Web Voice SDK

Run a voice agent from your own website UI with the headless StrideOps voice SDK.

The headless voice SDK lets a website talk to a StrideOps voice agent with its own buttons, visualizer and transcript, instead of the embedded chat widget. Audio runs over WebRTC directly between the visitor's browser and the agent.

It is a single script with no dependencies and no build step:

<script src="https://app.strideops.ai/widgets/voice-sdk.v1.js"></script>

It also loads as <script type="module">. Either way it defines window.StrideVoice.

The SDK connects through a Chat Widget that has the Voice AI channel turned on. In Grow > Chat Widgets:

Create a widget of kind Voice AI (or add the Voice AI channel to an all-in-one widget) and pick the voice agent.
Under Installation > Allowed website origins, add every exact origin that will run the SDK, for example https://example.com, https://www.example.com and, for local development, http://localhost:3000. Origins are exact: scheme, host and port must match, with no path.
Optional: raise Maximum simultaneous voice calls (default 2, max 25) if many visitors may call at once.
Publish the widget and copy its ID.

Requests from any origin that is not on the allowed list are refused, both by the browser (CORS) and by the server.

Quick start

<button id="talk">Talk to our AI</button>
<script src="https://app.strideops.ai/widgets/voice-sdk.v1.js"></script>
<script>
  const call = StrideVoice.create({ widgetId: "YOUR-WIDGET-ID" });
  const button = document.getElementById("talk");

  call.on("call-start", () => (button.textContent = "End call"));
  call.on("call-end", () => (button.textContent = "Talk to our AI"));
  call.on("volume-level", (agentLevel) => {
    button.style.transform = `scale(${1 + agentLevel * 0.2})`;
  });
  call.on("message", (m) => {
    if (m.type === "transcript" && m.transcriptType === "final") console.log(m.role, m.transcript);
  });
  call.on("error", (e) => alert(e.message));

  button.onclick = () => (call.isActive ? call.stop() : call.start());
</script>

Start the call from a click or tap so the browser allows microphone access and audio playback.

React

"use client";
import { useEffect, useRef, useState } from "react";

const SDK_URL = "https://app.strideops.ai/widgets/voice-sdk.v1.js";

function loadSdk(): Promise<void> {
  if ((window as any).StrideVoice) return Promise.resolve();
  return new Promise((resolve, reject) => {
    const script = document.createElement("script");
    script.src = SDK_URL;
    script.onload = () => resolve();
    script.onerror = () => reject(new Error("Voice SDK failed to load"));
    document.head.appendChild(script);
  });
}

export function useStrideVoice(widgetId: string) {
  const callRef = useRef<any>(null);
  const [active, setActive] = useState(false);
  const [level, setLevel] = useState(0);

  useEffect(() => {
    let cancelled = false;
    loadSdk().then(() => {
      if (cancelled) return;
      const call = (window as any).StrideVoice.create({ widgetId });
      call.on("call-start", () => setActive(true));
      call.on("call-end", () => { setActive(false); setLevel(0); });
      call.on("error", () => setActive(false));
      call.on("volume-level", (v: number) => setLevel(v));
      callRef.current = call;
    });
    return () => { cancelled = true; callRef.current?.stop(); };
  }, [widgetId]);

  return {
    active,
    level,
    start: () => callRef.current?.start(),
    stop: () => callRef.current?.stop(),
  };
}

API

`StrideVoice.create(options)`

Option	Type	Description
`widgetId`	`string`	Required. The published chat widget ID.
`apiBase`	`string`	Defaults to `https://app.strideops.ai`. Use your white-label app domain if you have one.
`contact`	`{ fields, consent?, message? }`	Only needed when the widget requires its contact form before voice. See below.

Returns a call object.

Call object

Member	Description
`start(options?)`	Requests the microphone, opens a visitor session (reusing it across calls in the same tab), and connects. Resolves when connecting has begun; `call-start` fires once audio is live. Rejects with `{ code, message }` on failure. `options` accepts `challengeToken` and `contact`.
`stop()`	Ends the call and releases the microphone. Safe to call at any time.
`setMuted(boolean)`	Mutes or unmutes the visitor's microphone.
`isMuted()`	Current mute state.
`isActive`	`true` while a call is connected.
`on(event, handler)` / `off(event, handler?)`	Subscribe or unsubscribe.

Events

Event	Payload
`call-start`	None. Audio is connected.
`call-end`	`{ reason }`: `stopped`, `ended`, `time_limit`, `connection_lost`, `page_exit` or `error`. Fires only for calls that connected; failures before that fire `error` instead.
`volume-level`	`(agentLevel, micLevel)`, both `0..1`, about 20 times a second. `agentLevel` is the agent's voice.
`message`	`{ type: "transcript", role: "user" \| "assistant", transcript, transcriptType: "partial" \| "final" }`. Partial user transcripts update as the visitor speaks.
`speech-start` / `speech-end`	The agent started or stopped talking.
`error`	`{ code, message }`.

Error codes

Code	Meaning
`origin_not_allowed`	The page origin is not in the widget's allowed origins.
`network_error`	StrideOps could not be reached. A blocked CORS request also shows up this way, so check allowed origins first.
`mic_denied`	The visitor blocked the microphone.
`unsupported`	The browser lacks WebRTC or microphone support, or the page is not served over HTTPS.
`rate_limited`	Too many calls in a short time. Limits: 3 voice calls per visitor per hour, and an hourly organization limit of 10x the concurrency setting (minimum 20).
`concurrency_limit`	The maximum number of simultaneous voice calls is in use.
`consent_required`	The widget requires its contact form first and no `contact` was provided.
`challenge_required`	The widget has the spam challenge on and no Turnstile token was provided.
`session_expired`	The visitor session expired and could not be renewed.
`unavailable`	Voice is unavailable (widget unpublished, out of credits, agent misconfigured, or the call could not connect).

Contact form and spam challenge

If the widget's contact form or required consent is on, voice only starts after the visitor submits it. Collect those details in your own UI and pass them in:

call.start({
  contact: {
    fields: { name: "Ada Lovelace", email: "ada@example.com" },
    consent: true,
    message: "Requested a demo call",
  },
});

The fields keys must match the widget's form fields. For a pure demo, simply turn the contact form off on that widget.

If the widget's spam challenge (Cloudflare Turnstile) is on, render Turnstile with the site key returned by the widget config and pass the token as call.start({ challengeToken }). Most sites leave the challenge off and rely on the built-in rate limits.

Billing and limits

Voice calls draw from the organization's credits exactly like calls through the embedded widget. Each call is capped at the widget's Maximum voice call minutes.

Demo

A minimal working page lives at https://app.strideops.ai/widgets/voice-sdk-demo.html. Copy it to your own site, set the widget ID, and add that origin to the widget.

Set up the widget