voicekey

Open-source voice dictation for macOS.
Hold Option. Speak. Release. Text appears at your cursor.

No subscription, no third-party servers. Your API key, one hotkey, and you're dictating.

Uses OpenAI's gpt-4o-mini-transcribe by default, with a pluggable provider system so you can swap in other transcription backends. Audio goes straight from your mic to the API over HTTPS, text comes back in under a second, gets pasted at your cursor. Costs about $0.003/minute with OpenAI.

Why this exists

Wispr Flow is good, but lots of companies (including OpenAI itself) block third-party services that process audio. If your employer won't let you install Wispr, or you'd rather not pay $10/month for dictation, this does the same thing with your own API key.

How it compares

	voicekey	Wispr Flow	macOS Dictation	whisper.cpp
Cost	~$0.003/min (API)	$10/month	Free	Free (local)
Privacy	Your API key, direct to provider	Audio goes to Wispr servers	Audio goes to Apple servers	Fully local
Allowed at work	Yes (your own key)	Often blocked by IT	Usually allowed	Yes
Accuracy	GPT-4o-mini-transcribe	Proprietary	Apple ML	Open-source Whisper
Latency	Sub-second	Sub-second	~1-2s	Depends on hardware
Paste at cursor	Yes	Yes	Yes	No (manual copy)
Works in any app	Yes	Yes	Yes	No (terminal only)
Open source	Yes	No	No	Yes
Setup time	2 minutes	Account + install	Built-in	Compile from source

How it works

Hold Option → Recording → Release Option → Transcribe → Paste at cursor

1. Hold the Option key. A red dot appears on screen, recording starts.
2. Talk.
3. Release Option. Audio gets sent to the transcription API, text streams back and gets typed wherever your cursor is.
4. Your clipboard is untouched. It's saved before the paste and restored after.

Install

Prerequisites

• macOS 13+
• Python 3.11+
• PortAudio: brew install portaudio

pipx (recommended)

brew install portaudio
pipx install voicekey
voicekey setup

From source with uv

brew install portaudio
git clone https://github.com/adamkhakhar/voicekey.git
cd voicekey
uv sync
uv run voicekey setup

Setup will ask for your API key (goes into macOS Keychain, never touches disk) and walk you through Accessibility and Microphone permissions.

Run

voicekey          # if installed via pipx
uv run voicekey   # if installed via uv

A mic icon shows up in your menu bar. Hold Option to dictate.

Permissions

macOS needs two permissions granted:

Permission	Why	How to grant
Accessibility	Global hotkey detection + simulating Cmd+V to paste	System Settings → Privacy & Security → Accessibility → add your terminal
Microphone	Recording audio	Dialog pops up on first use

voicekey setup checks both and opens the right Settings pane if either is missing.

Configuration

Config file is at ~/.config/voicekey/config.toml.

voicekey config                       # view all settings
voicekey config language en           # set language (ISO 639-1)
voicekey config model gpt-4o-transcribe   # higher-accuracy model (~$0.006/min)
voicekey config hotkey left_option    # trigger on left Option only

Key	Default	Options
provider	openai	openai (more coming)
model	gpt-4o-mini-transcribe	gpt-4o-mini-transcribe, gpt-4o-transcribe
hotkey	option (either)	option, left_option, right_option
language	"" (auto-detect)	Any ISO 639-1 code

Architecture

┌─────────────────────────────────────────────────────┐
│  Main thread (NSApp run loop)                       │
│  ┌──────────┐  ┌─────────┐  ┌────────────────────┐ │
│  │ CGEvent  │→ │ rumps   │  │ Overlay (NSWindow) │ │
│  │ tap      │  │ menubar │  │ red dot indicator  │ │
│  └──────────┘  └─────────┘  └────────────────────┘ │
├─────────────────────────────────────────────────────┤
│  Audio thread (sounddevice callback)                │
│  24kHz mono int16 PCM → WAV encoding                │
├─────────────────────────────────────────────────────┤
│  Transcription thread (per utterance)               │
│  Provider.transcribe() → paste at cursor            │
└─────────────────────────────────────────────────────┘

State machine: IDLE → RECORDING → TRANSCRIBING → INSERTING → IDLE

Providers are pluggable. The providers/ package defines a Protocol that any transcription backend can implement. OpenAI is the default. To add a new provider, implement transcribe() in a new module and register it.

There's a 200ms debounce on the Option key so it doesn't fire when you're typing special characters (Option+E for accents, etc).

Security and privacy

Your API key is stored in the macOS Keychain through the keyring library, not in a config file. Audio goes directly to your chosen provider's API over HTTPS with no middleman. There's no telemetry, no analytics, nothing phoning home. The clipboard gets saved before pasting transcribed text and restored right after. The whole thing is ~900 lines of Python you can read in 20 minutes.

Troubleshooting

"Failed to create event tap" - Your terminal doesn't have Accessibility permission. Add it in System Settings → Privacy & Security → Accessibility, then restart voicekey.

No audio captured - Check that your terminal has Microphone permission in System Settings → Privacy & Security → Microphone.

Text not appearing at cursor - A few apps with custom input handling don't respond to simulated Cmd+V. Most standard apps work.

PortAudio errors on install - brew install portaudio first.

Contributing

Contributions are welcome. Open an issue first if you want to discuss a change.

git clone https://github.com/adamkhakhar/voicekey.git
cd voicekey
uv sync
uv run pytest tests/ -v

License

MIT