GPT-Live-1 OpenAI Developers is now supported in LiveKit Agents. It's full-duplex, so it listens and speaks at the same time instead of taking turns, and it hands reasoning and tool calls off to a backend model so a slow lookup doesn't leave your agent sitting in dead air. Same... `AgentSession` API. New plugin: `GPTLiveModel`. Get started:show more

@OpenAIDevs Full-duplex is the difference between a walkie-talkie and a real conversation. Most "voice AI" is just very fast turn-taking.

@OpenAIDevs GPT-Live-1 in LiveKit Agents as full-duplex — listens and speaks at once, tools offloaded so a slow lookup doesn't leave dead air. Realtime stack that doesn't freeze mid-sentence.

@OpenAIDevs The way it hands off reasoning to backend models to avoid dead air is a game changer and we actually went deeper on this here:

@OpenAIDevs Full-duplex voice plus backend reasoning is a huge upgrade for real-time AI agents.

@OpenAIDevs And on an affordable pricing too, we tried this model when it was on beta and the price per minute was insane

@OpenAIDevs On phone interviews, full-duplex still cuts the agent mid-sentence when the candidate starts talking. We store that cancelled utterance with a trailing ellipsis so the next turn knows the sentence was cut off.

@OpenAIDevs So glad I saw your post. I switched to this today just in time to finalize my new iOS app I'm submitting tomorrow for approval.

@OpenAIDevs 全双工加后台工具调用，语音交互终于不那么像轮流发言了

@OpenAIDevs This is cool, still pretty expensive option though

@OpenAIDevs Please check your inbox

@OpenAIDevs Pretty nice - it only works with OAI agents / models for delegation , right? Or will there be an option for gpt live to delegate to a custom agent running on my backend server (in the livekit process)?