Connect and test the voice stack

Configure the local services

Open NautPin's Settings panel (menu bar icon → Settings).

Speech tab:

• Transcription model: Parakeet TDT 0.6B v3 with the European multilingual variant if you dictate in more than one language. (English-only? Stay on Whisper Base for now and upgrade later.)
• First use downloads the model. ~600 MB, one-time.

AI Enhancement tab → Providers:

• Add a new provider, type Custom (OpenAI-compatible).
• Display name: LM Studio or Ollama (whichever you set up in the model-server lesson)
• Base URL:
  • LM Studio → http://127.0.0.1:1234/v1
  • Ollama → http://127.0.0.1:11434/v1
• API key: literally any string — neither LM Studio nor Ollama check it
• Default model: whatever model ID you loaded / pulled in the model-server lesson

AI Enhancement tab → Assignments:

• Set both Transcription Enhancement and Note Enhancement to use LM Studio + your model.
• Prompt preset: leave on clean.

Voice Output tab:

• Kokoro server URL: http://127.0.0.1:8880 (or whichever interface from step 2)
• Default voice: bf_emma
• Hotkey for Read Selected Text: pick what you want. ⌘⇧S works well.

Test dictation: focus a text field, press your dictation hotkey, speak a sentence, press it again. The cleaned-up text should land at your cursor.

Test TTS: highlight any text, press your Read hotkey. Emma should read it aloud.

It’s working if

Dictation inserts text at your cursor, AI enhancement uses your local model, and the Read hotkey speaks selected text through Kokoro.