Docs
Set up local processing
Choose Local to run transcription, optional AI polish, and selected-text Orb Commands on your Mac. After downloading models, speech and text processing works offline. Local has no provider API usage charges and is included in the Listen Orb trial and Mac license.
Before you start
- An Apple silicon Mac running macOS 14 or later.
- An internet connection for the initial model downloads.
- About 2.43 GB for the default speech and text models, plus the app. Larger speech models require more storage and memory.
- An active trial or license and the usual Microphone, Accessibility, and Input Monitoring permissions.
Choose and download your models
Choose Local
During setup, select Local. If you already use the app, open Settings → AI Provider… and choose Local. You do not need an OpenAI or xAI API key.
Choose speech and text models
Start with the default Fast — Whisper Base speech model and Recommended — Qwen3 4B text model. The text model handles polish and Orb Commands. Other speech choices are Balanced and Accuracy-focused; a Lightweight text model is also available. Larger models do not guarantee better results for every recording.
Download & use Local
The app checks each download and loads the selected models before making them active. You can cancel downloads. If setup fails, the previous processing choice stays active.
Try a sentence
Hold Fn, speak, and release. Local transcription starts after recording ends; live partial transcription is a cloud feature. Review names, numbers, and instructions before sending.
What stays on your Mac
Local processes audio, transcript text, any optional Contextual polish excerpt, and Orb Command selections on your Mac. If local processing fails, the app does not automatically send that recording to a cloud provider. Audio Recovery keeps the recording on its original processing path.
Model downloads, license checks, updates, and optional online features still use the internet. Local describes speech and text processing; it does not turn off every network feature. See Privacy for the full data paths.
Change models or switch providers
Use Manage local models… to download, change, or remove models. Switch between Local, OpenAI, and xAI from AI Provider…. Cloud providers need your own API key and bill usage separately.
Models load in the background when dictation begins. The first short dictation can take longer while loading finishes. Models unload after 30 idle minutes, when switching to a cloud provider, or when quitting. Speed, memory use, and output quality depend on your Mac and chosen models.