vsip-sdk

JavaScript / TypeScript SDK for the VSIP Voice Events API — real-time turn detection, identity-gated barge-in, and speaker verification. Runs in the browser and Node 18+.

npm install vsip-sdk

Browser quickstart

import { VSIPClient, VoiceSession, type Turn } from "vsip-sdk";

const client = new VSIPClient({
  apiKey: "vsip_...",              // browser: sent as ?token= (proxy in prod)
  url: "wss://api.vsip.io/v1/stream",
});

const session = new VoiceSession(client, {
  onTurn: async (turn: Turn) => {
    if (!turn.shouldRespond) return;          // verified: not your enrolled user
    const text = await myStt(turn.audio);     // Int16Array PCM
    const reply = await myLlm(text);
    await session.speakGate();                // don't talk over a turn
    client.agentSpeaking();
    await myTtsPlay(reply);                    // stop this in onBargeIn
    client.agentIdle();
  },
  onBargeIn: () => myTtsStop(),
});

await client.connect();
session.start();

// Feed 16kHz mono Int16 PCM frames from the Web Audio API:
scriptProcessor.onaudioprocess = (e) => {
  const f32 = e.inputBuffer.getChannelData(0);
  const i16 = new Int16Array(f32.length);
  for (let i = 0; i < f32.length; i++) i16[i] = Math.max(-1, Math.min(1, f32[i])) * 0x7fff;
  session.sendAudio(i16);
};

What VoiceSession handles for you

Same field-learned patterns as the Python SDK: onset-lookback turn slicing (1.0s / 1.5s for barge-ins), pre-lock fast path + post-lock speaker_verification gating (turns held until the verdict, so an impostor is never answered), speakGate(), queued turn dispatch, and a capped idle buffer. Gate your agent on turn.shouldRespond.

VSIPClient alone gives you the typed event stream (client.on(handler)), control commands (agentSpeaking/agentIdle, setBargeInMode, toggleLock, reset), and transparent reconnect + resume.

Node usage

Node 22 has a global WebSocket. On Node 18–20, pass one:

import WebSocket from "ws";
const client = new VSIPClient({ apiKey, url, webSocketImpl: WebSocket as never });

Types & compatibility

Fully typed (discriminated VSIPEvent union). Unknown event types and fields parse without error, so this SDK survives additive protocol growth.