Dictation overview

Dictation is the simplest thing Hyprcore does. Set a shortcut. Press it. Speak. Your words appear wherever your cursor is—Slack, your code editor, an email, a search bar, anywhere a text field accepts input.

What makes it different

• Local by default. A whole family of speech models runs on your Mac—Whisper, Parakeet, Canary, Moonshine, SenseVoice, Breeze, and GigaAM. Audio never has to leave the machine.

• Works in any app. Hyprcore pastes via macOS Accessibility, so it works in native apps, Electron apps, browsers, and the terminal.

• Two modes. Toggle (press once to start, press again to stop) or push-to-talk (hold the shortcut while speaking).

• Optional cleanup. Run the transcript through an LLM—Claude, GPT, Llama, or your own Ollama—to fix grammar, change tone, or apply a custom template.

• Say it, don't type it. With Action Mode, lead with "Hyper …" and Hyprcore runs your words as an instruction—drafting the email or reply—instead of transcribing them.

• Recognizes your words. Add names, jargon, and acronyms to your custom vocabulary so the recognizer spells them the way you mean.

• History. Every transcription is saved locally so you can copy it again later.

How a dictation runs

1. You press the dictation shortcut.

2. Hyprcore opens the microphone and shows the recording overlay.

3. Silero voice activity detection filters silence so the model only sees real speech.

4. You press the shortcut again (or release it, in push-to-talk mode).

5. Your selected model transcribes the audio locally.

6. Optional post-processing runs.

7. The text is pasted into the app that was focused.

8. The session is saved to dictation history.

What's next

• Set up shortcuts — Toggle vs push-to-talk, paste behavior, and conflict-free hotkeys.

• Post-processing — Templates, LLM cleanup, and tone presets.

• Action Mode — Say "Hyper …" and run a command instead of transcribing it.

• Custom vocabulary — Teach the recognizer your names, jargon, and acronyms.

• History — Find old dictations, replay audio, manage retention.

• Speech models — Pick the right model for accuracy vs speed.