Android + AI Cookbook

Build AI features in Android apps with Kotlin and Jetpack Compose.

Want to add Gemini Live, use ML Kit, run AI on-device, or build an agent with ADK? Explore the runnable samples, then pick a topic to follow its learning path.

New preview: PocketCommunity — A2UI + DevEarth: native event and preparation surfaces inside one conversation. Live Gemini agent required; key stays on the companion server.

Sample apps

Each sample is an independent Android Studio project, focused on one Android + AI topic.

	PocketCards Turn your notes into knowledge. An AI flashcard app: paste study notes, generate questions and answers with Gemini, review the result, and study your saved deck. Topic: AI Fundamentals Level: Introductory AI integration Status: Runnable sample · learning content in progress Firebase AI Logic and structured model output Prompting, validation, cancellation, and failure handling Jetpack Compose, Material 3, MVVM, and Navigation 3 Local decks, favorites, daily review dates, and progress Light and dark themes; unit and device tests Browse & run → · Hands-on codelab · Learning roadmap Manual cards and the sample deck work without setup. AI generation uses your own Firebase project. Screenshots show the running app.
	PocketCook A little help, hands-free. A cooking companion: choose a recipe, follow the steps, and ask Gemini for spoken guidance while you cook. Topic: Gemini Live in Android Level: Intermediate realtime voice integration Status: Runnable core · academy update awaiting deployment Direct Gemini Live over WebSocket, without Firebase Microphone capture, PCM streaming, and speaker playback Interruptions, mute, transcripts, and lifecycle cleanup Jetpack Compose, Material 3, MVVM, and offline recipe progress Audio regression tests, physical-device checks, and light/dark themes Browse & run → · Learning roadmap · Codelab source (deployment pending) Recipes work offline. Voice uses your own Gemini key entered in the debug app; release voice is not yet available. Screenshot shows the real app before starting voice. Camera and tools are future extensions.
	PocketChat Explore ideas together. An Android Engineers AI learning companion: ask questions, watch Gemini answers stream, continue the conversation, and save useful ideas. Explore our roadmaps, codelabs, courses, and optional 1:1 mentorship through a curated resource directory. Topic: Gemini Chat for Android Level: Intermediate conversation and streaming integration Status: Runnable core · dedicated roadmap and codelab planned Direct Gemini text streaming over SSE, without Firebase Multi-turn context, Stop/Retry, and stale-response protection Room-backed history, saved answers, and draft recovery Android Engineers system instructions and curated learning links Jetpack Compose, Material 3, MVVM, and light/dark themes Browse & run → · Learning plan · Architecture Chat uses your own Gemini key entered in the debug app. History works offline. Screenshot shows the actual home screen before the branding update. Structured takeaways, tool calling, and production cloud access are future milestones.
	PocketStories Ordinary things. Extraordinary tales. A camera-to-story app in development, exploring Gemma on Android with local model downloads and a private bookshelf. Topic: Gemma on Android Level: Intermediate on-device multimodal integration Status: Building / Preview · end-to-end generation under validation CameraX capture, Photo Picker and bounded image preparation Resumable model downloads, SHA-256 verification and private storage LiteRT-LM integration, multimodal prompts, streaming and native cancellation Compose, MVVM, Room stories and draft checkpoints Model selection, hardware constraints and honest device verification Explore the preview → · Learning plan · Verification & limitations Actual app screenshot; illustration is bundled artwork. No successful AI story generation is claimed. Build and automated tests pass; the tested emulator hits a GPU buffer limit. Physical-device generation and the academy codelab remain unverified/unpublished.
	PocketCommunity Find your people. Get ready for your next developer event. Discover DevEarth events, explore venues, and prepare for your visit in one conversation. Gemini composes interactive native UI from a restricted component catalog. Topic: A2UI and agent-generated Android UI Level: Intermediate agent and UI integration Status: Live-agent preview · physical-phone flow verified · roadmap and codelab planned Official AndroidX A2UI parser, processor, and Material component catalog Surfaces, component trees, data bindings, and outgoing actions Conversation context and checkbox state sent back to Gemini Validated component graphs, controlled links, setup and retry handling Server-only Gemini credentials and an inspectable protocol flow Browse & run → · How A2UI works · Verification & limitations Actual live-generated UI on a physical Android phone. Artwork is a bundled illustration. Requires the companion server with your own Gemini key; no scripted runtime fallback. Production hosting and durable conversations are not implemented.

