FluidAudio provides fully local, low-latency audio AI capabilities on Apple devices by offloading inference to the Apple Neural Engine. It wraps state-of-the-art open-source audio models (MIT/Apache 2.0 licensed) including speech recognition, text-to-speech, voice activity detection, and speaker diarization into a Swift SDK. Developers can integrate these models with a few lines of code, with models optimized for always-on and ambient computing workloads, minimizing CPU usage while leveraging dedicated neural hardware.