Restaurant AI Agent — AI skill for Claude Code

Restaurant-Ai-Agent

Building an AI-Powered Restaurant Ordering Assistant with n8n — A Case Study The Problem

Small restaurants often rely on manual order-taking: a customer calls or messages, someone jots down the order, walks it to the kitchen, and later updates the inventory by hand. Every step is a chance for a mistake — a missed order, a stock count that's never updated, a customer who has to repeat their order because nobody wrote down their name or phone number correctly.

I wanted to see how far I could take this with pure automation: a customer messages a Telegram bot, the bot understands what they want in natural language, confirms the order, logs it, and keeps inventory in sync — with zero manual data entry.

The Solution

I built a multi-stage automation system in n8n, connected to Google Sheets and Telegram, with an AI agent at its core. The final piece — and the focus of this report — is the ordering assistant itself: a conversational AI that talks to customers, understands their order in plain language (in whatever language they write in), collects what's needed to complete it, and writes it straight into the restaurant's order log.

Technical Architecture

1. Telegram Trigger The flow starts the moment a customer sends a message to the restaurant's bot.

2. AI Agent This is the core of the system. It's given a system prompt containing the full menu (items, prices, ingredients) and a set of behavioral rules: reply in the customer's language, never assume a choice the customer hasn't confirmed, ask clarifying questions instead of guessing, and only mark an order as "chosen" once both a specific item and the customer's name and phone number have been provided.

The Agent is wired to three sub-components:

Model — an OpenRouter-connected chat model that powers the actual reasoning and replies. Mem