AUTOMATION

Claude on Telegram: Build an AI Agent That Knows Your Business

June 17, 2026 7 min read

Claude AI Telegram Bot for project management and automation

There are many ways to work with AI. You can open a browser, enter the interface, write a prompt, wait for a response. It works. But there’s another way to think about AI automation in business — one that feels more like sending a message to a trusted person who knows your business from the inside, and less like talking to an advanced search engine.

In this article, I want to show you how to build a Telegram bot connected to Claude — not just any generic Claude, but Claude oriented to your specific project, with all the context, files, and settings you’ve prepared in advance.

Wait — What Exactly is an ‘Agent’?

The word “agent” sounds technical and intimidating. But the idea itself is very simple.

An agent is simply a project with a dedicated folder on your computer, and inside that folder sits a markdown file — CLAUDE.md — that explains to Claude what this project is, the working rules, and how to act.

A concrete example: if your CLAUDE.md explains to Claude how to create and publish posts on your Instagram — including the aesthetic, tone, which account, and at what time — then that’s your Instagram agent. No magic there. There’s a folder, there’s a settings file, and there’s Claude who knows how to read it.

Those who work with a Second Brain architecture can take this a step further: create a separate agent for each active project, plus a main agent connected to the parent folder. Each one knows only its own world, without confusion. This is AI automation that fits how you actually work.

Why Do You Need a Telegram Bot?

True, you can talk to Claude directly through his app. So what’s special here?

The difference is one thing: your files.

Regular Claude knows nothing about your projects, your clients, the decisions you’ve made. To ask him a question based on the real reality of your business — you need to sit at the computer, open the tools, provide the files manually.

The bot solves this. It sits on your computer and is directly connected to the project files. You’re at the beach, in a meeting, on the road — you send a message — and the bot brings a response based on your real information.

You (Telegram) → Python Bot → Claude CLI → Project Files

Why Telegram and Not WhatsApp?

1. Open and Free API — no cost, no approvals. WhatsApp has a business API that requires approval from Meta.

2. Minutes to Set Up — the bot is ready within 5 minutes from when you open @BotFather.

3. Voice Messages — speak → Claude listens. Whisper transcribes automatically. Very useful when your hands are busy.

Why One Bot for All Projects Isn’t Enough

With several active projects, Claude needs to figure out “which project is this about?” with every question. This creates ambiguity and less precise answers. The solution: a dedicated bot for each project.

• You have a business — “What’s David’s status?” ← the bot opens the file and brings the answer
• Content channel — “What did I plan for this month and what haven’t I done yet?” ← the bot checks the list
• Small business — “What did the client request at the last meeting?” ← the bot pulls the notes

The Advantages

Access from Anywhere — the phone is always in your pocket. An idea at the beach — sent in a second.

Voice Input — Whisper transcribes automatically. Speak — Claude answers.

Precise Context — the bot knows, it doesn’t guess. Opens the file and brings the information.

Conversation Memory — the last 20 messages are saved during runtime.

Direct Writing to Vault — “Save this”, “Add to tasks” — Claude actually edits the files. This is the difference between AI that helps for a moment and AI automation that improves processes over time.

The Disadvantages — Let’s Be Direct

Not real-time — 10–30 seconds for a response. A work tool, not a chat conversation.

Memory isn’t permanent — but there’s a solution — Stopped the bot? Conversation memory is deleted. But the bot can write to files forever. “Save this” — written to your files. The permanent memory is the files.

The computer needs to be on — for 24/7 availability you need a server (VPS, Mac mini, etc.).

How to Build It — Step by Step

What you need before you start: Python 3 · Claude Code CLI · Telegram account

Step 1 — Create a Telegram Bot

Open @BotFather in Telegram and create a new bot:

/newbot
→ Bot name: My Project Bot
→ Username: myproject_bot
→ Receive token: 1234567890:ABCdef...

