Technology
A Restaurant Telegram Bot: Catalog, AI Orders and Kitchen Alerts
In Uzbekistan a guest will not install your app, but they open Telegram fifty times a day. That makes a bot a cheaper and faster channel than a mobile app. This article covers a bot's three jobs: catalog, ordering and staff alerts.
01A bot's three jobs — keep them apart
The first job is the catalog: the guest browses and orders. The second is the staff alert: a new order, a waiter call, a bill request, a payment. The third is AI conversation: the guest writes in plain language, the bot understands and places the order.
Cramming all three into one bot is the most common mistake. Staff messages get tangled with guest conversation and the guest cannot find the menu. The right split: a catalog bot of its own, staff notifications through a group.
In laCafe each organisation creates unlimited bots. That matters for multi-brand groups: every brand gets its own bot, logo and menu, while the orders all land in one admin panel.
02The catalog bot: a menu inside a Mini App
Old-style bots showed the menu as buttons: "Salads" → "Achik-chuchuk" → "Add to cart". Ten dishes meant fifteen taps. The Mini App closed that gap: a full site opens inside the bot, with photos, categories, a cart and modifiers.
For the guest the difference is no install and no sign-up: the phone number is shared in one tap. For the restaurant the difference is that the menu comes from one place, so prices and the stop-list match the website exactly.
03Staff notifications: routing by hall
A working restaurant drops hundreds of messages a day into a Telegram group. In one stream nobody reads them. The answer is routing: decide in advance which message goes to which group, or which forum topic.
A workable scheme: the "Outdoor hall" topic gets that hall's calls, "Kitchen" gets new orders, "Till" gets bill requests and payments. The message language is configurable too: uz, ru or en.
Telegram is also valuable here as a fallback. The waiter has the laWaiter app; the hookah attendant or the prep cook does not, but everyone has Telegram. It keeps the whole team informed without buying devices.
- New order → the "Kitchen" topic
- Waiter call → that hall's topic plus a laWaiter push
- Bill request and online payment → the "Till" topic
- Cancellation and complaint → the owner's private chat
04The AI bot: "two plovs, not spicy"
The AI bot works on different logic. The guest doesn't walk the menu, they just write: "Two plovs, one not spicy, deliver to Chilanzar". The bot, running on Anthropic's Claude, understands it, builds the cart, asks for the address and passes the order to the kitchen.
When is this channel useful? For repeat orders and for guests used to ordering by phone. They know the menu by heart; they don't want to browse, they want to say it. This is where operator call time is saved.
The AI agent starts at the Standart package (2 400 000 UZS a month, 5 branches, 3 000 orders). The reason is clear: every conversation is a call to an AI model. On a smaller package, it makes sense to plan the AI bot as a next step rather than the main channel.
A plain conversational bot (without AI) and an Instagram Direct AI assistant are still planned — we say "coming soon" because they do not work yet. What does work: the catalog bot, the AI bot and notifications.
05Getting the bot live in an hour
The steps are simple: create a bot with @BotFather, take the token, paste it into the admin panel. Then pick the bot's purpose: catalog, notifications or AI. The webhook connects itself; there is nothing to configure.
The notification bot needs one extra step: add it to the group, make it an admin, then map halls to topics in the admin panel. That mapping is what actually takes the time — 15 to 20 minutes.