Thanks — I'll answer your question about using VAD on top of Gemini Live to save costs during silence. Gemini Live already includes built-in VAD/turn detection and it’s enabled by default. Using a separate VAD or adding a LiveKit turn-detection pipeline that requires a streaming STT will either duplicate that functionality or add extra STT costs. So adding external VAD on top of Gemini Live does not reduce costs — using Gemini Live’s native turn detection is the most cost-effective option unless you need custom behavior that Gemini’s detector can't provide. If you need help deciding whether a custom turn detector is required for your use case, tell me: • what custom behavior you need (examples: language-specific rules, shorter/longer silence threshold, speaker merging), and • whether you’re already using a streaming STT provider. Sources: Turns overview | LiveKit Documentation | https://livekit.com/pricing.md|, with interactive elements