What do you want to build?

25 topics · 3 runnable samples · 2 previews · 20 planned apps. Start with PocketCards for AI fundamentals, PocketCook for realtime voice, or PocketChat for streaming conversations. PocketCook's expanded lessons and codelab are committed in the website repository and awaiting deployment. Each app README documents setup, verification, and remaining limitations.

Add AI to your Android app

Topic	What you’ll learn
Gemini Live in Android	PocketCook: direct voice conversations, PCM audio, interruptions, and explicit session restart.
ML Kit GenAI + Gemini Nano	Add on-device text and image features with capability checks.
On-device AI for Android	Run models locally and understand downloads, memory, and device support.
Gemini Chat for Android	PocketChat: streaming conversations, Stop/Retry, local history, saved answers, and an Android Engineers AI guide.
Firebase AI Logic	Connect an Android app to Gemini with Firebase and App Check.
Voice AI for Android	Build speech-to-text, AI responses, text-to-speech, and turn-taking.
Multimodal Android apps	Combine camera, images, audio, and text in an AI feature.

Explore open models, local search, and computer vision

Topic	What you’ll learn
Gemma on Android	PocketStories (Preview): camera-to-story integration, verified model downloads, local persistence and hardware compatibility.
LiteRT and LiteRT-LM	Run custom models locally and understand the inference runtime.
RAG and embeddings	Search personal notes and answer questions with source citations.
FunctionGemma and local tool calling	Turn a natural-language request into an approved app action.
ML Kit vision and text	Build OCR, barcode scanning, and other camera-based ML features.
MediaPipe for Android	Build live perception features such as gesture and pose detection.

Build agents and connect tools

Topic	What you’ll learn
A2UI for Android	PocketCommunity (Live-agent preview): Gemini-composed native UI, component catalogs, bound checklists, and action feedback.
ADK for Kotlin	Build JVM agents with tools, sessions, and orchestration.
ADK for Android	Put an agent in your Android app with lifecycle-aware sessions and local tools.
Koog for Kotlin	Build AI agents with the Kotlin-based Koog framework.
MCP for Android developers	Connect an Android experience to tools through a controlled backend.
Android AppFunctions	Expose app actions with typed inputs, permissions, and confirmation.

Learn the foundations and ship your app

Topic	What you’ll learn
AI for Android — start here	PocketCards: turn notes into flashcards and learn the AI fundamentals.
Production AI on Android	Handle failures, evaluation, privacy, costs, and rollout.
Gemini in Android Studio	Use AI assistance and Agent Mode to implement and debug features.
Android CLI	Build and verify Android projects from the terminal.
Android Skills	Use Android development skills to guide and check project changes.
Create agent skills for Android	Write reusable workflows for your coding agent.

Find your way around

android-ai-cookbook/
├── adk-android/
├── adk-kotlin/
├── agent-skills/
├── ai-fundamentals/
├── android-cli/
├── android-skills/
├── appfunctions/
├── firebase-ai-logic/
├── functiongemma/
├── gemini-chat/
├── gemini-in-android-studio/
├── gemini-live/
├── gemma/
├── koog/
├── litert/
├── mcp/
├── mediapipe/
├── ml-kit-genai/
├── ml-kit-vision/
├── multimodal-ai/
├── on-device-ai/
├── production-ai/
├── rag-and-embeddings/
├── voice-ai/
└── docs/                     # Feature index and contribution guidance

Browse all feature guides and their status. Each topic folder will hold its own dedicated project when implemented.

New to Android + AI?

Start with AI for Android fundamentals, then choose the topic for the app you want to build.

Build plan and engineering

Follow the app implementation order, engineering standards, and roadmap/codelab learning contract. Development uses official Android skills and project-local guidance.

Help grow the cookbook

Have a recipe idea or want to add a sample? Read the contribution guide and use the recipe template.

Maintained by Android Engineers · Android Engineers Academy · Apache 2.0