Discover and explore top open-source AI tools and projects—updated daily.
Summary
Box is an on-device AI suite for Android, forking Google AI Edge Gallery to provide a private, offline experience. It targets users and developers seeking to run advanced AI functionalities like conversational chat, voice interaction, vision analysis, and image generation directly on their mobile devices without network reliance. The primary benefit is enhanced privacy and immediate, local AI capabilities.
How It Works
Box integrates multiple native inference engines—llama.cpp for LLMs, whisper.cpp for speech-to-text, and stable-diffusion.cpp for image generation—alongside the LiteRT framework. It supports importing GGUF model files, allowing users to run various large language models and diffusion models locally. The architecture emphasizes a seamless, multimodal user experience, enabling features like real-time voice conversations and live camera-based AI interaction, all processed on-device.
Quick Start & Requirements
Highlighted Details
Maintenance & Community
Box is presented as a community effort, periodically merging upstream changes from Google AI Edge Gallery. While specific community channels like Discord or Slack are not detailed, the project is hosted on GitHub, facilitating issue tracking and contributions.
Licensing & Compatibility
Licensed under the Apache License, Version 2.0. This license is permissive and generally allows for commercial use and integration into closed-source applications.
Limitations & Caveats
NPU acceleration currently falls back to GPU/CPU for most models due to missing model-side AUX metadata. Custom ROM users require a specific build ("custom-rom-support") to work around missing system components like AICore and system TTS. NPU acceleration on custom ROMs for Snapdragon remains untested. Android 15+ is a strict requirement.
4 days ago
Inactive