Free and open sourceAGPL-3.0-or-later. Read the code on GitHub.

Free dictation for Mac, Windows and Linux.

You talk, Hushpen types it into any app.
No subscription. No account.

Coming soonAndroid and iOS keyboards, and models on your other devices.

Hears

A small native app. Light on memory.

See the benchmarks →

Messy speech. Clean text.

Fillers, repeats, and take-backs are removed.

The flow bar

Every state in one small capsule. Click it.

Listening

Rule cleanup

Fillers and take-backs go.

Heard

Pasted

Any key you like

Pick the keys for every action.

Dictate on your phone

Coming soon. A keyboard for Android and iOS, with the dictation on the phone.

Use your other devices

Coming soon. A phone or a server runs the models.

AI cleanup, when you want it

A local model rewrites the text for where it goes.

Heardhi team the docs the tests and the numbers all look good can we ship friday thanks

Command Mode

Select text, hold Right Option + Right Shift, and speak.

History

Search, play, and export what you said.

TXTSRTVTTJSON
A History entry: raw text, text after the rules, after AI cleanup, and final text

Dictionary and word packs

Names and terms, spelled your way.

Manual

Drop audio. It waits in a queue.

Drop audio files here
Interview.wavWaiting

Read aloud

Text, in a voice you pick.

Heart with 30% Adam

Dictionary and word packs

Spells your words right.

Names, places, and terms come out the way you write them. Use your own list, or a word pack.

Command Mode

Edit by voice.

Say the change, and a model on your computer rewrites the selected text in place.

Works offline

No internet needed.

Speech turns into text on your computer. It keeps typing with Wi-Fi off.

Read aloud

Reads text aloud.

Choose a voice, mix two, or clone your own. It reads text from any app.

AGPL-3.0 license

Free forever. Open source.

No subscription, no account, no word limit. Anyone can read the code and change it.

Benchmarks

30 minutes of talk, written out in 17 seconds. Lower is better.

Seconds to write out one recording on an Apple M4 Pro Mac, from a 5 second note to a 30 minute meeting.

0 s20 s40 s60 s80 s100 s120 s5 s15 s30 s1 min2 min5 min10 min30 minMacWhisper 95.5 sVoiceInk 94.2 sBuzz 69.0 sHushpen Turbo 64.8 sHushpen Small 34.5 sHushpen Parakeet 17.0 sHushpen Base 16.1 sFluidVoice 8.5 sHandy 4.3 s
0 s20 s40 s60 s80 s100 s120 s15 s1 min5 min30 minMacWhisperVoiceInkBuzzHushpen TurboHushpen SmallHushpen ParakeetHushpen BaseFluidVoiceHandy
Each point is the middle of 3 runs, with the model already loaded. The light marks show the fastest and slowest run. Handy was not run past 5 minutes. Hushpen Parakeet was timed again after a speed fix, on 5 s, 1, 5 and 30 minute files only. Buzz's times include about 13 s of app start and model load for every file. Hushpen Base was slower on the 15 second file than on the 30 second file in every run, and is shown as measured. See the benchmarks →

Local models. Downloaded once.

Parakeet hears 25 European languages. Whisper adds Hindi and more. S1-mini tidies the text.

Your voice stays on your devices.

On your own devices

Audio and text stay on your computer and the devices you pair, unless you pick an endpoint.

No telemetry

Hushpen sends no usage data. Update checks are opt-in. Update checks send the app name, platform, and version. We count them per day and keep nothing that identifies you.

Checked downloads

Models download only when you ask, and every hash is checked.

Open source

AGPL-3.0 licensed. Signed and notarized on macOS.

Your dictation, on every device.

Free on Mac, Windows and Linux today. Android and iOS are coming soon.

The same app on Linux.

X11 desktops, x64 and arm64.

Download for Linux
Prefer a terminal?

Install Hushpen. The script checks SHA256SUMS.

curl -fsSL https://hushpen.samhq.app/install.sh | sh
Other downloads

The same app on Windows.

Windows x64. No admin rights needed.

Download for Windows
Prefer PowerShell?

Install Hushpen. The script checks SHA256SUMS.

irm https://hushpen.samhq.app/install.ps1 | iex
Other downloads

Hushpen for Android.

A full keyboard with a voice bar. Dictation runs on the phone.

Coming soon
  1. 1
    Type on a full keyboardGboard-style keys, with the voice bar above them.
  2. 2
    Tap the voice bar and talkDictation and AI cleanup run on the phone. No cloud.
  3. 3
    Share its modelsTurn on Server mode. Your computer can use the phone's models over the local network, in the background or with the app open.

Hushpen for iOS.

A keyboard for every app. It listens in the Hushpen app.

