Introduction

React Native AI runs AI models on your users’ devices, through the Vercel AI SDK. Each provider uses the models its platform offers, or ones you ship, so you can swap the model per platform and keep the rest of your code.

import { apple } from '@react-native-ai/apple'
import { streamText } from 'ai'

const result = streamText({
  model: apple(),
  prompt: 'Plan a 3-day trip to Kyoto',
})

Why On Device

• Private. Prompts and replies stay on the device. Cloud models are opt-in.
• Offline. No network round trip, and it works in airplane mode.
• No server costs. No API keys for system models, and no per-token bills.

Providers

Provider	Platform	Models	Runtime
Apple	iOS	Built in: Apple Intelligence, speech and embeddings	Native, provided by the platform
Core AI	iOS 27	Yours: language, image, speech and vision models	Native, provided by the platform
ADK	Android	Built in: Gemini Nano, or cloud Gemini	Native, provided by the platform
Llama	iOS and Android	Yours: GGUF models from Hugging Face	Third-party binding, llama.rn

Built-in models need no download and work on supported devices right away. With Core AI and Llama, your app downloads or bundles the models, and we point you to ready-to-use ones.

Every provider needs the AI SDK’s polyfills.

Tools

• Conversation History keeps long chats within the small context windows of on-device models.
• DevTools shows every AI SDK call as a trace in Rozenite, for on-device and remote models alike.
• JSON UI builds React Native views from tool calls, and is designed for small models. It’s experimental.