How a restaurant would answer every table over a Telegram QR code

The starting point

In this scenario the dining room is full and three tables are trying to catch the same server's eye — one asking whether a dish is gluten-free, one wanting the wine that pairs with it, one just chasing their order. Every question is reasonable and every one of them pulls a server off the floor. The allergen questions are the tense ones: the honest answer is "let me go check the kitchen," and that round trip takes minutes the guest spends worrying.

What gets deployed

• A table agent reachable from a Telegram QR code printed on the table tent — the guest scans, Telegram opens, and they are talking to the restaurant. Nothing to install; almost everyone already has the app.
• The restaurant's menu, allergen matrix, and prep notes as the knowledge base, so the answer to "does the laksa contain shellfish" comes from the kitchen's own document, not a guess.
• Answers in the guest's language, using the platform's built-in languages — the same QR code serves a local and a tourist at neighbouring tables without a second menu.
• Remembered regulars: because each guest arrives with their Telegram account, the agent knows the person who always asks for the no-coriander version, and greets them as themselves rather than as a stranger every visit.
• A hand-off to staff for the things a bot should not decide — a substitution request, a complaint, "is chef in tonight" — filed with the table number so a server walks over already knowing why.

What changes

Because it never quotes an allergen answer it cannot find in the kitchen's own document — and hands the guest to a person the moment the stakes call for one — the risky questions stay safe, which is the only way a table would trust it with them.