Version 2

Product Introduction

1. MyDeviceAI Version 2 is a privacy-focused mobile application that enables local AI-driven search, summarization, and conversational interactions entirely on-device without requiring cloud connectivity.
2. The core value lies in delivering commercial-grade AI capabilities—including web search integration, personalized responses, and multi-model support—while ensuring 100% data privacy and offline functionality on iOS devices.

Main Features

1. Local Web Search & Summarization: Integrates a privacy-preserving SearXNG-powered search engine to retrieve and process web results locally, enabling users to query and summarize information without data leaving their device.
2. Thinking Mode with Qwen 3: Leverages the 1.7B-parameter Qwen 3 model for complex problem-solving tasks, providing step-by-step reasoning for technical, creative, or analytical challenges while maintaining offline operation.
3. Custom Model Support: Allows advanced users to import and run GGUF-format AI models from HuggingFace, supporting diverse use cases like coding assistance or creative writing through open-source model compatibility.
4. Persistent Chat History: Stores conversation logs locally for 30+ days with encrypted storage, enabling users to review past interactions without cloud synchronization or third-party data access.
5. Dynamic Personalization (Beta): Implements on-device fine-tuning to adapt responses based on user preferences and historical interactions, using locally stored context files to maintain privacy during customization.

Problems Solved

1. Privacy Risks in AI Services: Eliminates data exposure by processing all queries, search results, and personalization entirely on the user's iPhone or iPad, addressing concerns about cloud-based AI services harvesting sensitive information.
2. Target Audience: Serves privacy-conscious consumers, researchers, and professionals requiring offline AI access, as well as developers seeking open-source alternatives to commercial chatbot platforms.
3. Use Cases: Enables secure medical/legal research via private web summarization, supports brainstorming sessions in airplane mode, and provides a sandbox for testing open-source LLMs without cloud dependencies.

Unique Advantages

1. Architectural Differentiation: Unlike cloud-dependent tools like Perplexity or ChatGPT, MyDeviceAI executes all NLP tasks via on-device CoreML/MLX frameworks, ensuring zero data transmission to external servers.
2. Technical Innovations: Combines quantized model optimization (2-4bit precision) with asynchronous background loading to achieve sub-second response times on A15+ chipsets while maintaining full app functionality during model initialization.
3. Market Position: Only iOS app offering free open-source (Apache 2.0) access to both base code (via GitHub) and modular AI pipeline architecture, allowing community-driven enhancements to search providers, UI components, and model integrations.

Frequently Asked Questions (FAQ)

1. Does the personalization feature actually modify model behavior? Yes, the beta personalization uses LoRA-like adaptation techniques applied locally through user-provided context files, though effectiveness varies based on training data quality and device compute capabilities.
2. Can I use models larger than 2B parameters? While the app supports any GGUF model, optimal performance is achieved with 1.5B-3B parameter models on iPhones; larger models may require M-series iPads or cause latency on older A12-A14 devices.
3. How does local search differ from traditional engines? The integrated SearXNG instance routes queries through user-configurable privacy-focused providers (e.g., DuckDuckGo) and processes results via on-device extraction algorithms, avoiding search engine tracking or IP logging.
4. Is internet required for initial setup? Only required for downloading base AI models (1.5-2GB) during first-time installation; subsequent web searches require connectivity, but chat functionality works fully offline.
5. What iOS versions are supported? Requires iOS/iPadOS 18 or later due to dependencies on CoreML 6 frameworks and on-device ML acceleration features exclusive to Apple's latest operating systems.