Guides
Step-by-stepApril 8, 202612 min

AI Manager in Telegram: A Bot That Sells 24/7

This guide is based on a real project. The AI manager is running in production, serving customers and closing sales without human involvement.

The Problem

You have an online store or service. Clients message you on Telegram at any time. You can't respond 24/7. Hiring a manager is expensive. A regular button-bot annoys customers and doesn't close deals.

Result: lost orders at night and on weekends, slow responses, customers leave for competitors.

Solution: AI Manager

An AI manager is not a button-bot. It's a full-fledged salesperson that understands natural language, knows your entire catalog, and can carry on a conversation like a real person.

Understands any question: The client writes however they want — the bot understands context and responds on point
Knows the full catalog: 1000+ products, prices, availability, specs — all in the bot's memory
Carries conversations: Doesn't just answer — asks clarifying questions, suggests alternatives
Follows rules: SOUL file sets tone, boundaries, and communication style — the bot stays on track

Architecture (how it works)

Client → Telegram
        |
   Telegraf Bot (Node.js)
        |
   Express Server (API)
        |
   OpenAI GPT-4o-mini
   + SOUL.md
   + Product DB (Supabase)
        |
   Response → Client

5 Setup Steps

1

Create a SOUL file

SOUL.md is the instruction file for the AI. It defines who it is, how it communicates, and what it must NOT do.

2

Upload your product catalog

The bot needs to know what you sell. Load products into a database (Supabase, PostgreSQL, or even a JSON file to start).

Minimum fields: name, price, description, availability. The bot searches these when responding to clients.

3

Connect the Telegram bot

Create a bot via @BotFather, get the token. Telegraf (Node.js) is the simplest way to handle messages.

Each incoming message goes to the API → AI generates a response using SOUL + catalog → response goes to the client.

4

Set up context and memory

The AI needs to remember conversation history. Store the last 10-15 messages per chat and pass them as context with each request.

This lets the bot carry coherent conversations: "What size?" → the bot remembers which product they're discussing.

5

Launch and monitor

PM2 for production — auto-restart on crashes. Log all conversations for analysis and SOUL file improvement.

Read logs the first week and adjust instructions. By week's end, the bot will respond better than any human manager.

Results (real numbers)

MetricWithout AIWith AI Manager
Response time30 min — 12 hours< 3 seconds
Availability10-12 h/day24/7/365
Catalog knowledge~70% (humans forget)100% (database)
Cost/month€500-1500 (manager)~€15 (API + server)
Scalability1 chat at a timeUnlimited

Tech Stack

Node.jsTelegrafExpressOpenAI APISupabasePM2VPS

Important

An AI manager doesn't replace sales strategy. It replaces the routine of answering. If you have a bad product or no traffic — a bot won't help. Business first, then automation.

Want one like this?

We set up AI managers turnkey: from SOUL file to production launch. In 2 weeks you'll have a bot that knows your business better than any employee.

Discuss on TelegramView services