Privacy-First AI Assistant for iOS with Apple Intelligence Integration
OnDeviceAI is a revolutionary iOS application that seamlessly integrates Apple Intelligence with advanced MLX models - bringing powerful AI capabilities directly to your iPhone, completely offline and completely private.
- 100% On-Device Processing: No data ever leaves your iPhone
- Zero Analytics: No tracking, no telemetry, no user profiling
- No Account Required: Start chatting immediately after installation
- Offline Capable: Full functionality without internet connection
- Apple Intelligence First: Native integration with iOS 18.1+ AI capabilities
- On-Demand MLX Models: Download and run advanced models locally
- Smart Model Management: Automatic recommendations based on device capabilities
- Hybrid Processing: Seamless switching between Apple Intelligence and custom models
- Efficient Memory Management: Optimized performance with thermal protection
- Native SwiftUI Interface: Fluid, responsive, and beautiful
- Real-time Streaming: See responses as they're generated
- Smart Performance Monitoring: Automatic thermal and memory management
- Conversation History: Local storage with full user control
| Main Chat Interface | Real-time Streaming | Model Management |
| App Settings | Performance Monitoring | Privacy Controls |
| Resources & Legal | Privacy Documentation | Device Optimization |
| Advanced AI Features |
Key Interface Features:
- ✨ Clean, modern SwiftUI design following iOS guidelines
- ⚡ Real-time AI response streaming with performance stats
- 🧠 Smart model management with device optimization
- 🛡️ Privacy-first controls and transparent documentation
- 📊 Intelligent performance monitoring and thermal management
OnDeviceAI features a revolutionary hybrid architecture that seamlessly combines Apple Intelligence with advanced MLX models:
AI Processing Flow:
User Input → Intelligent Router → {
│
├─ Apple Intelligence (Default)
│ └─ iOS 18.1+ Native AI
│ • Instant availability
│ • Zero setup required
│ • Optimal performance
│
└─ MLX Models (On-Demand)
└─ Custom Downloaded Models
• Specialized capabilities
• Advanced reasoning
• Domain-specific tasks
}
OnDeviceAI/
├── 🧠 Core/ # Hybrid AI Engine
│ ├── AppleFoundationLLM.swift # Apple Intelligence integration
│ ├── LocalLLM.swift # Universal AI protocol
│ ├── MLXLLM.swift # MLX model implementation
│ ├── LLMSelection.swift # Intelligent model routing
│ ├── ModelManager.swift # Lifecycle & optimization
│ └── ModelDownloader.swift # On-demand model acquisition
├── 🎨 Views/ # Native SwiftUI Interface
│ ├── ChatView.swift # Real-time conversation
│ ├── ChatVM.swift # State management & logic
│ ├── QuickModelSelector.swift # Instant model switching
│ └── ModelPickerView.swift # Advanced model management
└── 📦 Resources/ # Privacy docs & legal
| Component | Technology | Purpose |
|---|---|---|
| Primary AI | Apple Intelligence | Native iOS AI (iOS 18.1+) |
| Enhanced AI | MLX Swift | Advanced on-device models |
| UI Framework | SwiftUI | Native iOS interface |
| Model Router | Swift Protocols | Intelligent AI selection |
| Concurrency | Swift Async/Await | Non-blocking AI processing |
| Storage | iOS FileManager | Local model & conversation storage |
| Performance | iOS System Frameworks | Thermal, memory & performance monitoring |
| Downloads | URLSession | Secure model acquisition |
- Xcode 16.0+
- iOS 26.0+ deployment target
- iPhone with A16+ chip (for Apple Intelligence)
- 8GB+ RAM for optimal MLX model performance
- iOS 18.1+ for full Apple Intelligence integration
- iPhone 15 Pro or later for best performance
- 12GB+ RAM for large MLX models
- 50GB+ free storage for extensive model library
-
Clone the repository
git clone https://github.com/simonazoulay/ondeviceai.git cd ondeviceai -
Install dependencies
# MLX Swift will be automatically resolved via Swift Package Manager -
Open in Xcode
open OnDeviceAI.xcodeproj
-
Build and run
- Select your target device
- Press ⌘+R to build and run
- Instant Start: Begin chatting immediately with Apple Intelligence (iOS 18.1+)
- No Setup Required: Zero configuration, no downloads, no waiting
- Optional Enhanced Models: Download additional MLX models for specialized tasks
- Smart Optimization: Automatic performance tuning based on your device capabilities
| Device Capability | Primary AI | Enhanced Models | Performance |
|---|---|---|---|
| Apple Intelligence | Native iOS AI | All MLX Models Available | Instant + Excellent |
| S-Tier (16GB+) | Apple Intelligence | SOLAR 10.7B, Gemma 2 9B | Perfect |
| A-Tier (12-16GB) | Apple Intelligence | Llama 3.1 8B, Mistral Nemo | Excellent |
| B-Tier (8-12GB) | Apple Intelligence | Gemma 2 2B, Phi-3 Mini | Very Good |
| Entry (6-8GB) | Apple Intelligence | Qwen 2.5 0.5B, TinyLlama | Good |
Apple Intelligence provides instant, high-quality responses on all supported devices. Additional MLX models enhance capabilities for specialized tasks.
