How to dictate in any Mac app without sending your voice to the cloud

You can dictate into Slack, Gmail, Notion or your code editor on a Mac without a single second of audio leaving your computer. Open speech models like Whisper and Parakeet, running on Apple Silicon, have made local dictation a normal choice rather than a compromise. This guide shows what your options are in October 2026, which ones really keep your voice on your Mac, and how to set one up in about five minutes.

Disclosure: I make MyFlow, one of the apps in this guide. I've tried to be fair and every competitor claim links to its source. Prices were checked on official pages on 11 October 2026.

Why it matters where your voice goes

When you dictate, you say things you'd never type into a search box: client names, health questions, salary numbers, half-formed ideas. With cloud dictation, that audio is uploaded and transcribed on someone else's server. Many providers handle it responsibly, but you are trusting their retention rules, their staff access and their security.

Local dictation removes the question. The speech model is a file on your Mac. Audio goes from your microphone into memory, becomes text and is gone. No account, no upload, and it works on a plane.

There's also a speed reason to dictate at all. In a Stanford study, speech input on a phone was about three times faster than typing in English, with fewer errors (Stanford, 2016). On a keyboard, a large study of 168,000 people measured an average of about 52 words per minute (Aalto University, CHI 2018), while conversational speech runs at around 160 to 170 words per minute (Language Log analysis of the Fisher corpus).

Option 1: Apple Dictation, built into macOS

Apple's dictation is free and already on your Mac. Press the dictation key (or the shortcut you set) and talk.

Where your voice goes depends on your Mac and your language. Apple's macOS guide says: "On a Mac with Apple silicon, Dictation requests are processed on your device for supported languages" and no internet connection is required. On an Intel Mac, or in a language without on-device support, "your dictated utterances are sent to Apple" (Apple Support). Apple also says server-processed dictation isn't stored unless you opt in to Improve Siri and Dictation (Apple privacy).

Good for: quick messages, zero setup, zero cost.

Limits: you get a plain transcript. There's no cleanup of "um" and false starts, no per-app tone, no custom snippets, and you can't choose a different speech model.

How to check it's on-device: open System Settings, then Keyboard, then Dictation. The current Apple guide tells you to check there whether your requests are "processed on your device and not sent to Siri servers" (Apple Support).

Option 2: a dictation app that runs the speech model locally

A local dictation app lets you pick an open speech model, runs it on your Mac's chip, and pastes the result wherever your cursor is. Most add a global shortcut, an on-screen indicator while you talk and optional text cleanup.

These are the main ones, with what their official pages said on 11 October 2026:

For contrast, Wispr Flow, the best-known cloud option, costs $15 a month (or $12 a month billed yearly) with a free plan of 2,000 words a week on desktop. Its pricing page doesn't say that speech can be processed on your Mac.

Which speech model to pick

The model decides accuracy, speed and language. These are the ones worth knowing:

• NVIDIA Parakeet TDT 0.6B v2: English only, 600 million parameters, very fast and accurate on Apple Silicon. License CC-BY-4.0 (model card).
• NVIDIA Parakeet TDT 0.6B v3: extends v2 "from English to 25 European languages" with automatic language detection (model card).
• OpenAI Whisper: the multilingual workhorse, trained on 680,000 hours of audio (paper), with code and weights under the MIT License (GitHub). It's the one to use for Urdu, Hindi, Arabic and most languages outside Europe. Sizes run from tiny to large, plus a faster "turbo" version.

A simple rule: English speakers start with Parakeet v2. Everyone else starts with Whisper Turbo, and moves to Whisper Large if accuracy matters more than speed. Accents are handled better by bigger models, so if a small model keeps mishearing you, step up a size before giving up.

How to set up private dictation with MyFlow

Here's the full setup. It takes about five minutes, most of it waiting for the model download.

1. Download MyFlow from the download section and drag it into Applications. You need a Mac with Apple Silicon (M1 or newer).
2. Open it once. MyFlow isn't notarized by Apple yet, so the first time macOS asks you to confirm. Click Done, then go to System Settings, Privacy & Security, and click Open Anyway.
3. Allow Microphone and Accessibility. Accessibility is what lets MyFlow paste text into other apps.
4. Pick a speech model. Downloads range from about 30 MB to 1.5 GB. This is the only time MyFlow needs the internet for dictation.
5. Hold ⌥ Space and talk. Let go, and the text appears where your cursor is. Double-tap the shortcut for Hands-free mode when you have a long thought.

At this point nothing you say leaves your Mac. You can turn Wi-Fi off and keep dictating.

Optional: clean-up without giving up privacy

A raw transcript keeps every "um", false start and spoken number. AI cleanup turns it into text you'd actually send. In MyFlow, cleanup is part of Pro and you choose where it runs:

• Fully local: Apple Intelligence on supported Macs, or any model you run with Ollama. Nothing leaves your computer.
• A cloud provider with your own key: OpenAI, Anthropic, Groq, OpenRouter or Cerebras. Only the text is sent, straight from your Mac to the provider you picked. The audio never goes anywhere.

A quick privacy checklist for any dictation app

Before you trust an app with your voice, check five things:

1. Where does speech recognition run? Look for a clear "on your Mac" statement, not just "private".
2. Does it work with Wi-Fi off? Try it. This is the fastest honest test.
3. What gets sent for cleanup? Text only, or audio too? To whom?
4. Is there an account? Accounts usually mean server-side history.
5. What happens to history? Stored locally, synced, or kept by the company?

If an app's website doesn't answer these plainly, assume the answer is "the cloud".