SDKs
Packages that put a live call with your character in your own site or app, and help your server listen to it and check webhooks.
The SDKs speak the call's two WebSockets and check our signatures for you, so a live call in your own page is a few lines. They talk to the /v1 API, whose reference is generated from our OpenAPI file.
The packages are ready but not published: they stay private until the product has its name. The code on these pages shows how they will be used. The names @acme/client and @acme/react, and the address https://talk.madein10.com, are placeholders that will change with the name.
| Package | For | What it does |
|---|---|---|
@acme/client | the browser | Call: the microphone, the AI's voice with echo cancellation, the face, mute, end, the call's state and events |
@acme/client/server | your server (Node 22+, Deno, Bun, workers) | ControlSocket for the call's live events (transcript, status, say, context, end); readWebhook, readToolCall, verifyWebhook; forBrowser; the API's types |
@acme/react | React 18+ | useCall() and a drop-in <Conversation>: the face, the AI badge (always shown), mute, end, a joining state |
A call, end to end
Your server starts the call
With its secret key, and keeps the live events socket for itself:
import { ControlSocket, forBrowser, type CallStart } from "@acme/client/server";
const res = await fetch("https://talk.madein10.com/v1/calls", {
method: "POST",
headers: { authorization: `Bearer ${process.env.ACME_SECRET_KEY}`, "content-type": "application/json", "idempotency-key": key },
body: JSON.stringify({ character_id: "chr_…", metadata: { ticket: 7 } }),
});
const start = (await res.json()) as CallStart; // 201; a refusal is { error: { code, message } }
const events = new ControlSocket(start.events_url!);
events.on("transcript", (line) => console.log(line.who, line.text));
await events.connect();
return forBrowser(start); // to the page, without events_urlThe page joins it
It opens the microphone first, straight from the click, then asks your server, then connects:
import { Conversation } from "@acme/react";
import "@acme/react/theme.css"; // optional: the default look
<Conversation name="Alex" getCallStart={() => fetch("/api/calls", { method: "POST" }).then((r) => r.json())} />Without React:
import { Call } from "@acme/client";
const call = new Call();
button.onclick = () => call.start(getCallStart);
call.on("state", render);Webhooks arrive at your server
Signed, and checked in one line:
const event = await readWebhook(secret, req.headers, rawBody);Why in this order
- The microphone first. A refused microphone never starts a call, so it never holds money on your balance.
- The events socket stays on your server. Whoever holds
events_urlcan make the character say anything and end the call.forBrowser(start)takes it out before the CallStart goes to the page, as our own widget does. Call'sevents: trueoption shows the transcript in the page fromevents_url, for tools where the person on the call is trusted with that (your own team's, say).