Speak naturally. Zansay writes it cleanly.

A voice dictation app for Mac. It lives in your menu bar; you hold a shortcut, speak naturally, and it types clean text into whatever app is focused: your email, chat, notes, browser or editor. It also reads text aloud. Speech recognition runs on your Mac by default.

Yes. Zansay Alpha is the free Zansay product: not a trial, not a promotion, and not free-until-we-change-our-minds. It keeps receiving updates while remaining Alpha, and Alpha will remain free.

No. Zansay Alpha has no subscription, no tiers and no account. Download it and use it.

Alpha is the name of the free Zansay product line, the original Zansay. It's a product name, not a warning label: the app is signed, notarized, kept up to date, and used daily by the person who builds it.

Apple Silicon Macs (M1 or later) running macOS 14 or later. There is no Intel build. Apple's on-device cleanup model additionally requires macOS 26; the other cleanup options work from macOS 14.

Yes, that's the point. Zansay pastes the finished text at your cursor in the app you're already using: Mail, Messages, Notes, Slack, browsers, documents, coding tools. It works anywhere macOS lets an app insert text at the cursor, which is almost everywhere. Where pasting isn't possible, the text is on your clipboard and the app says so. Secure password fields are detected and deliberately never typed into.

Not by default. Audio is recorded only while you're dictating and is transcribed by a model running on your Mac. If you opt into a cloud engine (for speech, cleanup or read-aloud), the relevant content is sent to that provider under your own API key, and the app's Privacy screen states this in plain words whenever it's true.

Only for three things: downloading the app, fetching your chosen voice model on first use, and checking for updates. Dictation itself (recognition, cleanup passes, history, the on-device read-aloud voice) works entirely offline with the default engines.

The language picker offers English, Spanish, French, German, Italian, Portuguese, Dutch, Yoruba, Arabic, Hindi, Chinese, Japanese and Korean, plus auto-detect. The fast on-device engine covers English and major European languages; the Whisper engine covers the full list. The app tells you if a chosen combination isn't supported rather than guessing.

On-device: NVIDIA's Parakeet family running as CoreML models on the Apple Neural Engine (the default), and OpenAI's Whisper running through whisper.cpp on the GPU. Model downloads are SHA-256-verified against pinned upstream releases. Optional cloud engines from providers such as OpenAI, Groq, Deepgram and ElevenLabs can be connected with your own API keys.

macOS dictation types raw words. Zansay is built around what happens after recognition: filler removal, punctuation, number formatting, your own vocabulary and correction rules, per-app writing styles, so what lands is ready to send, not ready to edit. It also keeps a searchable history, never loses a failed dictation, and reads text back to you.

Accessibility is what lets Zansay paste text into other apps at your cursor. Without it, Zansay still works: transcriptions are copied to the clipboard and the app says so. It also uses this permission to detect password fields, so it never types into one.

Not yet. Zansay Alpha is built for Mac; a good Windows version would be its own substantial piece of native work, so we're not promising one.

Updates are checked for and installed from inside the app, over a signed update feed; your settings, history and dictionary carry straight over. For support, email hello@zansay.com and you'll be talking to the person who builds it.