Use on-device text embeddings in React Native. Our Vercel AI SDK provider leverages Apple's built-in NLP framework, adding powerful AI features to your app with zero extra downloads or increased bundle size.

Callstack Blog

Callstack has expanded their on-device AI SDK provider for React Native to support text embeddings via Apple's built-in Natural Language framework. The integration uses Apple's NLContextualEmbedding API, a BERT-based model producing 512-dimensional vectors for up to 256 tokens, with no additional model downloads required. The library wraps this into the Vercel AI SDK interface using embed and embedMany functions. Token-level vectors are aggregated into sentence embeddings via mean pooling, accelerated by Apple's vDSP library. This enables semantic search, text classification, clustering, and recommendation features in React Native apps with zero bundle size increase.

On-Device Text Embeddings in React Native With Apple NLP Framework

From token vectors to a single sentence embedding