A practical quick-start guide for OpenAI's GPT Realtime API, covering how to create ephemeral session tokens, choose between WebSocket and WebRTC connections, and send your first live message. Includes a working Node.js WebSocket example, session update patterns, common gotchas (like the correct field name `output_modalities`), and guidance on when to use GPT Realtime versus the simpler GPT Audio API for request-response use cases.
Table of contents
What this GPT Realtime guide coversGet your API key readyCreate a realtime session tokenPick the right connection type firstSend your first live GPT Realtime messageYou can update the session without reconnectingBuild something useful: a low-latency support assistantThe field name that is easy to get wrongWhat GPT Realtime supports wellWhen GPT Realtime is the right choiceCommon mistakes with GPT Realtime4 Impressions