Thanks — I’m responding to your three questions: availability, how to use it now, and inference gateway compatibility. • Availability: We do not have information about when GPT Realtime 2 will be available in LiveKit. • How to use right now: You can use OpenAI’s current Realtime API with LiveKit via the OpenAI Realtime plugin (

openai.realtime.RealtimeModel

) in LiveKit Agents (Python or Node). Example flow: a. Install the plugin (Python or Node package). b. Set

OPENAI_API_KEY

in your environment. c. Create an

AgentSession

with

llm=openai.realtime.RealtimeModel(...)

(you can pick a voice or use

modalities=["text"]

and supply a separate TTS in the session for a half-cascade architecture). For implementation details, see the Realtime models quickstart and examples in the LiveKit Agents docs: https://docs.livekit.io/agents/start/voice-ai/ and https://docs.livekit.io/agents/models/realtime/ • Inference Gateway: The Inference Gateway does not list or confirm support for Realtime model variants. Realtime models are used via the Realtime plugin and are not shown as available through the gateway’s

inference.LLM

list (which covers non‑realtime OpenAI LLMs). So you cannot currently rely on the Inference Gateway to proxy Realtime models. If you’d like, tell me which SDK (Python or Node) and whether you want streaming voice or text-only, and I’ll give a short example snippet you can drop into your project. Sources: Voice AI quickstart | LiveKit Documentation | Realtime models overview | LiveKit Documentation | OpenAI LLM models | LiveKit Documentation, with interactive elements