google-ai-edge/gallery
A chatbot app that runs entirely on your phone, no internet needed for chat, image questions, or voice transcription
Google AI Edge Gallery is an app that lets people run large language models (LLMs) directly on their smartphone or macOS device instead of a server. The latest version adds support for the Gemma 4 model family along with features like chat, image-based Q&A, live voice transcription, and prompt experimentation. Everything runs locally on the device, so no data is sent to an external server.
What it does
- Distributed via Google Play, App Store, and a macOS download, with an apk available directly from GitHub releases for those without Google Play access
- Adds official support for Gemma 4, and introduces a Thinking Mode that shows the model's step-by-step reasoning while answering, currently starting with the Gemma 4 family
- Includes Ask Image for object recognition and visual questions using the camera or photo gallery, Audio Scribe for real-time voice-to-text transcription, and Prompt Lab for testing prompts with adjustable parameters like temperature and top-k
- Ships experimental features such as Mobile Actions, which uses a finetune of FunctionGemma 270m to control the device offline, and Tiny Garden, a mini-game where you plant and harvest using natural language
- Lets users browse and download models from Hugging Face or load their own custom models, manage a model library, and run benchmark tests to see actual performance on their own hardware
Why it matters
It gives developers and everyday users a hands-on way to test how practical open-source LLMs really are on mobile hardware. Because inference happens fully on-device without cloud servers, it's a useful way to explore private, offline generative AI before deciding whether it fits a real product.
Terms in this repo
- On-Device · Processing that happens on the phone or computer itself rather than on a cloud server
- LLM (Large Language Model) · An AI model trained on large amounts of text to perform tasks like conversation and writing
- Thinking Mode · A feature that reveals the model's step-by-step reasoning before it gives a final answer
- finetune · Additional training applied to an already-trained model to adapt it for a specific task
- LiteRT · A lightweight runtime software optimized for running AI models efficiently on-device
Repository description (English)
A gallery that showcases on-device ML/GenAI use cases and allows people to try and use models locally.
Open on GitHubTrending repos
- vorssaint/vorssaint-utilsOne free menu bar app replaces a dozen paid Mac utilities
- Alishahryar1/free-claude-codeA local proxy that lets coding AI agents run on 49 free or cheap model providers instead of one paid service
- freestylefly/awesome-gpt-image-2A library of 532 reverse-engineered prompts that turn GPT-Image2 into a predictable image-making tool
- block/buzzAn open-source workspace where humans and AI agents chat, code, and review in the same rooms
- NousResearch/hermes-agentNous Research's Hermes is an AI agent that gets smarter the more you use it
- virgiliojr94/book-to-skillA tool that turns technical book PDFs into on-demand reference skills for AI coding agents
- VoltAgent/awesome-agent-skillsA single hub collecting over 1000 'how-to' manuals that make AI coding assistants act like experts
- anthropics/claude-plugins-communityA shared shelf where anyone's Claude add-ons get listed for install