Coming soon
  1. 1
    Tap the mic on the keyboardIt starts a flow session in the app. A Live Activity and the Dynamic Island show that it listens.
  2. 2
    Go back and tap the check markSwipe back to your app. The text goes in where you were typing.
  3. 3
    Serve in Server modeKeep the app open and charging, with the screen dimmed, and your computer can use the phone's models.

Use models on your other devices.

Let a phone or a server do the work. Everything stays on your own devices.

Coming soon
  1. 1
    Pair a deviceScan a QR code (nearby, encrypted with Noise), or connect over SSH or Tailscale.
  2. 2
    Pick where models runTurn on Use models on other devices. Speech to Text and Text cleanup each list devices in order, with latency and status.
  3. 3
    Dictate as usualThe first device that answers does the work. This computer is the fallback.
  • Your phoneAndroid serves in the background or with the app open. iOS serves in Server mode, with the app open, charging, and the screen dimmed.
  • Your serverhushpen-host is a headless program for servers. Add one over SSH (for example ssh oci2) or Tailscale.
  • No cloudAudio and text move only between your own devices. Or choose Do not load models on this computer.

Questions.

Answers on privacy, languages, and memory. Not here? Open an issue on GitHub.

Is Hushpen free?

Yes. Hushpen is free and open source under the AGPL-3.0-or-later license. There is no subscription, no account, and no word limit.

Is Hushpen open source?

Yes. The code is on GitHub under the AGPL-3.0-or-later license. You can read it, build it, and change it.

Does my voice leave my computer?

No. Hushpen records and transcribes on your computer. Your audio and text leave it only if you set up your own endpoint for AI cleanup. With Devices (coming soon), audio goes only to the phones and servers you paired. There is no cloud.

Does it work offline?

Yes. Dictation runs on your computer with Parakeet v3 or a Whisper model, so it works with Wi-Fi off. You need the internet only to download a model, and only when you ask. The update check is off until you turn it on.

Does Hushpen use Whisper?

Yes. It offers Whisper Large v3 Turbo, Whisper Small, and Whisper Base, run through whisper.cpp. It also offers Parakeet v3, the recommended model.

Which systems does it run on?

macOS 13 or later on Apple silicon. Linux x64 and arm64 with an X11 session and a Vulkan or OpenGL driver. Windows x64, as a preview. On Wayland, the global key and paste are not available. Android and iOS apps are coming soon.

Is there an Android or iPhone app?

Not yet. Both are coming soon. The Android app is a full keyboard with a voice bar, and dictation and AI cleanup run on the phone. The iPhone app is a keyboard that starts a flow session in the Hushpen app, and the Dynamic Island shows it while it listens.

How is Hushpen different from Wispr Flow, Superwhisper, or MacWhisper?

Hushpen is free and open source, and speech to text runs on your computer. Wispr Flow transcribes in the cloud and has a free plan with a weekly word limit. Superwhisper and MacWhisper also run models on your computer, and their paid plans add features. Read the comparison of dictation apps.

Can my phone run the models for my computer?

Yes, with Devices, which is coming soon. Turn on Use models on other devices, then pair a phone with a QR code (nearby, encrypted with Noise), or a computer with SSH or Tailscale. Speech to Text and Text cleanup each list devices in order, and Hushpen uses the first that answers. This computer is the fallback. An Android phone can serve in the background or with the app open. An iPhone serves in Server mode, with the app open, charging, and the screen dimmed.

What is hushpen-host?

A headless program for servers. It serves models to your computer the same way a phone does. It is part of Devices, which is coming soon.

Which languages?

Nemotron Unified, the default model, and Parakeet v3 both hear English and 24 other European languages. Whisper Large v3 Turbo covers Hindi and many more. Whisper Small translates speech to English.

Which keys does it use?

Left Option + Tab starts hands-free dictation, and the same keys stop it. Hold Right Option to talk, and let go to finish. Right Option + Right Shift is Command Mode, Ctrl + Cmd + V pastes the last text, Ctrl + Cmd + S reads aloud, and Esc cancels. On Linux and Windows, double tap Right Alt (or press Right Alt + Space) for hands-free. Hold Right Alt to talk. Change any of them in Settings.

What does AI cleanup do?

It rewrites the text for where it goes, with the General, Message, Email, or Notes profile. It uses a local model (S1-mini by Superwhisper for English, or Qwen3.5 2B) or your own OpenAI-compatible endpoint, such as Ollama or LM Studio. It is off until you turn it on. Rule-based cleanup is on by default.

How much memory does it use?

The app used about 15 MB after a day of running. The speech engine runs as its own program, hushpen-asr, and holds the model: about 400 MB with Parakeet kept loaded. In Settings, the engine can unload the model after 5, 15, 30, or 60 idle minutes.

Hushpen is free.

Open source, and it runs on your own computer.

Downloading HushpenOpen it from Downloads.