LLM polish
After dictation completes and before the text is inserted, an LLM rewrites the spoken transcript. It is configured separately from the voice-command intent model, and can reuse the same Keychain keys per provider.
Styles
Section titled “Styles”| Style | Role |
|---|---|
| Verbatim | Fix clear ASR errors only |
| Light polish | Drop fillers / self-corrections, light formatting |
| Deep polish | Clear written form, keep the original meaning |
Path: Settings → Text processing → LLM polish.
Providers
Section titled “Providers”Configure OpenAI, Anthropic, NVIDIA NIM, DeepSeek, Qwen, Gemini and more (see the in-app list). A valid API key is required; on failure the raw transcript is usually inserted instead.
Local Only Mode
Section titled “Local Only Mode”If Local Only Mode is on and the polish API is not on localhost, polish is skipped.
Relationship to translation
Section titled “Relationship to translation”Voice translation uses a separate hotkey and flow. It shares the LLM provider configuration from the polish side, but is a separate call.