Stream Analysis

For live calls and voice agents, stream audio to Voxtar over a WebSocket and receive incremental biomarker readings as the patient speaks — instead of uploading a finished file. Ideal for turning an in-progress triage or monitoring call into a screening with no extra step.

Connecting

Open a socket to the streaming endpoint and authenticate in the first frame. Send raw PCM (16kHz, mono) or Opus frames; receive JSON events.

const ws = new WebSocket(
  "wss://api.voxtar.hallelx2.com/v1/analyze/stream",
);
ws.onopen = () => {
  ws.send(JSON.stringify({
    type: "start",
    api_key: process.env.VOXTAR_API_KEY,
    patient_id: "pt_8kQ2v9",
    encoding: "pcm_s16le",
    sample_rate: 16000,
    panels: ["respiratory", "neurological"],
  }));
};
// then stream audio chunks
mic.on("data", (chunk) => ws.send(chunk));

Events

The server emits partial readings every ~1s and a final complete event with the same shape as POST /analyze when the stream closes.

{ "type": "partial", "elapsed_ms": 2000,
  "panels": { "respiratory": { "score": 0.55, "drift": "watch" } } }
{ "type": "complete", "id": "an_5Tz7Lm",
  "flags": [{ "code": "respiratory_drift", "severity": "review" }] }

Send { "type": "stop" } to finalize, or just close the socket. The completed analysis is persisted and retrievable via GET /analyses/{id}.