Loop Qwen3-Omni native voice prototype

connecting…

Hold the button (or press & hold Space) to speak. Release to send.

TTFA
—
Total
—
Audio chunks
—
About this prototype

This is a push-to-talk web client for qwen3-omni-flash with Loop's SOUL.md + STYLE_GUIDE.md injected as system prompt. You're talking to the same personality the physical sloth would have, via native voice-to-voice model.

Evaluate these:

  1. Latency: does TTFA feel natural?
  2. SOUL adherence: does it only ask questions, never advise?
  3. Emotional awareness: try saying the same sentence in different tones (happy / tired / flat) — does the response differ?