Google AI Edge Gallery is an open-source Android and iOS application from Google's AI Edge team that brings large language model inference entirely onto mobile devices — no internet connection required, no cloud API, no data leaving your phone. With 19,000+ GitHub stars and nearly 900 gained in a single day as Gemma 4 launched, it has become the most-watched on-device LLM showcase on GitHub. The app supports multi-turn AI Chat with visible chain-of-thought thinking mode, multimodal Ask Image (point your camera or upload photos and ask questions), real-time Audio Scribe for voice transcription and translation, Agent Skills that augment the local LLM with integrated Wikipedia and maps tools, and Mobile Actions that lets a FunctionGemma 270M fine-tune control device functions offline. Prompt Lab provides real-time generation parameter controls (temperature, top-k, top-p) and a built-in Model Benchmark tool measures performance across device hardware. The latest release (v1.0.11, April 2026) emphasizes the Gemma 4 family for advanced reasoning, multi-step logic, and creative tasks within mobile memory constraints. Built on Google's LiteRT inference runtime (formerly TensorFlow Lite), models are discovered and downloaded via Hugging Face integration, giving users access to the broader open-source model ecosystem. On-device AI eliminates cloud inference latency, preserves privacy completely, and enables reliable offline operation — advantages that become more significant as mobile neural processing units grow more capable. Gallery targets Android 12+ and iOS 17+, with hardware acceleration via NPUs and GPUs on supported devices. As a production-quality open-source reference implementation backed by Google's AI Edge team, it provides developers a high-fidelity blueprint for building private, offline-capable AI features on mobile platforms.