Expo Blog

On-device AI is now practical for mobile apps thanks to more powerful hardware and smaller models. The react-native-executorch library by Software Mansion wraps Meta's ExecuTorch inference engine, letting React Native developers run AI models locally without ML expertise. The post covers the benefits (privacy, zero API cost, offline support, low latency) and real limitations (battery drain, RAM constraints, storage size, performance variance across devices). A step-by-step tutorial builds a real-time speech transcription app using OpenAI's Whisper Tiny model running entirely on-device via Expo, react-native-audio-api, and the useSpeechToText hook. The post also previews a growing ecosystem including react-native-rag for offline RAG workflows.

The future of AI apps is on the device: How to run AI models with React Native ExecuTorch

Building a real-world app: Voice transcription with on-device AI

The future of on-device AI and React Native ExecuTorch