I Built the World’s 1st Offline AI That Can See & Talk on A Phone 🤯

I'm Jed, and in this video, I’m thrilled to introduce MNN Chat—an open-source Android app and I contributed to its development. it delivers a 1:1 real-time multimodal voice and video chat experience matching ChatGPT. Powered by the MNN inference engine, it runs entirely locally on your device. No cloud, no internet required, and zero latency. 🚀 Key Features of MNN Chat: Massive Model Selection: Run over 40 different multimodal models completely offline, including DeepSeek, LLaMA, Gemma, and Qwen (from 0.6B to 35B). True Real-Time Conversation: Features acoustic echo cancellation so you can interrupt the AI mid-speech, just like talking to a real human. Live Video Chat: Enable your camera to let the AI see you and analyze the world around you in real-time. 100% Privacy: Because it runs entirely offline on-device, your voice and camera data never leave your phone. Forever Free & Open-Source: No subscriptions, no usage limits, no ads, and no catches. Just a pure, secure on-device AI companion. 🔗 Resources & Download Links: Want to try it out or dive into the code? Everything is open-source and available below: 📱 Download the App (APK): https://github.com/JedLee6/MNN/tree/m... 💻 GitHub Repository: https://github.com/JedLee6/MNN 🛠️ How I Built It (All Pull Requests): https://github.com/alibaba/MNN/pulls?... 📄 Technical Docs (Voice Interruption & Video Chat Principles): https://github.com/JedLee6/MNN/blob/m... ⏱️ Timestamps: 0:00 - The Problem with Cloud-based AI 0:45 - Meet MNN Chat (100% Offline AI) 1:30 - Real-Time Voice Chat Demo (Speed Test) 2:15 - Testing the Interruption Feature 2:45 - Live Video Chat Demo (Front & Back Camera) 3:30 - How the Code Works (Open Source) 4:00 - Future Plans (Multi-language & sherpa-onnx) If you found this project helpful, please drop a Like, Subscribe for more Android AI development, and let me know in the comments what features you want me to add next!