ai

Why we included this project

Mobile teams that want LLM features without running their own inference server get a genuinely useful on-device layer here. The library wraps Apple's Foundation Models, llama.rn for GGUF files, and the MLC runtime behind an API that mirrors the Vercel AI SDK, so web developers can carry over familiar patterns to React Native instead of learning a new stack. Beyond chat and text generation it also handles embeddings for local RAG, plus transcription and speech synthesis, which fits offline-first or privacy-sensitive apps where user data should not leave the phone. Model handling is refreshingly explicit: Apple models run immediately on iOS, while GGUF and MLC models need a download step with progress reporting and an explicit prepare and unload cycle to manage memory. There is also an optional dev-tools profiler that captures OpenTelemetry spans from AI SDK requests, handy for teams debugging mobile inference in the field.