Layer diagram of ShoutFlow's on-device stack: your app, ShoutFlow, MLX clean-up, WhisperKit and Apple Silicon, all running locally.

WhisperKit and MLX: the on-device stack behind ShoutFlow

A technical tour of the two open-source projects that let ShoutFlow transcribe and clean up speech entirely on Apple Silicon, with no cloud round trip.

ShoutFlow's local mode stands on two open-source projects. WhisperKit turns speech into text on the Neural Engine; MLX runs the language model that turns that raw text into clean prose. Both run entirely on your Mac. This note is a tour of what each layer does and why it is fast enough to feel instant. [1] [2]

WhisperKit: speech to text on the Neural Engine

WhisperKit takes OpenAI's Whisper models and compiles them for Core ML, Apple's inference stack. The heavy lifting lands on the Neural Engine, which is designed exactly for this shape of work, leaving the CPU and GPU free for everything else you are doing. Real-time transcription of a quantised model is comfortably achievable on any recent M-series chip. [1]

Streaming, not batch

Because inference is local, there is no batch boundary to wait behind. WhisperKit can begin emitting tokens a fraction of a second after you start speaking, which is why ShoutFlow's overlay can show a live preview while you are mid-sentence.

MLX: the clean-up pass

Raw transcription is accurate but literal: filler words, false starts, spoken punctuation. The clean-up stage is a small language model running on MLX, Apple's array framework for machine learning on its own silicon. MLX shares memory between CPU and GPU, which keeps a multi-gigabyte model responsive on a laptop. [2]

WHAT EACH LAYER IS RESPONSIBLE FOR
LayerInputOutput
WhisperKit16 kHz microphone audioLiteral raw transcript
MLX clean-upRaw transcript plus tonePunctuated, de-fillered prose
ShoutFlowClean prose plus app contextText inserted at the cursor

Why open source matters here

A privacy claim is only as good as your ability to check it. Both WhisperKit and MLX are public, audited and maintained in the open: WhisperKit by Argmax, MLX by Apple's own machine learning research team. ShoutFlow's privacy story does not depend on trusting a black box. [3]

The best privacy architecture is the one a sceptical engineer can verify over a weekend.

Honest limits

On-device models are not magic. Very rare vocabulary and heavy accents still favour the largest cloud models, which is precisely why BYOK exists as an opt-in. The full local pipeline requires Apple silicon, and Intel Macs are not currently supported. [4]

FREQUENTLY ASKED

What are WhisperKit and MLX?

WhisperKit is an open-source project that runs OpenAI's Whisper speech models natively on Apple Silicon via Core ML. MLX is Apple's own open-source array framework for running language models efficiently on M-series chips. ShoutFlow uses WhisperKit to transcribe and MLX to clean up the text, all on-device.

Does on-device dictation work offline?

Yes. Because WhisperKit and MLX run entirely on your Mac, local mode works with no network connection at all: on a plane, in a dead zone, or with Wi-Fi switched off. Only the initial model download and optional BYOK providers need a connection.

Which Macs support the full local pipeline?

The full local pipeline requires Apple silicon: any M-series Mac running macOS 14 or later. Intel Macs are not currently supported.

REFERENCES

  1. WhisperKit: on-device speech recognition for Apple SiliconArgmax, Inc. · accessed 1 July 2026
  2. MLX: efficient and flexible machine learning on Apple SiliconApple Machine Learning Research · accessed 1 July 2026
  3. Robust speech recognition via large-scale weak supervision (Whisper paper)arXiv · published 6 December 2022 · accessed 1 July 2026
  4. Apple Platform Security: Neural EngineApple Inc. · accessed 1 July 2026

Talk faster than you type.

ShoutFlow turns natural speech into clean text in any Mac app. On-device by default, pay once, no subscription.

$25 ONCE · YOUR VOICE NEVER LEAVES YOUR MAC BY DEFAULT