Skip to content

OpenAI Realtime API: using the VIDAI Control Plane as the backend

The Realtime API is WebSocket-based (bidirectional voice + text streaming to a persistent connection). The control plane's current shipping surface is HTTPS request/response; WebSocket proxying isn't in scope today.

For real-time voice/text agents you have two paths:

  • Point your Realtime client directly at OpenAI's wss://api.openai.com endpoint. Your Chat Completions / Responses traffic can still route through the control plane for cost tracking, guardrails, and chargeback — you'd have two connection paths in the same app.
  • Use HTTPS streaming instead. For most agentic conversation flows (including tool-calling flows), streamed Chat Completions or the Responses API delivers responses fast enough that Realtime's WebSocket path isn't strictly needed. Both HTTPS-streaming shapes ARE routed by the control plane. See OpenAI SDK and OpenAI Responses API.

What we're considering

Realtime API proxying is on our roadmap. Priority depends on customer demand — if your project needs it, please raise an issue below so we can weigh your use case.

If something's off

Raise an issue at github.com/vidaiUK/vidai-quickstart/issues describing the Realtime pattern you'd like to route. We'll respond with a timeline (or a workaround, if one exists for your specific case).

Where to go next