React Native Executorch

React Native ExecuTorch is a toolkit from Software Mansion for running AI models directly on a phone or tablet inside a React Native app. Models execute locally through the ExecuTorch runtime, so features work offline, keep user data on the device and avoid per request API fees. The library covers several workloads. Computer vision includes object detection, segmentation and image classification with models such as YOLO, RF DETR, MobileNet and Segment Anything. Voice features pair Whisper for speech to text with Kokoro for text to speech. Image and text embeddings support semantic search, similarity matching and local retrieval augmented generation without a cloud vector database. Large language models and vision language models, including Llama and Phi, power chat, autocomplete and image aware assistants. An OCR plus LLM pipeline extracts text from a photo and passes it to a local model for translation or document questions. It is aimed at mobile developers building private or offline first products such as audiobook readers, voice note apps, travel and translation tools, note taking apps and photo managers. The project integrates with Expo. The site lists no pricing or paid tiers, and the toolkit is presented as a developer library rather than a hosted service.