🎙 Talk to your vault
Hold the mic, speak your question, release. Replies stream back as text and audio.
🛰 Real-time speech-to-speech via OpenAI gpt-realtime.
Tap Start, then just talk — the model hears you and
replies in the same audio stream. Network latency only; no whisper +
TTS round-trip. Conversation history stays in this panel only.
🌐 Realtime voice via Inworld's pipeline (STT → LLM router → TTS-2, end-to-end <1s latency). Audio relays through this server's WebSocket proxy because Inworld auth uses API keys, not ephemeral tokens. From ~$0.015/min — far cheaper than OpenAI Realtime.
🤖 Realtime speech-to-speech via xAI grok-voice-think-fast-1.0
(voice Ara). OpenAI-Realtime-compatible over a
WebSocket proxied through this server. Flat ~$0.05/min — far cheaper than
OpenAI Realtime. Conversation history stays in this panel only.