cactus-react-native — React Native SDK for on-device inference

Run local LLMs and vision models on iOS and Android with a high-performance React Native SDK powered by Nitro Modules.

Cactus-react-native delivers sub-50ms inference by running GGUF models directly on mobile hardware. It uses optimized C++ kernels and Nitro Modules to provide native speeds for models like Llama 3 and Qwen. The SDK supports 2-bit quantization to fit large models into mobile RAM and includes an optional cloud fallback for high-complexity tasks. By processing data locally, you eliminate server costs and network latency while maintaining total user privacy.

https://github.com/cactus-compute/cactus-react-native