and Qwen thinking models

Supported Models

• Llama 3.2 1B & 3B
• DeepSeek-R1-Distill-Qwen-1.5B (Reasoning)
• DeepSeek-R1 14B (Reasoning)
• Qwen 2.5 3B & 7B
• Qwen 3 4B & 8B (Vision + PDF)
• Mistral 7B
• Gemma 2 2B
• Phi-4 14B
• SmolLM2 360M & 1.7B

Project Structure

silicon/
├── Features/              # App features (Siri Shortcuts, etc.)
├── Models/                # Data models & business logic
│   ├── Data.swift        # SwiftData models (Thread, Message, Folder)
│   ├── LLMEvaluator.swift # MLX inference engine
│   ├── ModelInfo.swift   # Model metadata
│   └── Models.swift      # ModelConfiguration extensions
├── Services/              # Business logic layer (future)
├── Utilities/             # Helper utilities
│   └── UI/               # UI-specific utilities (Theme)
└── Views/
    ├── Chat/             # Chat interface & conversation views
    ├── Onboarding/       # First-run experience
    ├── Settings/         # App settings
    │   └── Models/       # Model selection UI
    └── Shared/           # Reusable components
        └── Components/   # UI components (logos, etc.)

Architecture

• SwiftUI - Modern declarative UI framework
• SwiftData - Local persistence for chat history and folders
• MLX Swift - Hardware-accelerated ML inference
• AppIntents - Siri Shortcuts integration
• Observation - State management with @Observable

Requirements

• macOS 14.0+ / iOS 17.0+ / visionOS 1.0+
• Apple Silicon (M1/M2/M3/M4) or A17 Pro+ chip
• Metal 3 support
• Xcode 15.0+

Building

1. Clone the repository
2. Open silicon.xcodeproj in Xcode
3. Let Swift Package Manager resolve dependencies
4. Select your target device
5. Build and run (⌘R)

Credits

Silicon is made possible by MLX Swift from Apple. MLX is an array framework for machine learning research on Apple silicon.

Additional acknowledgments:

• Hugging Face community for model quantizations