5月3日 的 LLM Hub - Local AI Assistant 應用程式分析

Advanced AI runs completely on your phone - no internet required, no data shared, complete privacy guaranteed. State-of-the-art Large Language Models in your pocket! ✨ EIGHT POWERFUL AI TOOLS 📝 SMART CHAT Multi-turn conversations with context awareness, RAG memory, optional web search, text-to-speech output with auto-readout, and support for text, images, and audio input. Create custom AI personas with personalized instructions. ✍️ WRITING AID Summarize documents, expand ideas, rewrite content, improve grammar, and generate code from natural language descriptions. 🎨 IMAGE GENERATOR Create stunning images from text prompts using Stable Diffusion 1.5. Generate multiple variations with swipe-through gallery. All processing happens on-device. 🌍 TRANSLATOR (50+ Languages) Translate text, images (OCR), and audio in real-time. Works completely offline with bidirectional support. 🎙️ TRANSCRIBER Convert audio to text with high accuracy. Supports WAV. All processing on-device. 🎵 VIBECODE Creator Create, customize, and share coded AI personas. Build personalized instruction sets that control AI behavior across all chat sessions. Design your perfect AI assistant with system prompts and descriptions. 💻 CODE ASSISTANT Generate, debug, and explain code with VibeCode-enhanced AI models. Optional web search for documentation and best practices. 🛡️ SCAM DETECTOR Analyze suspicious messages and images for phishing attempts. Get clear risk assessments and detailed explanations. 🚀 CUTTING-EDGE AI MODELS Supported Model Formats: GGUF, ONNX, LiteRT, MediaPipe (MediaPipe, ONNX Runtime, and Nexa SDK inference backends) Featured Models: • Gemma-3 1B (Google) - Fast and efficient • Gemma-3n E2B/E4B (Google) - Multimodal: text, vision, audio • Llama-3.2 1B/3B (Meta) - Powerful open-source • Phi-4 Mini (Microsoft) - Optimized for mobile • Ministral 3B/8B (Mistral AI) - High-performance multilingual with vision support • GPT-OSS Family - Fast, efficient models with ONNX optimization • LFM-2.5 Thinking (Thinking) - Reasoning and long-form generation • Granite Models (IBM) - Enterprise-ready language models • Stable Diffusion 1.5 - On-device image generation All models run 100% on-device with GPU/NPU acceleration. 🔐 PRIVACY & SECURITY • Zero data collection - everything stays on your device • No internet required for AI inference • No accounts, no tracking, no cloud uploads • Open-source and transparent ⚡ ADVANCED FEATURES • Text-to-Speech with auto-readout for hands-free listening • GPU/NPU acceleration for fast performance • Multimodal: text, images, and audio • RAG with global memory for enhanced responses • Create custom AI personas (VibeCode) with personalized instructions • Import custom models (.gguf, .onnx, .task, .litertlm, .mnn) • Direct downloads from HuggingFace for GGUF and ONNX models • Multi-runtime inference: MediaPipe, ONNX Runtime, and Nexa SDK backends • Beautiful Material Design UI • 17 language interfaces 📱 REQUIREMENTS Minimum: Android 8.0+, 2GB RAM, 1-5GB storage Recommended: 6GB+ RAM for best performance 💡 HOW IT WORKS 1. Download AI models in-app (one-time) — GGUF, ONNX, LiteRT, and MediaPipe formats supported 2. Create custom AI personas with VibeCode for personalized behavior 3. Choose your tool: Chat (with personas), Writing Aid, Image Generator, Translator, Transcriber, Code Assistant, or Scam Detector 4. Use AI completely offline with full privacy 5. Optional: Enable web search, upload images/audio, use text-to-speech, or import custom models 🌟 PERFECT FOR • Privacy-conscious users valuing data security • Professionals needing offline AI assistance (VibeCode Custom Personas) • Developers generating and debugging code • Artists and creators generating AI images • Users wanting hands-free voice responses • Students working on writing and research • Travelers requiring offline translation • Anyone protecting against scams 📖 OPEN SOURCE (MIT License) github.com/timmyy123/LLM-Hub