// Automatic AI routing based on:
- Device capabilities (Apple Intelligence availability)
- Task complexity (simple queries → Apple AI, advanced → MLX)
- Device thermal state and available memory
- User preferences and model availability
- Performance optimization for optimal experience🚀 Instant Availability
- Zero setup required - start chatting immediately with Apple Intelligence
- No model downloads needed for basic functionality
- Seamless experience from first launch
⚡ Enhanced Performance
- 80% faster app launch (lightweight architecture)
- Reduced storage footprint (no bundled models)
- Intelligent resource management
🧠 Smart AI Integration
- Apple Intelligence for instant, high-quality responses
- MLX models for specialized tasks and advanced reasoning
- Automatic model selection based on query complexity
🔒 Privacy & Security
- Complete Apple ecosystem integration
- Enhanced privacy with native iOS AI processing
- Optional advanced models downloaded securely on-demand
Unlike cloud-based AI assistants, OnDeviceAI ensures your conversations never leave your device. No servers, no data collection, no privacy concerns.
Built specifically for iOS with cutting-edge optimizations:
- Apple Intelligence Integration: Native iOS AI with optimal performance
- Metal Performance Shaders: GPU acceleration for MLX models
- Neural Engine Utilization: Maximum hardware efficiency
- Intelligent Model Switching: Automatic selection based on task complexity
- Advanced Thermal Management: Prevents overheating with smart throttling
- Lightweight Architecture: Faster app launch and reduced storage footprint
Pioneering the future of on-device AI with hybrid architecture:
- Apple Intelligence Integration: First-class support for iOS native AI
- Advanced MLX Framework: State-of-the-art models running locally
- Intelligent Model Router: Automatic selection between Apple AI and custom models
- Real-time Streaming: Instant responses with performance monitoring
- Smart Resource Management: Efficient memory and thermal optimization
- Zero-Install Experience: Immediate functionality without setup delays
This repository contains a demonstration version of OnDeviceAI with certain implementation details abstracted for public sharing. The core architecture and interfaces are fully functional, showcasing the app's capabilities while protecting proprietary implementations.
// Core AI Protocol
protocol LocalLLM {
func load(modelURL: URL) throws
func generate(prompt: String, onToken: @escaping (String) -> Void) throws
func unload()
}
// Performance Monitoring
class PerformanceMonitor: ObservableObject {
var shouldThrottleGeneration: Bool { /* Implementation */ }
var statusText: String { /* Implementation */ }
}This is a demonstration repository. For feature requests or technical discussions, please reach out:
📧 Contact: @SimonAzoulayFr on X/Twitter
OnDeviceAI is built with privacy as its fundamental principle:
- ✅ No data collection of any kind
- ✅ No analytics or tracking
- ✅ No cloud processing
- ✅ Full user control over local data
- ✅ GDPR & CCPA compliant by design
OnDeviceAI integrates cutting-edge AI technologies:
- Apple Intelligence - Native iOS AI framework
- MLX Swift - Apple's machine learning framework
- Qwen Models - Advanced language models (optional)
- Llama Models - Meta's open-source models (optional)
- Various community contributions to the MLX and AI ecosystem
Full license attributions available in LICENSES.md.
- 🆘 Support & Help - How OnDeviceAI works
- 🚀 Product Info - Features and capabilities
- 🛡️ Privacy Policy - Our privacy commitment
- ⚙️ User Choices - Privacy controls
- Apple Intelligence Integration (iOS 18.1+) ✅
- Hybrid Model Architecture with intelligent switching ✅
- Zero-Setup Experience with instant availability ✅
- Vision Model Support for image understanding
- Advanced Code Models for developer assistance
- Multi-language Interface support
- Shortcuts Integration for workflow automation
- Background Processing for continuous AI assistance
Developer: Simon Azoulay
X/Twitter: @SimonAzoulayFr
For technical questions, feature requests, or general inquiries, DM me on X/Twitter for the fastest response.
OnDeviceAI - AI that respects your privacy
Made with ❤️ for iOS • Built in France 🇫🇷
This project is proprietary software. The code in this repository is shared for demonstration and educational purposes only.
- ✅ Viewing and learning from the code is encouraged
- ✅ Academic research and analysis is permitted
- ❌ Commercial use is prohibited without permission
- ❌ Redistribution of the code is not allowed
- ❌ Creating derivative works requires explicit authorization
For licensing inquiries, contact @SimonAzoulayFr.
© 2025 Simon Azoulay. All rights reserved.