Running an LLM on an Android Phone - Mostly nerdless

An open-source Android app runs AI models locally using the MediaPipe API and exposes them through a server on port 8005. The app supports Google's Gemma models, object detection with EfficientDet Lite 2, and provides REST APIs for text generation, image capture, and sensor data. Developers can access these AI capabilities from other apps or terminal environments like Termux, enabling use cases like creating a fortune clone that generates random adages using local LLMs with 2-3 second response times.