Hold hotkey
Press and hold to start dictating anywhere.
Hold one hotkey and talk. Because OfflineVoice runs entirely on your Mac, there's no cloud round-trip — so it's faster, and more private, than tools that send your voice away. And it's fully open source.
100% on-device and open source. No cloud, no account, no subscription.
You say
Today I wanted to give a quick update on the project. We've been working on the new onboarding flow and it's almost ready.
Text appears (on-device)
Today I wanted to give a quick update on the project. We've been working on the new onboarding flow and it's almost ready. We still need to fix a few edge cases and some Chinese copy. Let's review it together.
Press and hold to start dictating anywhere.
Talk at your own pace. Everything is transcribed on your Mac.
Recognized text is pasted at your cursor the moment you release.
Dictate into any text field on your Mac. The same fast, on-device flow everywhere you type.
Because everything runs on your Mac, there is no cloud round-trip. Text shows up near-instantly the moment you stop talking.
Dictate straight into the focused text field of any native Mac app — Mail, Notes, Cursor, Slack, ChatGPT, and more.
Speech recognition runs locally with a multilingual model bundled in the app, or optional Whisper. No cloud, no account, no subscription.
Audio is captured only while you hold the hotkey, and stays on your Mac.
Recognition runs locally — a bundled multilingual model, or optional Whisper.
No audio sent, no text synced, no data used for training. Ever.
Press, speak, release, done. There's no upload, no server queue, and no network latency — text lands at your cursor the moment you stop talking. Built for repeated daily writing, not one-off recordings.
A bundled on-device model that detects Chinese, English, Japanese, Korean and Cantonese by itself — mixed Chinese and English included. Near-instant, nothing to download.
Whisper (large-v3 turbo) for sharper results on English and technical content. Downloads once on first use, then works offline.
Dictation runs on your machine. Speed mode ships with its model and works out of the box; the optional Whisper model downloads once, then runs offline too. Fewer privacy tradeoffs, and fewer broken writing sessions.
OfflineVoice is designed for a one-time Mac app purchase rather than another recurring voice subscription.
Download, open the menu-bar app, grant permissions, and hold Right Option to dictate into any text field. Signed and notarized — just double-click to open.
Download for MacYes. It pastes into the focused text field across native Mac apps — anywhere you can type.
Speed (the default) uses SenseVoice, a multilingual model bundled inside the app: it recognizes Chinese, English, Japanese, Korean and Cantonese without being told which one — even mixed Chinese and English in one sentence — and is near-instant. Accuracy uses Whisper (large-v3 turbo) as an alternative for English and technical content; it downloads once, then runs offline. Native uses Apple's own recognizer, which only understands your Mac's system language.
No. Recognition runs on your Mac. Speed mode works immediately with the model that ships in the app; the optional Whisper model downloads once and then runs fully offline.
Yes. OfflineVoice is signed with a Developer ID and notarized by Apple, so you can just double-click the download to open it — no Gatekeeper warnings.
Last updated June 16, 2026
Nothing. OfflineVoice has no account, no sign-in, no analytics, and no telemetry. We do not operate a server that receives your data, and the app does not phone home.
Microphone audio is captured only while you hold the dictation hotkey, is transcribed on your Mac, and is discarded after the text is produced. The recognized text is placed on your clipboard and pasted into the app you choose. Audio and text are never uploaded to us or any third party.
Transcription runs on-device. Speed mode uses SenseVoice, a model that ships inside the app (see the NOTICE file in the app bundle for its licence). Accuracy mode uses Whisper (large-v3 turbo); the first time you use it the model is downloaded from its public host and then cached locally for offline use. Native mode uses Apple's built-in recognizer. None of this sends your speech or text off the device.
OfflineVoice asks for Microphone access (to capture your voice) and Accessibility access (to detect the global hotkey and paste into the focused app). These are used solely for dictation and for nothing else.
Your settings are stored in a plain file at ~/.config/offlinevoice/config.json on your Mac. You can read, edit, or delete it at any time.
Questions about privacy? Reach out via offlinevoice.ai.