Grammar cleanup is optional. Set grammar.enabled = false for transcription-only mode, or choose a backend with grammar.backend. Backends are registered in src/whisper_voice/backends/__init__.py. Non-selected backends must stay lazy and must not initialize at startup.

​
Backends

BackendRequirementsRuntime
Apple IntelligencemacOS 26+, Apple Silicon, Apple Intelligence enabledOn-device Foundation Models
OllamaOllama installed and runningLocalhost at http://localhost:11434
LM StudioModel loaded and local server startedLocalhost at http://localhost:1234 by default; private LAN allowed
DisabledNoneTranscription only

​
Apple Intelligence

Apple Intelligence is the default configured backend, but grammar is disabled by default. Enable it only when macOS and Apple Intelligence support are available. 
[grammar]
backend = "apple_intelligence"
enabled = true

​
Ollama

ollama pull gemma3:4b-it-qat
ollama serve
wh backend ollama

[ollama]
url = "http://localhost:11434/api/generate"
check_url = "http://localhost:11434/"
model = "gemma3:4b-it-qat"
keep_alive = "60m"

​
LM Studio

Download and load a model in LM Studio, then start the local server from the Developer tab. 
curl http://localhost:1234/v1/models
wh backend lm_studio

[lm_studio]
url = "http://localhost:1234/v1/chat/completions"
check_url = "http://localhost:1234/"
model = "google/gemma-3-4b"

​
Selected-text modes

When grammar is enabled, Local Whisper can transform selected text:
ShortcutMode
Ctrl+Shift+GProofread
Ctrl+Shift+RRewrite
Ctrl+Shift+PPrompt engineer
Results go to the clipboard.