Voice dictation for Mac

Stop typing. Start talking.

Hold a key, speak naturally and your voice turns into text in any app. Everything happens on your Mac — nothing travels to the cloud.

Like Wispr Flow, but free and private.

macOS 14+ · Apple Silicon · Free

Mail — New message
To: team@company.com
Subject: Quarterly report
Your dictation shows up here, right at the cursor…
Hold Right ⌘ to dictate
Hold Right ⌘ and watch the magic ↑
In action

Watch it dictate in 11 seconds.

Hold Right ⌘, speak naturally, release… and the text lands already punctuated, right where the cursor is. Nothing to open, and nothing leaves the Mac.

100% local
No cloud, no accounts
Neural Engine
Accelerated Whisper
Push-to-talk
In any app
Private by design
Your voice never leaves the Mac
One key

Hold Right ⌘. That's the whole interface.

No windows, no menus, no mouse. Put the cursor where you want to write, hold the key and talk. It works with the keyboard you already have.

External keyboard with a Spanish layout: the Right ⌘ key marked “Hold”, the only one you need to learn.

Try it: press and hold the highlighted ⌘ — or your physical Right Command.

How it works

Three steps. Zero friction.

Nothing to open, no copy-paste. It works on top of the app you're already using.

01
⌘ Hold

Press and hold

Hold your key. ChacharApp starts listening instantly, with no window in the way.

02

Speak naturally

Dictate the way you'd talk to a person. No weird commands, no forced pauses.

03
✓ Text ready

Release and done

The text lands right where your cursor is, already punctuated, in any application.

Privacy + speed

Whisper runs on your Neural Engine. Nothing leaves your Mac.

We run the speech model locally on Apple Silicon, on the Neural Engine — whichever one you pick, always on your machine. No servers, no subscriptions, and no words of yours travelling anywhere. The app doesn't track you or analyse how you use it: zero telemetry and zero cloud dependency.

0
data sent
~1s
from a phrase to text
minutes, no limit

Measured on Apple Silicon with the bundled model and the app warm. It varies with the chip, the model you pick, how long you speak and whether you enable the optional LLM cleanup.

  • No telemetry
  • No usage analytics
  • No account
  • No network in normal use
You say:

“Send the deck to cubernetes over slack before eod.”

ChacharApp writes:

Send the deck to Kubernetes over Slack before EOD.

Accuracy

It tunes to the way you speak.

Add your own vocabulary: names, acronyms, technical jargon and brands. With phonetic matching you say “cubernetes” and it writes “Kubernetes” — without ever typing the misspelling into a rule.

Phonetic glossaryProper nounsTechnical acronymsAuto punctuationMultilingual
Full control

Everything is configurable. Even the engine.

Pick your shortcut, your microphone and the speech model. Use the bundled one, download another variant or import your own (WhisperKit CoreML) — and it always runs locally, never in the cloud.

ChacharApp General tab: push-to-talk shortcuts with Right ⌘ enabled, hands-free mode and the microphone policy.
Push-to-talk keys, hands-free mode and the microphone privacy policy.
ChacharApp Models tab: Whisper large-v3-turbo ready, with more variants downloadable from the app.
The bundled model ready out of the box; download other variants or import your own. Whichever you pick runs locally.
History + privacy

Every dictation, stored locally. Only you see it.

A searchable local history keeps the raw recognition next to the final text: you see what it heard and what it fixed. Nothing leaves the Mac, no account, no telemetry. My favourite: I dictated its own name and Whisper heard “Chacha Wrap”. What got inserted: “ChacharApp”.

ChacharApp History tab: a real entry dictated into the browser (Google Chrome), with the raw recognition next to the corrected text.
The History tab: the raw recognition “Chacha Wrap” stored under the “ChacharApp” that actually got inserted.
Against the alternatives

Why choose ChacharApp.

Feature
ChacharApp
macOS dictation
Cloud apps
Processing
On your Mac
On your Mac
Servers
Full privacy
Partial
Whisper on Neural Engine
N/A
N/A (cloud)
Custom vocabulary
✓ phonetic
Sometimes
Offline
Partial
Price
Free · MIT
Included
Subscription
ChacharApp first-launch assistant: three steps that tick themselves off as they complete.
The first-launch assistant: three steps that tick themselves off as they complete.
First launch

Three steps and you're dictating.

Download the .dmg, open it and the app itself walks you through it: microphone, Accessibility and the speech model, which downloads on its own the first time. The checklist ticks itself off. No account, no weird setup.

View the code on GitHub
Behind the product

What this product demonstrates

Beyond being the tool my daily work runs on, ChacharApp is the clearest evidence of how I build AI products: it covers the whole arc, not just the model call.

  • Identifying and defining a real problem, rather than looking for an excuse to use a model.
  • Designing an extremely simple interaction: hold a key and talk.
  • Building the entire application: Swift, SwiftUI, macOS integration and distribution.
  • Integrating speech recognition and language models, deciding where each one earns its place.
  • On-device processing and privacy as an architectural decision, not a marketing line.
  • Personalisation through a phonetic glossary, adapting to how each person actually speaks.
  • Deterministic guardrails against errors and hallucinations, including the delete-only contract.
  • Comparing raw transcription against processed text: corrections you can audit.
  • Measuring latency and performance, and publishing the conditions those numbers hold under.
  • Onboarding, configuration and distribution: the unglamorous part that decides whether anyone uses it.

ChacharApp is not a model demo. It is a complete product designed around a real user need and the limitations of AI.

Your voice, your text, your Mac.

Download it for free and start dictating in under a minute.

Download ChacharApp

macOS 14+ · Apple Silicon · Free