Nabla launches Mac dictation that keeps the audio on the computer

Nabla released a dictation app for Mac that does the speech recognition on the computer itself, so the audio never leaves the machine. That removes a whole conversation with the privacy office about where recordings go. Nabla says 44 percent of the clinicians using it work on Apple devices.

By Atlas ... AI Editorial Collaborator

What Nabla Launched With Its On-Device Mac Dictation Product

Nabla announced Nabla Dictation for Mac on July 28, 2026, describing it as a clinical dictation application built natively for macOS that runs its full speech pipeline on the device with no audio sent to servers. The company says 44 percent of the clinicians using Nabla today work on Apple devices, and that the app works with existing dictation hardware including PowerMic and SpeechMike microphones. Nabla emphasizes a deep Epic integration and says M Health Fairview has deployed the product at scale. Chief product officer Laurent Landowski credits Apple's Core ML and Neural Engine for making the on-device pipeline possible. The announcement states no accuracy figures and makes no FDA claim.

Why On-Device Dictation From Nabla Matters to Compliance Teams

Where the audio goes is the question compliance actually asks, and on-device processing answers it in one sentence instead of a data processing agreement. The honest gap is accuracy: Nabla published no error rate for this product, and on-device models are usually smaller than server models, so a careful buyer tests dictation accuracy against whatever it is replacing rather than assuming the privacy win is free. The Epic integration is the part that decides whether clinicians actually use it.

Where the Nabla Update Comes From

Nabla is the original record behind this update. It tells us what the company published. This brief adds the market context and the method we would use to test the development against other evidence.

Nabla original source.

How We Would Research the Nabla Mac Dictation Product

The source gives us the starting point. This is how we would build the next layer of research around it.

We would start by testing word error rate on real specialty vocabulary against the dictation tool the organisation uses now.
Then we would confirm in writing what, if anything, leaves the Mac, including text, usage telemetry and crash reports.
We would check which Mac hardware generations are supported, since on-device speech leans on the newer Apple chips.