I already have my own AI, I just need the WhatsApp side

People arrive with the same sentence: the AI assistant already works, I just need WhatsApp. That is one of the most straightforward ways to use WACO. Our built-in AI Operator does not need to be switched on, and there is no AI cost from us. You use the official number, the webhook, and the API.

Three pieces

1. Incoming message is posted by WACO to your URL as message.received.
2. Your AI thinks on your own infrastructure.
3. The reply goes out through POST /api/v1/balas.

Set up once

• An API key waco_… from the Settings page.
• Your webhook URL (HTTPS required) on the Developer page. WACO shows a whsec_… secret once; keep it, you need it to verify signatures.

A complete example, Node and Express

const express = require('express');
const crypto = require('crypto');
const app = express();

// The signature is computed over the RAW body, so do not use express.json() on this route.
app.post('/waco', express.raw({ type: '*/*' }), async (req, res) => {
  const expected = 'sha256=' + crypto.createHmac('sha256', process.env.WACO_WEBHOOK_SECRET)
    .update(req.body).digest('hex');
  const got = String(req.get('X-WACO-Signature') || '');
  if (got.length !== expected.length ||
      !crypto.timingSafeEqual(Buffer.from(got), Buffer.from(expected))) {
    return res.sendStatus(401);
  }

  // 1) Answer 200 FIRST. A slow AI must not hold the webhook.
  res.sendStatus(200);

  const ev = JSON.parse(req.body.toString('utf8'));
  if (ev.event !== 'message.received') return;      // ignore status and echo
  if (alreadySeen(ev.pesan.id)) return;             // deliveries can repeat

  // 2) Queue per number so one customer is handled in order
  queue(ev.dari, async () => {
    const answer = await yourAI(ev.dari, ev.pesan.teks);   // your own brain

    // 3) Send the reply back through WACO
    await fetch('https://waco.id/api/v1/balas', {
      method: 'POST',
      headers: {
        Authorization: 'Bearer ' + process.env.WACO_API_KEY,
        'Content-Type': 'application/json',
      },
      body: JSON.stringify({ nomor: ev.dari, teks: answer }),
    });
  });
});

app.listen(3000);

Four things that surprise people

• Answer 200 first, think afterwards. If you wait for the AI before replying 200, the delivery counts as failed and is retried, and the customer can get a double answer.
• The 24-hour window is Meta's. More than 24 hours after the customer's last message, free text is refused and you must use a template. Templates can be created through POST /api/v1/template, but still wait for Meta's approval.
• Media arrives as an ID, not a URL. Fetch the file with GET /api/v1/media/:id if your AI needs to see the image.
• Deliveries can repeat. Remember processed pesan.id values for at least a day.

If your AI is an agent

WACO also serves MCP at /mcp, so your agent can call WACO as a tool (prepare a confirmed blast, read history, send messages) without writing an HTTP integration.

Details in the API guide and the webhook guide. OpenClaw users have their own page.