connectOpenAi fails with server_error code 42 — backend still on shut-down OpenAI Realtime Beta API?

[Abdulatif90]

Describe the bug

streamVideo.video.connectOpenAi() always fails with server_error code 42.
The agent never joins the call.

Endpoint: wss://video.stream-io-api.com/video/connect_agent
Response:

{"type":"server_error","code":"42","message":"Error:  code "}

Environment

• @stream-io/node-sdk: 0.4.24
• @stream-io/openai-realtime-api: 0.4.0 (latest)
• Model: gpt-realtime
• Call type: default
• App API key: zvgv9fafmjg3

What we already ruled out

1. OpenAI key is valid and funded (Tier 1): a direct WebSocket to
   wss://api.openai.com/v1/realtime?model=gpt-realtime with the same key
   returns session.created successfully.
2. Not an app-config issue: reproduced identically on a brand-new Stream
   app created 2026-07-05 with default settings.
3. Client already upgraded: with openai-realtime-api 0.1.4 → 0.4.0 the
   error is unchanged. The v0.4.0 client's own error message says the
   /video/connect_agent backend must be updated to the GA Realtime API and
   to contact support if the error persists after upgrading — it persists.

Repro

const call = client.video.call("default", callId);
await call.getOrCreate({ data: { created_by_id: "user-1" } });
await client.video.connectOpenAi({
  call,
  openAiApiKey: process.env.OPENAI_API_KEY!,
  agentUserId: "agent-1",
  model: "gpt-realtime",
}); // → server_error code 42

Expected

The agent connects to the call via the GA OpenAI Realtime API.

Is GA support rolled out per-app/region? Is there a flag we need enabled?
Thanks!

[oliverlaz]

hi @Abdulatif90, thanks for raising this. Let me check with our infra/backend team, and I'll get back to you.