WhatsApp assistant for a small business — live simulator

Two phones, one engine. Left: a customer writes to the business on WhatsApp. Right: the owner's WhatsApp. The assistant answers from the knowledge base (About & opening hours, Price list, Delivery, Policies), collects a complete order step by step, and every reply first goes to the owner as a draft with Send / Skip / «I'll reply myself» buttons.

Meta WhatsApp Cloud API payloadsOpenAIhuman approvalorder collection → registerphoto understandingcustomer's languagehandoff to a human

The simulator replaces only the network call to Meta: the same JSON that would go to graph.facebook.com is rendered here. Connect a real number and the code is the same.

🌿
Green Leaf Flowersbusiness account
customer's phone
Hi! Do you deliver on Sunday?I'd like to order a bouquet for my momПривет! Сколько стоит букет из пионов?Can I talk to a person?
👩‍💼
OwnerGreen Leaf Flowers · WhatsApp

Buttons under a draft are real WhatsApp reply buttons. Typing text while a draft is pending sends your text instead.

What happens under the hood: customer message → webhook → the engine stores it, calls the model with the knowledge base and the conversation, gets strict JSON (reply, action, collected order fields) → a draft with reply buttons is sent to the owner; only the approved text reaches the customer. A complete order becomes a row in the register (SQLite, optional Google Sheet) and a notification with Confirm / Cancel buttons. «Can I talk to a person?» pauses the bot for that chat and forwards everything to the owner until /bot.