hold-dictate GitHub

Hold a key. Speak. Let go.

Your audio goes to Groq for transcription and isn’t stored here.

Voice typing for your whole computer. Hold Alt+Space, talk, and your words are typed at the cursor, in any app. Any language, or several in one sentence.

curl -fsSL https://hold-dictate.vercel.app/install.sh | sh
irm https://hold-dictate.vercel.app/install.ps1 | iex

Free with a Groq keyOpen source

One shortcut. Every app.

No window to open, nothing to click. hold‑dictate waits in the background until you hold the keys.

  1. 01

    Hold Alt+Space

    Anywhere you can type. A small pill appears while it listens.

  2. 02

    Speak

    Say what you want to write, in your language, pauses and all.

  3. 03

    Let go

    About a second later the text is typed where your cursor is.

Set up in a minute.

One command installs it and starts it at login. Then a settings page opens in your browser for your key. Updates install themselves.

Terminal
$ curl -fsSL hold-dictate.vercel.app/install.sh | sh
==> Installing uv (Python tool installer)
==> Installing hold-dictate v0.5.3
Installed 1 executable: hold-dictate
==> Starting hold-dictate
Will start at every login, and is running now.
==> Opening the settings in your browser. Add your API key there.
Opening the settings in your browser:
http://127.0.0.1:47811/?t=…
$

Groq

Recommended

Free tier · about 1 s

Whisper large-v3. Fast, and free for everyday use. Key steps below.

OpenAI

Paid · 1–2 s

If you already have an OpenAI account.

Get a key

Your server

Varies

Anything with an OpenAI-style /audio/transcriptions endpoint.

Offline

Free · slower

Runs on your computer. No account, larger download.

Get a free Groq key

  1. 1Open console.groq.com/keys and sign in.
  2. 2Click Create API Key and copy it.
  3. 3Paste it into the settings page and click Check and save. It stays on your computer.

Premium

Three extras on top of plain dictation, running on your own Groq key. Premium isn’t for sale yet.

Edit by voice

Select, then speak

Select text in any app and say what to change: “shorter”, “more polite”, or “reply that I can’t make it”. hold-dictate rewrites it for you.

Writing that fits the app

Automatic

Short and casual in chat, full sentences in email, and commands kept exactly as you said them in a terminal or code editor.

A dictionary that learns

Automatic

Fix a misheard name once and it’s spelled right from then on. It learns from your corrections by itself.

Questions

Short answers. The full guide is in the README.

What is hold-dictate?

A free, open-source push-to-talk dictation (speech to text) app for Windows, macOS and Linux. Hold a hotkey, speak, let go, and the text is typed at your cursor in any app.

Is hold-dictate free?

Yes. It needs a free account (sign in with Google once) and a speech-to-text key. Groq has a free tier, and offline mode needs no key.

Is hold-dictate open source?

The base is. Everything on GitHub is open source under the MIT license, and you can run it by itself, with no account. The version you install from this site adds a private package with the account and the Premium features. That package isn’t open source.

Which languages does it support?

Any language Whisper supports, around 100, including English, Spanish, French, German, Arabic, Hindi, Chinese, Japanese and Hebrew. Pick one or several in the settings and mix them in one sentence.

Which computers does it run on?

Windows, macOS and Linux, on both Wayland and X11. It works in any app where you can type, including chat, email, terminals and code editors.

Can it work offline?

Yes. Install the local extra and it transcribes on your computer with faster-whisper: no key, no audio leaves your machine. The first run downloads the model.

How is it different from the dictation built into my system?

It is the same shortcut in every app on all three systems, hold-to-talk instead of a mode you switch on, uses Whisper large-v3 on Groq in about a second, and is open source.

What happens on first run?

macOS: allow Microphone and Accessibility / Input Monitoring when asked. Linux: log out and back in once so keyboard access takes effect. Windows: nothing extra; if SmartScreen or your antivirus asks, allow Python.

Is my audio stored?

Not by hold-dictate. Audio goes straight to the provider you chose, or stays on your computer in offline mode. Check your provider’s privacy policy.

How do I change the key, hotkey, languages, model or look?

Open the settings: click the hold-dictate icon next to the clock (Windows) or in the menu bar (macOS), choose hold-dictate settings from the app menu (Linux), or run hold-dictate settings.

Which model does it use?

Automatic picks the one we recommend for your service: Whisper large-v3 on Groq. To choose another, open the settings and click Choose another next to Model.

How do I get updates?

Automatically. hold-dictate checks every 10 minutes and installs a new version while you are not dictating. You can turn this off, or update right away, in the settings.

Something doesn’t work.

Run hold-dictate doctor. It checks the microphone, keyboard access, paste tools and your key, and says how to fix each.

How do I uninstall?

hold-dictate service uninstall, then uv tool uninstall hold-dictate.

Stop typing what you can say.

Copy install command