SayIt

Free, native macOS voice dictation with AI polish.
Hold a key, speak, and clean, polished text lands at your cursor — in any app.

⬇️ Download the latest SayIt.dmg · All releases

English · 简体中文

SayIt turns your voice into polished text anywhere on your Mac. It's built for people who already talk to AI all day — vibe coders, writers, anyone who would rather speak than type. Hold the key, say what you mean, and the cleaned-up text appears right where your cursor is.

Your speech is transcribed locally on your Mac, accelerated by the Apple Neural Engine — the same NPU that powers on-device AI. It's quick, and it stays quick no matter how long you spoke: measured on an Apple M4 Pro with the recommended large-v3-turbo model, transcription runs in about 0.4 seconds — an ~9-second clip became text in ~0.4 s, roughly 20× faster than real time, and a short phrase finishes just as fast. Exact speed depends on your Mac, but the work always runs on-device, fully offline, and your audio never leaves your machine.

On macOS 26 and later, SayIt defaults to Apple's built-in speech engine — the new on-device SpeechAnalyzer model, with nothing to download. It's faster still: transcription is effectively instant, so quick that the "Transcribing…" indicator barely has time to appear before your words land at the cursor. Older macOS — or your own preference — falls back to the bundled WhisperKit model above.

If you vibe-code, you almost certainly already pay for ChatGPT / Codex or Claude. SayIt polishes your dictation through that same account or API — cleaning up filler words, punctuation, and phrasing — so there's no second subscription and nothing extra to pay.

Local transcription is free and offline. Polishing rides on the AI plan you already have. You never pay SayIt anything — and polishing is optional, so you can turn it off and SayIt stays 100% free and fully offline.

1. Hold or tap the trigger key (right ⌥ Option by default).
2. Speak.
3. Release or tap again — the polished text appears at your cursor.

That's it. Trigger key, interaction mode, which AI polishes your text, language, and microphone are all in Settings.

Speech — Apple Speech (macOS 26+), local model, or cloud API	Polish — ChatGPT/Codex login or your own API key, plus style and model

• macOS 14 or later, Apple Silicon recommended.
• macOS 26+: nothing to download — the speech model is built into the system.
• macOS 14–15: about 1–2 GB of free disk for the local WhisperKit model, downloaded automatically on first launch.

Download SayIt.dmg from the latest release, open it, and drag SayIt into Applications. It's Developer ID–signed and notarized, so it opens with no Gatekeeper warning.

On macOS 26 and later, SayIt is ready immediately: it uses Apple's built-in speech model, so there is nothing to download and dictation is near-instant from the very first try. On macOS 14–15, SayIt automatically downloads the local WhisperKit model (~1–2 GB, one time) on first launch — you can watch the progress in Settings and the menu bar — followed by a one-time preparation (the model is compiled for the Apple Neural Engine, a minute or two that first run); after that, dictation stays near-instant. Either way, grant Microphone and Accessibility in System Settings → Privacy & Security — Accessibility lets SayIt listen for the global hotkey and insert text into other apps (on macOS 26 you'll also be asked once for Speech Recognition).

To turn on AI polish, sign in to your ChatGPT account manually in Settings → Polish — this is the default and uses your existing ChatGPT/Codex subscription (you can also switch to your own OpenAI, Anthropic, or DeepSeek API key). Until you sign in, SayIt still works and inserts the raw transcription, just without polishing.

• Local speech recognition is 100% on-device and offline — your audio never leaves your Mac.
• Cloud speech and AI polish run only if you turn them on, using your own account or key.
• Your clipboard is preserved — SayIt restores whatever you had copied after inserting text.

Issues and pull requests are welcome. A couple of conventions: the Xcode project is generated by XcodeGen from project.yml (add files under App/ or SayItCore/Sources/ and re-run xcodegen generate), and docs, code comments, and commit messages are in English.

MIT

Built with Claude Code.