blumi can speak replies (TTS) and accept spoken input (STT). It works in the web UI and the blugo phone app; configuration lives in the voice section of settings.json and is editable from the in-app Control Center → Voice.

Two providers:

1. Control Center → Voice → enable, pick provider elevenlabs.
2. Paste your ElevenLabs API key.
3. Tap "Authenticate & load voices" — this validates the key and fills a dropdown of your account's voices. Pick one.
4. Save. Tap the 🔊 on any assistant message to hear it.

Pick provider openai, paste a TTS API key, and set a voice (e.g. alloy). Save.

Equivalent settings.json:

"voice": {
 "enabled": true,
 "tts_provider": "elevenlabs",
 "tts_api_key": "...",
 "tts_voice": "<voice_id>",
 "tts_model": "eleven_multilingual_v2"
}

Mic input uses an OpenAI-compatible Whisper endpoint. In Control Center → Voice, set the Mic key (and the app fills in the Whisper endpoint/model). Then tap the 🎤 in the composer, speak, and the transcript is dropped into the message box.

"voice": {
 "voice_api_key": "sk-...",
 "stt_base_url": "https://api.openai.com/v1",
 "stt_model": "whisper-1"
}

• Keys are write-only over the API: the app shows saved ✓ but never returns the stored key. To change a voice later, re-enter the key to re-authenticate and reload the dropdown.
• TTS is synthesized on the gateway (which holds the key) and streamed to the phone, so the key stays on your machine.
• Voice is optional — everything else works without it.