Realtime

WS /v1/realtime transparently proxies OpenAI Realtime events. Server-to-server clients work by changing base URL and API key only.

Authentication#

Pass Authorization: Bearer $PLEUM_API_KEY or x-api-key on the handshake. Do not put the key in the query string.

Connection and session flow#

After the connection opens, configure the session with session.update (for example, output_modalities and audio.output.voice), then exchange Realtime events. Server events are relayed unchanged. The proxy aggregates response.done usage and settles when the connection closes.

Node.js example#

import WebSocket from "ws";

const model = "gpt-realtime-2.1";
const url = `wss://apirouter.pleum.ai/v1/realtime?model=${model}`;

const ws = new WebSocket(url, {
  headers: {
    Authorization: "Bearer " + process.env.PLEUM_API_KEY,
  },
});

ws.on("open", () => {
  ws.send(JSON.stringify({
    type: "session.update",
    session: {
      type: "realtime",
      output_modalities: ["audio"],
      audio: {
        output: { voice: "alloy" },
      },
    },
  }));
});

ws.on("message", (data) => {
  const event = JSON.parse(data.toString());
  console.log(event.type);
});

Connection errors and recovery#

Billing and limits#

Credits are held on connect and settled from response.done usage (text/audio tokens) when the session ends. A successful upstream session is billed at least 1 KRW, rounded up to a whole won.