Box is a fork of Google AI Edge Gallery that provides a private on-device AI environment for Android devices. It integrates llama.cpp for LLM inference with GGUF model support, whisper.cpp and SenseVoice for speech-to-text, stable-diffusion.cpp for on-device image generation, and Supertonic for text-to-speech. The app includes RAG capabilities, MCP server support, vision AI with multimodal LLMs, and document analysis. Hardware acceleration via LiteRT enables CPU, NPU, and GPU execution.