Nexa SDK, Run, build and ship local AI in minutes (sdk.nexa.ai)

NexaML announced that its Nexa SDK now supports Apple’s Neural Engine (ANE), bringing NVIDIA’s Parakeet v3 automatic speech recognition (ASR) to iPhone, iPad, and Mac. After previously porting Parakeet v3 to the Qualcomm Hexagon NPU, Nexa has enabled the same model to run locally on Apple silicon — delivering real-time, low-latency speech recognition that works offline and preserves user privacy by keeping audio on device. This expansion is significant because it demonstrates cross-NPU portability: developers can ship the same advanced ASR model across major mobile NPUs without routing audio to the cloud. Technically, the move implies Nexa’s toolchain handles model optimizations and NPU-specific compilation so Parakeet v3 can leverage ANE’s hardware acceleration for fast inference. For the AI/ML community, that means faster prototyping, reduced server costs, better privacy guarantees, and broader reach for on-device ML apps — a practical step toward standardized local AI deployment across heterogeneous hardware.