Step 2 — Install the Tools

Python 3 — the programming language the bot is written in. On Mac — usually already installed. Check: python3 --version

Claude Code CLI — the Terminal version of Claude that can read and write files.

brew install python3
npm install -g @anthropic-ai/claude-code
pip3 install python-telegram-bot openai-whisper
brew install ffmpeg

Step 3 — The Code

💡 No need to copy and paste code manually. Just explain to Claude what you want — he’ll write the code. Only two lines need to be changed: the token and the project path.

import asyncio, subprocess, tempfile
from collections import defaultdict
from telegram import Update
from telegram.ext import Application, MessageHandler, filters, ContextTypes
import whisper

TOKEN      = "YOUR-TOKEN-HERE"        # <- change this
VAULT_PATH = "/your/project/folder"   # <- and this

conversation_history = defaultdict(list)
whisper_model = whisper.load_model("small")

async def get_claude_response(user_id, message):
    history = conversation_history[user_id]
    history_text = ""
    if history:
        history_text = "Conversation history:\n"
        for msg in history:
            role = "User" if msg["role"] == "user" else "Claude"
            history_text += f"{role}: {msg['content']}\n"
        history_text += "\n---\n"
    full_prompt = f"{history_text}User: {message}"
    loop = asyncio.get_event_loop()
    result = await loop.run_in_executor(None, lambda: subprocess.run(
        ["/opt/homebrew/bin/claude", "-p", full_prompt,
         "--dangerously-skip-permissions",
         "--allowedTools", "Read,Write,Edit,Glob,Grep,Bash(*)"],
        capture_output=True, text=True, cwd=VAULT_PATH, timeout=120
    ))
    reply = result.stdout.strip() or "Could not get a response."
    conversation_history[user_id].append({"role": "user", "content": message})
    conversation_history[user_id].append({"role": "assistant", "content": reply})
    if len(conversation_history[user_id]) > 20:
        conversation_history[user_id] = conversation_history[user_id][-20:]
    return reply

async def handle_text(update, context):
    await update.message.reply_text("Thinking...")
    response = await get_claude_response(update.effective_user.id, update.message.text)
    await update.message.reply_text(response)

async def handle_voice(update, context):
    await update.message.reply_text("Transcribing...")
    voice_file = await update.message.voice.get_file()
    with tempfile.TemporaryDirectory() as tmp:
        ogg, wav = f"{tmp}/voice.ogg", f"{tmp}/voice.wav"
        await voice_file.download_to_drive(ogg)
        subprocess.run(["ffmpeg", "-i", ogg, wav, "-y"], capture_output=True)
        text = whisper_model.transcribe(wav)["text"].strip()
    if not text:
        await update.message.reply_text("Could not transcribe.")
        return
    response = await get_claude_response(update.effective_user.id, text)
    await update.message.reply_text(response)

def main():
    app = Application.builder().token(TOKEN).build()
    app.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, handle_text))
    app.add_handler(MessageHandler(filters.VOICE, handle_voice))
    app.run_polling(drop_pending_updates=True)

if __name__ == "__main__":
    main()

Step 4 — Activation

python3 bot.py

The bot is running. Open Telegram, find the bot you created, and send the first message.

Step 5 — Additional Project? Additional Bot

For each additional project — create a new bot in @BotFather, save a new file, and change only two lines:

TOKEN      = "new-token"
VAULT_PATH = "path-to-other-project"

Run each bot in a separate Terminal window. Both bots run simultaneously, each connected to its own world.

Frequently Asked Questions

Does a Telegram bot cost money?

As of today — completely free. Telegram's API is open at no cost. Given growing popularity, a change may come in the future, but right now — there's no reason not to try.

Is it complicated to build?

Not at all. If you can run a command in Terminal and change two lines in a text file — you can build this bot. An hour at most the first time, and less than 10 minutes for each additional bot.

Do you need to know how to program?

No. Even if the code above looks foreign to you — just explain to Claude what you need, and he'll write it for you. The code is simple enough that Claude can prepare it without effort.

Can the bot help with content creation, not just information management?

Absolutely. If your CLAUDE.md defines the style, audience and platforms — the bot can write posts, edit texts, and prepare drafts directly to your files. Almost anything Claude knows how to do in the regular interface — he can do through the bot too, with the real context of your project.

Summary

What we built: a simple architecture that connects your phone to Claude who knows your projects in depth.

You don't need to be at a computer to work with your information. Every idea is saved. Every question gets an answer based on your real context. Every project gets the bot that understands it.

This isn't just another chatbot — it's the assistant that knows your business.