Introduction

Ultra Whisper turns your voice into text right on your Mac. Press a key, speak, and your words are transcribed — and optionally cleaned up by an LLM — then pasted straight into whatever app you're using.

How it works

Every dictation runs through the same short pipeline:

1. Record — a global hotkey starts recording from your microphone.
2. Transcribe — speech is turned into text. By default this happens on your Mac with a local Whisper model; no audio leaves the device.
3. Refine (optional) — a Profile can send the transcript to an LLM to fix grammar, remove filler, or match a tone. This step is off unless you enable it.
4. Replace — your deterministic word Replacements are applied so names and terms always come out spelled the way you want.
5. Paste — the result is pasted into the frontmost app.

What makes it different

• Local-first. The default Whisper model transcribes entirely offline. Cloud speech-to-text and LLM cleanup are opt-in and use your own API keys (BYOK).
• Profiles. Bundle a voice model, an optional LLM, and a tone, then let the right Profile activate itself based on the app you're in.
• Your vocabulary. Teach it names with Custom Words and fix spellings with deterministic Replacements.

Requirements

• macOS 14 Sonoma or later (Apple silicon recommended; Intel Mac also supported)
• 8 GB of memory (16 GB recommended)
• Microphone and Accessibility permissions
• Optional: an OpenAI or Groq API key for cloud models

Intel Macs are supported, but local speech recognition (Whisper) is faster on Apple silicon. If speed is an issue on an Intel Mac, see Troubleshooting.

Next, see Install & set up to get up and running.