Talk instead of type.
It reads the screen.
Press a hotkey, speak, and finished text appears at your cursor — in Mail, Slack, Word, or wherever you already work. Hush can read the window you are in and draft the reply for you. Transcription runs on your Mac; your voice is never uploaded.
Free 30-day trial · Requires macOS 26 (Tahoe) or later on Apple Silicon (M1+)
Press fn, speak — the text appears at your cursor.
Speech in. Finished text out.
Dictation is the starting point. Hush also reads what is on screen, finishes the draft, and turns meetings into follow-ups — on your Mac by default.
Text where your cursor is
Press Fn (or your own shortcut), speak, and the transcription is inserted in the app you are already using. Your words appear live in the notch as you talk. No copy-paste, no switching windows.
Rough speech into ready-to-send text
Fix grammar, shorten a paragraph, draft an email, pull out to-dos, or translate. Hush ships with its own local model, so this works on every Apple Silicon Mac with nothing to set up — or use Apple Intelligence, your own LM Studio or Ollama server, or Claude.
Reply to what is on screen
Select text and dictate a response — or select nothing at all: Hush reads the window you are in (the email thread, the ticket, the chat) and drafts the reply against it. Password and secure fields are never read.
Meeting notes without note-taking
Record a call or meeting, get a speaker-separated transcript, and receive a summary — a short overview or a full protocol with decisions and action items.
Import audio you already have
Drag in an existing recording — a voice memo, podcast, or call — and get the same transcript and summary.
It learns how you write
Teach Hush names, terms, and jargon you use at work — and when you fix a word after pasting, Hush notices and offers to remember it. Past dictations stay on your Mac and are searchable.
One hotkey from speech to text.
Download, grant microphone access, and pick a shortcut. After that, Hush stays in your menu bar.
Press your hotkey
From any app. Hush starts recording immediately.
Speak
Talk at normal speed. Your words appear live in the notch as you go, and transcription runs on your Mac — Wi‑Fi can be off.
Text is inserted
The result appears at your cursor. Use an AI action if you want it cleaned up, rewritten, or turned into a reply before pasting.
You do not have to
explain the situation.
Press ⌥⌘T without selecting anything, and Hush reads the window in front of you — the email thread, the ticket, the chat — then writes against it. Say what you want to happen and the details are already there. This is what separates Hush from a dictation box: it knows what you are replying to.
- ⌥⌘T with nothing selected — the focused window becomes the context, no copy-paste
- Say the intent (“yes to both, propose Thursday”), get the whole reply
- Runs through the model you picked, so with a local one nothing leaves your Mac
- Password and secure fields are never read; switch the whole thing off in Settings
Can you confirm the delivery date, and whether the maintenance window is included in the fixed price?
Hi Anna, yes to both — the delivery date is confirmed, and the maintenance window is included in the fixed price. Could we do Thursday for the handover?
Speak freely.
Paste polished text.
You do not have to dictate perfect sentences. Describe what you need — a reply, a shorter version, or a proper email — and Hush drafts it on your Mac. Review the result, refine it in plain language, then paste. Two built-in actions cover most of the work, and you can add your own for the things you send every week.
- Works on selected text, on what you dictate next, or on the window you are in
- Two built-ins — compose a message, rewrite text — that adapt to what is selected
- Add your own actions for the formats you write again and again
- One shortcut (⌥⌘T) from any app — it uses your selection, or the whole window when there is none
Ready out of the box.
Yours to change.
Hush ships with its own model, so the AI features work the moment you install it — no setup, no API key, no account. Prefer something else? Use Apple Intelligence where your Mac supports it, point Hush at LM Studio or Ollama on this Mac or another machine on your network, or connect Claude when you want a larger model. Hush uses exactly what you picked and never switches on its own.
Built-in model
Qwen 3.5 (4B) ships inside Hush and runs fully on-device. Works on every Apple Silicon Mac with nothing to configure.
Apple Intelligence
On Macs with Apple Intelligence enabled, choose it for the fastest on-device responses.
Your own server
LM Studio, Ollama, or any OpenAI-compatible endpoint. Hush finds it, lists your models, and loads one with a click.
