Most voice agents still take turns: listen, respond, listen again.
Interrupt one mid-sentence and it talks over you or resets. Kick off a slow lookup and you’re sitting in dead air until it’s done.
GPT-Live-1 from OpenAI for Developers doesn’t work that way. It’s full-duplex, listening and speaking at once, so it handles real interruptions and hands reasoning and tool calls off to a backend model instead of freezing the conversation while it works.
We’ve brought it into LiveKit Agents as a new plugin, GPTLiveModel, same AgentSession API you’re already building on.
Watch the demo:
