Talk instead of typing. Press Fn twice and your words land on the clipboard.
Dictation that runs entirely on your Mac. No API key, no subscription,
no account, and your voice never leaves the machine.
Plus the clipboard history you already wanted.
Install · Voice Input · Features · Settings · Build from Source
One command:
curl -fsSL https://raw.githubusercontent.com/anuragxxd/popy/master/install.sh | bashOr download the DMG manually — open it, drag Popy to Applications, done.
- Dictate anywhere — press Fn twice, speak, press Fn twice again
- Transcribes on-device in under a second; nothing is uploaded
- Remembers your last 25 text copies
- Click any entry to copy it back (or paste it directly into the active app)
- Cmd+Shift+V to open from anywhere, no mouse needed
- Survives restarts — your history is saved
- No dock icon, no windows, no clutter
Press the Fn (globe) key twice to start listening. The menu bar icon turns into a red microphone. Press Fn twice again to stop — a second later the transcript is on your clipboard, ready to paste.
The transcript is stored exactly as spoken. Nothing rewrites, summarises, or "cleans up" your words.
If a recording contains no speech, it is discarded without transcribing — speech recognisers otherwise hallucinate filler text ("you", "Thank you.") when handed silence. Detection is based on the audio level, never on pattern-matching the transcript, so real words are never dropped.
First run downloads a 142 MB speech model. After that it works offline.
One setup step: macOS binds the Fn key by default. Go to System Settings → Keyboard → "Press 🌐 key to" and choose Do Nothing, otherwise the emoji picker will open every time you dictate.
Transcription happens entirely on your Mac using
whisper.cpp (MIT) with OpenAI's
base.en model. There is no API key, no subscription, no per-use cost, and
your audio never leaves the machine. Typical latency is well under a second
for a short utterance on Apple silicon.
Clipboard history is stored in the macOS Keychain (encrypted at rest, app-scoped). No plaintext history is written to UserDefaults.
Voice audio is written to a temporary file, transcribed locally, and deleted immediately afterwards. No audio is ever transmitted.
All toggleable from the menu:
- Click to Copy or Click to Paste Directly — choose what happens when you click an entry
- Sound on Copy — subtle audio feedback
- Launch at Login — start Popy automatically
- Dictate with Fn Fn — enable or disable voice input
- Paste Transcript Directly — auto-paste after dictating instead of clipboard-only
- Mute Audio While Dictating — silence system output while recording so it doesn't bleed into the microphone (on by default)
Either "Paste Directly" mode simulates Cmd+V into whatever app you're using, so macOS asks for Accessibility permission the first time.
Dictation itself only needs Microphone access — watching for the Fn key requires no special permission.
macOS 12 (Monterey) or later.
Issues and PRs welcome at github.com/anuragxxd/popy.
To build from source:
git clone https://github.com/anuragxxd/popy.git
cd popy
bash create-signing-identity.sh # once — keeps macOS permissions stable
bash build-whisper.sh # once — builds the voice engine (~5 min)
bash setup.sh # builds Popy.app
setup.sh alone is enough to get a working build. The other two are one-time
setup and both degrade gracefully: without the engine, voice input is simply
reported as unavailable; without the signing identity, the app falls back to
ad-hoc signing.
Why the signing step matters. macOS records permissions against a code
signature, not an app name. With ad-hoc signing that means a content hash,
which changes on every rebuild — so Accessibility silently lapses and "paste
directly" stops working, while System Settings still shows Popy as allowed.
Signing with a certificate ties the permission to the certificate instead, so
it survives rebuilds. Run bash create-signing-identity.sh --remove to undo.
MIT