LLM Hub

Experience the next generation of AI with LLMHub, the most powerful private AI assistant for iPhone and iPad. Runs cutting-edge large language models (LLMs) entirely on-device upright, with zero data sharing and no need for the internet. Powered by the RunAnywhere SDK and deeply optimized for Apple chips.
Eight powerful on-device tools
• SMART CHAT: Supports multi-turn conversations, context awareness, RAG memory, and optional text-to-speech auto-reading. Supports text, image, and audio input. Powered by the RunAnywhere inference engine for extreme performance.
• WRITING AID: Instantly summarizes documents, rewrites content, and improves grammar. Processes long texts locally without sending any single word to the cloud.
• IMAGE GENERATOR: Creates stunning visuals using Stable Diffusion. Optimized for Apple Neural Engine (ANE) and Metal for fast local image synthesis directly from text or image prompts.
• TRANSLATOR (50+ language translation): Real-time two-way translation for text, audio, and image (OCR). Great for secure, private communication while traveling or working.
• TRANSCRIBER: High-precision speech-to-text conversion using native iOS recognition capabilities. All audio processing is done on the device to ensure absolute privacy.
• VIBECODE CREATOR: The ultimate AI personalization engine. Build, share, and use custom AI characters with deep behavior instructions throughout all chats.
• CODE ASSISTANT: Generates, interprets, and debugs code using VibeCode-enhanced models like Gemma-4. Has local inference capabilities to understand your specific programming style.
• SCAM DETECTOR: Analyzes suspicious messages and phishing images to assess risks. Gets clear local assessments to protect your digital identity security.
Cutting-edge technology stack
LLMHub is built specifically for iOS, leveraging the latest achievements in mobile AI research milling:
• RunAnywhere SDK: High-performance inference backend supporting GGUF, ONNX, and LiteRT formats with full Metal and Accelerate acceleration.
• Apple base model support: Seamlessly integrated with native Apple Intelligence base models for system-level speed and efficiency.
• Metal & Neural Engine optimization: Achieves desktop-level performance on your iPhone through Apple chips.
Selected on-device models
• Gemma-4 (Google): Represents the latest in multimodal performance, with E2B and E4B variants for advanced text and visual tasks.
• Apple Foundation Model: Apple system-integrated intelligence optimized for your specific hardware.
• Llama-3.2 (Meta): Powerful 1B and 3B inference models.
• Phi-4 Mini (Microsoft): Model optimized for mobile work.
• Ministral (Mistral AI): High-performance multilingual support.
Provides absolute privacy guarantee
• No internet needed: All AI inferences are completed 100% on your device.
• Zero data sharing: We never collect, store, or view your data.
• No need to register an account: Start using advanced AI immediately without any tracking or cloud uploads.
• Open source: Transparent and secure (MIT license).
Advanced iOS features
• Multimodal intelligence: Interact via text, voice, or photo.
• Download directly from HuggingFace: Import any compatible GGUF or ONNX model directly into your personal library.
• Native SwiftUI interface: Beautifully designed for iOS, focusing on smoothness and speed.
• Thinking model: Experience a new generation of inference that allows the model to think deeply before "speaking".