Skip to content
Get packs

ML Kit On-Device Agent API

Build on-device Android agents with Gemini Nano, multi-turn sessions and tool calling via AppFunctions

Source Download skill
Repository
android/skills
Stars
10,278
Licence
Apache-2.0
Files
3 files

What it does

This skill gives an AI coding agent detailed, authoritative patterns for building Android apps that run an autonomous AI agent entirely on the device using Google's ML Kit GenAI Agent API (powered by Gemini Nano through AICore). It tells the agent how to set up the Gradle dependencies, create an Agent and AgentRunner, keep multi-turn chat sessions alive in a ViewModel, let the model call typed app functions declared with AppFunctions, and handle GenAiException errors. Once installed, the agent writes Android code that follows these rules rather than guessing at APIs.

Great for

  • Android developers adding a local, privacy-preserving chat or assistant feature to an app
  • Teams experimenting with on-device tool calling through AppFunctions
  • Developers who need guidance on session lifecycle and cleaning up model memory

Skip it if

  • Cloud-hosted Gemini or Vertex AI projects, which this skill explicitly excludes
  • Simple one-shot text generation with no agent loop; the skill itself suggests the Prompt API instead
  • Non-Android platforms such as iOS or web, or apps relying on an unreleased SDK: the skill targets the AICore Developer Preview and alpha library versions, so APIs may change

Try asking

  • “Add a multi-turn chat screen to my Android app that uses on-device Gemini Nano and keeps history across rotation”
  • “Let the on-device model call my app's functions to book reminders, using AppFunctions tool discovery”
  • “Write a ChatViewModel that streams tokens into a Compose UI and handles GenAiException gracefully”
  • “Run a single ephemeral query with runner.runSingleTurnAsync and free the session afterward”
Step-by-step guide: how to use ML Kit On-Device Agent APIInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 3 original files, the Apache-2.0 licence and an attribution note.

    mlkit-agent-api.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip mlkit-agent-api.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip mlkit-agent-api.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Pin the exact alpha library versions listed in the skill and check them against current ML Kit and AppFunctions release notes, since the skill targets preview APIs.
  • Ask the agent to follow the lifecycle rules: keep the session in a ViewModel and call session.close() in onCleared to free RAM.
  • Review generated code yourself and test on a supported physical device, because on-device model availability varies by hardware.

What's inside

  • references/session_lifecycle.md
  • references/tools_and_functions.md
  • SKILL.md

Redistributed unmodified from android/skills under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.