Show HN: Realtime, expressive AI personas that you can video call (playground.keyframelabs.com)

A Show HN demo showcases a lightweight platform for realtime, expressive AI personas that you can video-call and interact with live. The web-playground lets you stream and converse with “Persona-1,” tweak LLM-related settings (with some options locked in the public playground), and adjust turn-taking controls such as “persona speaks first” and “allow interruptions.” The interface emphasizes low-friction experimentation: start a live stream, converse naturally, and modify how the agent behaves during a session. This matters because real-time multimodal persona interfaces change how researchers and product teams evaluate conversational agents: controlling initial-turn behavior and interruption handling lets you experiment with natural dialogue flow, turn arbitration, and responsiveness — all crucial for human-like interaction in support agents, tutors, or companion bots. From a technical standpoint the demo highlights practical engineering tradeoffs for live AI calls: latency and streaming synthesis constraints likely explain why some LLM/turn settings are locked, and the turn controls surface how dialogue policy primitives (who initiates, when to yield) shape user experience. The project is useful as a rapid prototyping tool for studying conversational dynamics, but it also raises familiar operational and safety considerations around moderation, privacy, and persona consistency in always-on, real-time deployments.