Intro to Profiles
What a Profile is and how to configure its models and tone.
A Profile bundles everything that decides how a dictation is handled:
- a Voice Provider and Voice Model (how speech becomes text),
- an optional LLM Provider and LLM Model (how the transcript is refined),
- and the instructions (tone/format) given to the LLM.
Choosing the voice model
Each Profile picks its own voice model in the Profile editor.
- Local Whisper (default) runs on your Mac, fully offline. Pick any installed model from the dropdown. Models that aren't downloaded yet appear greyed out with a download button; downloading happens right there. Some models are language-specific and appear greyed out unless the Profile's recognition language matches.
- Apple Speech runs on-device speech-to-text using Apple's built-in engine — no API key, and audio never leaves your Mac. It needs macOS 26 or later; on earlier macOS it falls back to local Whisper. Its language assets download from the Voice Models screen, like Whisper models.
- OpenAI or Groq run cloud speech-to-text using your own API key (BYOK).
See Voice Models and LLM Models.
Choosing the LLM (optional)
Turn on LLM cleanup to refine the raw transcript. Pick an LLM Provider (Groq, OpenAI, Anthropic, Gemini, or DeepSeek) and a model, then write instructions describing the tone and format you want — for example "Clear, professional email prose with proper punctuation."
If LLM cleanup is off, or the provider key is missing, Ultra Whisper simply pastes the raw transcript. Your word Replacements are still applied either way.
Starting from the Default profile
Ultra Whisper ships with a single Default profile that does plain voice-to-text (local Whisper, no LLM cleanup). From there you build your own profiles for the contexts you write in — for example an email profile with LLM cleanup for clear prose, a chat profile whose instructions keep it short and casual, or a code profile that captures symbols and terms verbatim. Give each one its models, tone, and the apps it should activate for.
The active Profile is shown in the recording HUD while you dictate.