Nabla Launches Dictation for Mac

Nabla, providers of artificial intelligence assistants in clinical care, has introduced a fully on-device clinical dictation app, Nabla Dictation for Mac, combining on-device speech recognition with deep Epic electronic health record integration.

For health systems operating on Apple devices, Nabla is combining ambient documentation with medical-grade dictation on Mac in a single AI-native experience. The offering supports documentation, communication, and care coordination tasks through one unified voice layer while maintaining compatibility with existing dictation hardware, including PowerMic and SpeechMike microphones.

"For decades, clinicians have been forced to adapt their workflows to software," said Laurent Landowski, chief product officer of Nabla, in a statement. "We believe it's time for software to adapt to clinicians. Voice is becoming the most natural interface for clinical work, but healthcare has been held back by fragmented tools and workflows that weren't designed for the AI era. Our vision is a future where clinicians can move through their entire day using a single, intelligent voice layer built around how care is actually delivered."

To deliver the Mac app, Nabla rebuilt its full speech stack, including voice activation, speech-to-text, and medical post-processing using Apple native technologies like Swift and Core ML, and taking advantage of the power of Apple silicon. Using Apple's Core ML Tools, the team migrated years of PyTorch-based speech models onto Apple Silicon. On-device, the models are optimized for the Apple Neural Engine. This enables the low latency and efficiency required for medical-grade dictation , with no audio sent to servers and no cloud dependency.

"Core ML and the Apple Neural Engine made it possible for us to bring our entire speech pipeline on-device for Mac without compromising the performance clinicians expect from medical-grade dictation," Landowski said. "That means clinicians get the speed and responsiveness they need, while keeping speech processing entirely on their device."