- Auto-detects LM Studio and Ollama on this Mac
- Change the base URL to reach a server elsewhere on your local network
- Load, unload, and switch models from inside Hush
- Claude available through your own Anthropic key — clearly marked as leaving your Mac
Stay in the conversation.
Read the notes later.
Record a meeting or call on your Mac. Afterward, you get a readable transcript — who said what, with timestamps — and a short summary with action items. Useful when you need to pay attention instead of typing.
- Microphone, system audio, or both
- Speakers separated automatically; rename them afterwards
- A short overview or a full protocol — decisions, risks, action items
- Import an existing audio file the same way
Speech never leaves.
The rest is your call.
Most dictation tools send your audio to a server. Hush transcribes on your Mac — always, with no setting that changes it. The AI text work runs where you choose: Apple Intelligence, the model bundled with Hush, or your own LM Studio or Ollama server. Pick a cloud provider like Claude and the text of those requests goes there — Hush tells you before it does, and never switches on its own.
Transcription is always local
Speech recognition and speaker separation run on your Mac and never touch the network. There is no option that sends your audio away.
Built for confidential work
Nothing you say is sent to a third-party transcription service. Helpful if you handle client, patient, or internal data.
You pick where AI runs
Apple Intelligence and the bundled model stay on your Mac and work offline. Connect a remote model and Hush says so plainly, in the settings and before it sends.
Try it for 30 days
Download, install, and use every feature. No account or card required for the trial.
Free 30-day trial · License required after · Requires macOS 26 (Tahoe) or later on Apple Silicon (M1+)
One payment.
No subscription. Your license does not expire — keep using the version you bought for as long as it runs on your Mac.
Small business under § 19 of the German VAT Act — no VAT is charged or shown.
Important: The email address you enter at Stripe is where your license key will be sent. Use an address you can actually access — there's no other way to recover the key.
Lifetime
Pay once. No renewal needed.
- License does not expire
- All features included
- Use it on your Macs
- Email support (best effort)
Secure checkout by Stripe · License key delivered to your inbox · Updates and support on a best-effort basis while the project is maintained
Common questions
It is transcribed on your Mac. Recordings are never uploaded to a cloud transcription service, and there is no setting that changes that.
Not your audio — transcription is always local. For rewriting, replies, and summaries you choose the model: Apple Intelligence and the model bundled with Hush stay on your Mac. If you deliberately connect Claude or another remote endpoint, the text of those requests goes to that provider. Hush shows you which model is active and never switches by itself.
No, not for day-to-day use. Models are downloaded once; after that, dictation, rewriting, and summaries work offline — unless you have connected a remote model on purpose.
During setup, you grant microphone access and, optionally, Accessibility permission so Hush can paste at the cursor. After that, your hotkey records, transcribes, and inserts the text.
If you start a compose action without selecting anything, Hush reads the text of the window you are in — the email, the ticket, the chat — and uses it as context for the draft. Password fields and secure inputs are skipped, and you can turn the feature off in Settings.
No. Dictation is the main workflow, but Hush also rewrites text, drafts replies against what is on your screen, transcribes meetings, summarizes recordings, and imports audio files.
Yes. In Settings → AI Provider choose Server, then LM Studio or Ollama. Hush finds the local server, lists your models, and lets you load one. Point the base URL at another machine on your network if the server runs there instead. There is also a Custom option for any OpenAI-compatible endpoint, and an Anthropic option if you want to use Claude with your own API key.
That is a common reason people choose it: your speech is transcribed on your machine, and if you stay on Apple Intelligence or the bundled model, the text never leaves either. This is not legal advice — check your own policies.
30 days free, all features, no card. After that it is a one-time payment — currently €14.99 launch price (regular €24.99).
No. You only enter an email at checkout so we can send the license key.
Apple Silicon (M1 or newer) with macOS 26 (Tahoe) or later.
25 languages for transcription, without manual switching.
The key does not expire. You can keep using the version you bought. Updates and macOS compatibility are provided on a best-effort basis while the project is maintained.