Restaurant ai agent customer support

Overview

This project is an AI-powered restaurant customer support and order management system built with n8n, Telegram, OpenRouter, Google Sheets, Supabase Vector Store, and OpenAI Embeddings.

The workflow receives customer messages through Telegram, analyzes the user’s intent, and routes each request to the appropriate specialist:

General Customer Support

Restaurant Questions Expert

Orders Management Expert

The AI Agent can answer restaurant and menu questions using a Supabase knowledge base, create new orders, retrieve existing orders, cancel orders, and maintain conversation context across multiple messages.

Key Features

Telegram-based customer support chatbot

AI-powered intent classification

Automatic routing to specialized prompts

Menu and restaurant information retrieval

Retrieval-Augmented Generation using Supabase

New order creation

Order status lookup

Order cancellation

Unique 4-digit order ID generation

Google Sheets order database

Conversation memory per Telegram user

Structured LLM output for reliable routing

Automated customer responses

Workflow Architecture

Telegram Customer Message

↓

Basic LLM Intent Classifier

↓

Structured Output Parser

↓

Switch Router

┌──────┼─────────┐

↓ ↓ ↓

General Orders Questions

Prompt Prompt Prompt

└──────┼─────────┘

↓

AI Agent

┌──────┼──────────────────────────────┐

↓ ↓ ↓ ↓ ↓

Memory Read Order Add Order Cancel Restaurant

Order Order Knowledge Base

↓

Telegram Response

Intent Classification

The first language model analyzes each incoming message and assigns it to one of three categories.

Generalist

Handles general messages such as:

Hello

Good morning

Thank you

General conversation

Questions Expert

Handles questions related to:

Menu items

Prices

Ingredients

Restaurant information

Opening hours

Delivery

Restaurant policies

Orders Expert

Handles requests related to:

Creating a new order

Providing customer details

Asking about an existing order

Checking order status

Cancelling an order

Providing an order ID

Selecting specific meals or items

Order Management Capabilities

Create an Order

The AI collects the required customer information:

Customer name

Phone number

Delivery address

Selected items

Order price

Before saving the order, the assistant confirms the details with the customer.

A code tool generates a unique four-digit order ID, then the order is added to Google Sheets with the following fields:

TIME

ITEM

PRICE

ORDER ID

NAME

PHONE NUMBER

ADDRESS

STATUS

New orders are automatically assigned the status:

kitchen

After successfully saving the order, the customer receives the generated order ID.

Retrieve an Order

The assistant asks for or extracts the customer’s order ID from the conversation.

The Google Sheets read tool searches the ORDER ID column and returns the matching order information.

The assistant can then answer questions about:

Ordered items

Customer details

Order status

Existing order information

Cancel an Order

The assistant confirms the cancellation request and collects the order ID.

The Google Sheets cancellation tool searches for the matching order and updates its status to:

cancel

The assistant then confirms the cancellation to the customer.

Restaurant Knowledge Base

The project uses Supabase Vector Store as a restaurant knowledge base.

Restaurant information is converted into embeddings using OpenAI Embeddings and stored in Supabase. The AI Agent searches this knowledge base before responding to restaurant-related questions.

The knowledge base can contain:

Menu items

Prices

Ingredients

Restaurant opening hours

Delivery areas

Payment methods

Reservations

Current offers

Restaurant policies

Allergy information

Frequently asked questions

This Retrieval-Augmented Generation setup helps reduce hallucinations and keeps responses grounded in restaurant data.

Conversation Memory

The workflow uses Simple Memory with the Telegram chat ID as the session key:

Telegram Chat ID → Unique Conversation Session

This enables the assistant to remember previous customer messages, including:

Selected meals

Customer name

Phone number

Address

Order ID

Previous questions

Ongoing order details

The customer does not need to repeat information during the same conversation.

Technologies Used

n8n

Telegram Bot API

OpenRouter API

Large Language Models

Structured Output Parser

AI Agent

Google Sheets API

Supabase

Supabase Vector Store

OpenAI Embeddings

Retrieval-Augmented Generation

JavaScript Code Tool

Conversation Memory

Workflow Automation

Main Nodes

Telegram Trigger

Receives customer messages from Telegram.

Basic LLM Chain

Classifies customer intent and selects the appropriate specialist.

Structured Output Parser

Forces the classifier to return structured routing data.

Switch Node

Routes the message to the correct system prompt.

AI Agent

Handles customer conversations and decides which connected tool to use.

Simple Memory

Maintains conversation context for each Telegram customer.

Orders Sheet Read Tool

Retrieves orders using the customer’s order ID.

Orders Sheet Write Tool

Adds newly confirmed orders to Google Sheets.

Order Cancellation Tool

Updates the order status to cancelled.

Code Tool

Generates a random four-digit order ID.

Supabase Vector Store

Retrieves restaurant and menu information from the knowledge base.

Telegram Send Message

Sends the final AI response back to the customer.

Design Goals

The project was designed around the following principles:

Fast customer support

Natural conversational interaction

Accurate restaurant information

Automated order processing

Reduced manual work

Organized order records

Reliable intent routing

Context-aware conversations

Low hallucination through RAG

Easy workflow scalability

Example Use Cases

Restaurant Question

Customer:

How much is the Classic Beef Burger?

The assistant searches the Supabase knowledge base and returns the stored menu price.

New Order

Customer:

I want two chicken burgers.

The assistant asks for the missing customer details, confirms the complete order, generates an order ID, and saves the order to Google Sheets.

Order Status

Customer:

What happened to order 4821?

The assistant retrieves the matching order from Google Sheets and returns its current status.

Cancellation

Customer:

Please cancel order 4821.

The assistant confirms the request and changes the order status to cancel.

Future Improvements

WhatsApp Business integration

Payment gateway integration

Restaurant website chatbot

Automated delivery tracking

Kitchen notification system

Customer authentication

Order modification support

Multi-language conversations

Voice ordering

Table reservations

CRM integration

Customer feedback collection

Human agent handoff

Analytics dashboard

Multi-branch restaurant support

Project Goal

The objective of this project is to automate restaurant customer service and order management through a single intelligent conversational assistant.

By combining intent classification, AI Agents, conversation memory, Google Sheets tools, and a Supabase knowledge base, the system can answer customer questions, create and track orders, cancel existing orders, and provide a consistent support experience without requiring constant human intervention.