Orders come in the way
your customers already text.

Your restaurant lists its WhatsApp Business number on Google, Instagram, delivery flyers. Customer taps to chat. Meta delivers every incoming message to a webhook URL you register once during setup.

POST /api/whatsapp/webhook
X-Hub-Signature-256: sha256=a1b2c3...
{ from:"+971 50 ••• 4412",
  text: "Hi, I want to order for
         delivery to Marina Walk",
  timestamp: 1728324521 }

Each phone number gets a persistent session. First message triggers your restaurant's welcome, subsequent messages append to the running conversation. Menu, prices, and dialect notes are cached and prepended as system context.

const session = sessions.getOrCreate({
  from: "+971 50 ••• 4412",
  restaurant: "al_manara_grill",
  ttl: 7200 // 2h session window
});
session.append({ role:"user", text: msg.text });

Same Claude Sonnet 4.6 model that answers phone calls, same prompt engineering, same tool-use to add items to a running order. Language detected automatically from the incoming text. Order state maintained turn-by-turn until customer confirms.

tool_use: "add_to_order"
{ items: [
    { qty:1, name:"Chicken Shawarma" },
    { qty:1, name:"Fattoush Salad" },
    { qty:1, name:"Orange Juice" } ],
  language: "ar",
  delivery: "Marina Walk" }

The moment the customer says "yes", we call the same createKOT() function the voice channel uses. Kitchen ticket prints. Order gets a number. Confirmation goes back to WhatsApp with ETA. Same monitor portal shows it live — so you can edit the ticket before the kitchen starts if the AI misheard.

const kot = await createKOT({
  restaurant: session.restaurant,
  items: session.order.items,
  total: 77,
  channel: "whatsapp",
  delivery: "Marina Walk"
});
sendWhatsApp(from, `Confirmed ✓ Order #${kot.n}`);