Local Model Chat and Thoughtful Prompting
AI Chat is the central workspace for running language models directly on an Android device. Choose a model from the available cards, let it initialize, and use a multi-turn conversation for drafting, explanations, brainstorming, or other text tasks. The app keeps the main chat action easy to find while making local inference the defining experience.
Model cards can include practical details such as model size, recommended status, license links, and a Try it entry. Chat settings also expose controls such as thinking support and generation parameters, so experienced users can explore how different models behave while beginners can start with a recommended option. This makes the app useful for private experiments, lightweight work away from a desktop, and learning how model choice affects responses.
Agent Skills for Practical Mobile Tasks
Agent Skills adds a task-oriented layer to local language-model use. Instead of stopping at a conversation, you can open a skill, give it an instruction, and let the model work through a defined task with tools or structured guidance. Examples include fact-grounding with Wikipedia, interactive maps, visual summary cards, and community-built skills.
The app also leaves room for customization: skills can be loaded from a URL or imported locally, and creators can use the project guidance to shape their own skill files. This is a natural fit for developers testing agent patterns, learners exploring tool-assisted prompting, and users who want repeatable workflows on a phone. The experience stays exploratory, with each model and skill combination offering a different balance of speed, capability, and task depth.
Image Questions, Audio Notes, and Prompt Lab
Ask Image and Audio Scribe turn the same local model concept toward visual and spoken input. In Ask Image, add a picture from the camera or photo library and ask for an explanation, identification, or visual answer. Audio Scribe accepts a recording and turns speech into text, with translation available for supported workflows.
Prompt Lab provides a more controlled space for single-turn experiments. You can try reusable tasks such as summarizing, rewriting, or drafting and adjust model parameters when you want to compare outputs. Together, these entry points make Gallery more than a text chat: it can support quick visual questions, spoken notes, language practice, and prompt experiments when you have suitable models and enough device resources.
Model Library, Benchmarks, and Everyday Experiments
Google AI Edge Gallery also works as a hands-on model library rather than a single-chat client. Browse available open-source models, download compatible choices, import LiteRT-LM models through Hugging Face model-card URLs, and keep the options you want ready for later tasks. A model-management view helps separate discovery from actual use.
For deeper experiments, benchmark tools let you compare model performance on your own hardware, while Mobile Actions and Tiny Garden demonstrate how local models can connect language prompts with device-oriented tasks or a small interactive experience. These tools suit developers, AI enthusiasts, and technically curious Android users who want to evaluate on-device behavior, understand hardware trade-offs, and move between serious testing and approachable demos in one place.
