Browser SDK

Text-voice in a vanilla HTML page

Drop the JS SDK into a single HTML file. The mic button captures audio, streams it to Vaani, and prints the live transcript next to the AI's spoken reply. No build tooling — just a script tag.

The SDK ships an ES module bundle. For a quick demo, point a <script type="module"> at the CDN build; for production, install with npm and bundle yourself.

<!doctype html>
<html>
<body>
  <button id="talk">Hold to talk</button>
  <pre id="transcript"></pre>

  <script type="module">
    import Converlane from "https://cdn.jsdelivr.net/npm/vaanivoice/+esm";

    const vv = new Converlane({ publicKey: "vv_pub_REPLACE_ME" });
    const session = await vv.textvoice.connect({ language: "hi-IN" });

    session.on("transcript", (chunk) => {
      document.getElementById("transcript").append(chunk.text + "\n");
    });

    const btn = document.getElementById("talk");
    btn.addEventListener("mousedown", () => session.startListening());
    btn.addEventListener("mouseup", () => session.stopListening());
  </script>
</body>
</html>

Browser sessions authenticate with a short-lived public token; never ship vv_live_ keys to the client.

Open in Replit Download as zip JS SDK reference

Replit and zip downloads are placeholders for now — bucket wiring is tracked as a follow-up.