Google’s AI Edge Gallery app enables users to run large language models locally and privately on their Android or iOS devices without needing a developer account, supporting various AI interactions including chat, multimodal inputs, and customizable agent skills. The app emphasizes privacy by processing data on-device, offers features like voice-controlled device actions and preset prompt templates, and is designed for both casual users and developers seeking offline AI capabilities.
The video tutorial introduces Google’s AI Edge Gallery, an app available on both Android and iOS that allows users to run large language models (LLMs) locally and privately on their devices without needing a developer account or waiting list. The app enables downloading and running various AI models directly on your phone, making it a convenient tool for on-device AI interactions. The presenter highlights device compatibility, recommending phones with at least 8 GB of RAM for optimal performance, especially for running advanced models like Gemini 4, while older or lower RAM devices can handle smaller models.
The core feature of the app is the AI chat, where users can download different LLM variants and interact with them in a chat interface. The models vary in size and capability, with larger models offering better reasoning but requiring more powerful hardware. Users can customize settings such as temperature, top K, and top P to control the randomness and creativity of the AI responses, though the default settings are suitable for most users. The app runs entirely on-device, ensuring privacy as no data is sent to external servers, but it currently does not save full chat histories, only input text history.
Beyond basic chat, the app includes “agent skills,” which are preset prompts designed to guide the AI in performing specific tasks or reasoning styles. These skills can be imported or created by users, allowing for tailored AI behavior such as generating video scripts or fitness coaching. The AI also supports multimodal inputs, including images and audio. Users can upload or capture images for the AI to analyze and describe, and audio clips can be transcribed and responded to, although audio handling requires restarting conversations for each new clip.
Another experimental feature is “mobile action,” which allows the AI to control device functions like turning the flashlight on or off through voice commands. While still in early stages and not fully reliable, this feature hints at future possibilities for AI-driven device control. The app also includes a simple game called Tiny Garden and a “Prompts Lab,” which offers preset prompt templates for common tasks like summarizing text or rewriting tones, making it easier for users to get quick and useful AI-generated outputs without complex setup.
Overall, Google’s AI Edge Gallery provides a powerful and private way to run advanced AI models locally on mobile devices. It supports a variety of input types and customizable AI behaviors, making it suitable for both casual users and developers interested in experimenting with on-device AI. The app’s ease of use, combined with its privacy benefits and expanding feature set, positions it as a promising tool for mobile AI applications, especially for users who need offline access or want to keep their data secure.