CoincideFeature
Dictation
Coincide’s composer has a microphone. Speak a prompt instead of typing it, with the engine you prefer: Apple’s speech recognition on your Mac, a Whisper model that runs offline, or OpenAI or Groq with your own API key.
Dictate a message
- Click the microphone in the composer, or choose Edit › Dictate. The first time, macOS asks whether Coincide may use the microphone.
- Speak. With Apple’s engine, the words appear as you talk.
- Click again to stop. The text waits in the composer for you to check and send, or goes straight out if you turn on “Send when I stop”.
Everything is in Settings › Voice: the engine, the language (automatic by default), which microphone to use, and a level meter to test it. Testing the microphone records nothing.
Choose an engine
| Engine | Where it runs | Good to know |
|---|---|---|
| Apple (on device) | On your Mac | The default. Partial text appears as you speak. macOS downloads a language’s speech model the first time you dictate in it, or when you click Download Now. |
| Whisper (on this Mac) | On your Mac, offline | You download a model once (see below), then it works without a network. |
| OpenAI | OpenAI’s transcription API | Uses your OpenAI API key. Models: GPT Transcribe (recommended), GPT-4o Transcribe, GPT-4o mini Transcribe and Whisper. |
| Groq | Groq’s API | Uses your Groq API key. Models: Whisper Large v3 Turbo (faster) and Whisper Large v3 (more accurate). |
Local Whisper models
Whisper runs through WhisperKit. Coincide bundles no models: they download from Hugging Face only when you click Download in Settings › Voice, are checked against their published checksums, and run offline afterwards.
| Model | Download | Trade-off |
|---|---|---|
| Tiny | about 79 MB | Fastest, least accurate |
| Base | about 149 MB | Fast, good for clear speech (the default) |
| Small | about 489 MB | Slower, more accurate |
| Large v3 Turbo | about 648 MB | Most accurate; best on Apple silicon with 16 GB of memory or more |
API keys for OpenAI and Groq
Paste your key in Settings › Voice and choose Test Key; Coincide keeps it in your Keychain. If Codex is signed in with an OpenAI API key, “Use Codex’s Key…” copies that key from ~/.codex/auth.json into Coincide’s own Keychain item. It never reads ChatGPT sign-ins, and changing the key in Codex later doesn’t change Coincide’s copy.
Dictate with a keyboard shortcut
Turn on the keyboard shortcut in Settings › Voice to dictate without reaching for the mouse. It’s off by default, and while it’s off Coincide watches no keys for dictation. The default is the 🌐 fn key; you can record another key with ⌘, ⌃ or ⌥, or a single modifier on its own.
- Hold to talk, tap to toggle (the default): hold the key while you speak and let go to stop, or tap once to start and again to stop.
- Hold only or Tap only, if you prefer one.
The shortcut works while a Coincide window is in front. If macOS also does something when you press 🌐, set “Press 🌐 key to” to Do Nothing in System Settings › Keyboard; Coincide’s Voice pane links there.
Privacy
- With Apple’s engine or Whisper, your audio stays on your Mac.
- With OpenAI or Groq, the recording is sent to that provider to be transcribed, under your account and its data policies.
- Recordings are held in memory only while they’re transcribed.
The privacy policy covers dictation in the same terms.
Works with every agent
Dictation fills Coincide’s composer, so it works the same whichever agent the thread uses. It pairs well with appshots: capture a window, then say what you want done with it.