Tap the orb and speak. Whisper transcribes, a small local LLM replies and your browser reads it aloud. Models run on your device, no server and no API keys.