Bell gets your agent’s reply ready sooner.

const bell = new WebSocket("wss://api.rabellabs.com/v1/conversation");
bell.onopen = () =>
  bell.send(JSON.stringify({ key: process.env.RABEL_API_KEY }));

// your speech-to-text's words so far, each time it has new ones
stt.on("words", (text) =>
  bell.send(JSON.stringify({ type: "heard", text })));
// its final words, when it ends the turn
stt.on("endOfTurn", (text) =>
  bell.send(JSON.stringify({ type: "end_of_turn", text })));

bell.onmessage = (e) => {
  const m = JSON.parse(e.data);
  if (m.type === "ready") stt.start();
  // start your reply on Bell's guess
  if (m.type === "guess") reply.start(m.text);
  // the guess that matched, or null
  if (m.type === "end_of_turn") reply.keep(m.guess_id);
};

URL = "wss://api.rabellabs.com/v1/conversation"

async with websockets.connect(URL) as bell:
    await bell.send(json.dumps({"key": os.environ["RABEL_API_KEY"]}))

    # your speech-to-text's words so far, each time it has new ones
    async def on_words(text):
        await bell.send(json.dumps({"type": "heard", "text": text}))

    # its final words, when it ends the turn
    async def on_end_of_turn(text):
        await bell.send(json.dumps({"type": "end_of_turn", "text": text}))

    async for msg in bell:
        m = json.loads(msg)
        if m["type"] == "ready":
            stt.start(on_words, on_end_of_turn)
        elif m["type"] == "guess":
            reply.start(m["text"])  # start your reply on Bell's guess
        elif m["type"] == "end_of_turn":
            reply.keep(m["guess_id"])  # the guess that matched, or None