The bot answered messages briskly all evening. In the morning the laptop was closed, its owner got on the metro — and the bot fell silent. Users keep writing, the bot does not react, and by evening some of their questions will never be seen.

This is not a Telegram failure. A bot is a program that runs on someone's computer, and Telegram merely passes messages to it. While the program runs, the bot is alive. So "running a bot 24/7" means answering three questions: where the process lives, who brings it back after a crash and where the token is kept.

24 hours
How long Telegram keeps updates the bot has not yet received.
4 ports
Supported for webhooks: 443, 80, 88 and 8443.
1 instance
How many copies of a bot should receive updates on one token.

How it works

Why the bot falls asleep

Telegram does not run bot code. It accepts users' messages and holds them for your program, and the program collects them and replies. If the program is not running, messages pile up on Telegram's side, but for no longer than 24 hours. Whatever the bot has not collected within a day is lost.

Hence a simple rule: a bot needs a computer that never switches off, and a program that restarts itself after a crash.

How a bot gets messages

Polling or webhook


Long polling (getUpdates)

Webhook (setWebhook)

Who starts

The bot asks Telegram whether there are new messages

Telegram sends messages to your address

What you need

Any computer with internet access

A public HTTPS address on port 443, 80, 88 or 8443

Pros

Simple: no domain or certificate needed

No need to hold a constant connection; suits platforms that wake code on request

Cons

The process must run all the time

Needs HTTPS, a domain and a check that the request came from Telegram

The two methods are mutually exclusive: while a bot has a webhook set, getUpdates does not work. To go back to polling, the webhook is removed with deleteWebhook. The bot from "How to create a Telegram bot: BotFather, your first Python code, and one rule about the token" uses long polling — that is the one we will move to a server.

Top: the bot requests messages. Bottom: Telegram sends them to the bot server.
Read from top to bottom. Bot on the left, Telegram on the right. The bot initiates polling; Telegram initiates a webhook. These are alternatives.

Options

Where to host it

Option

Suitable mode

Pros

Cons

Your own computer

Polling

Free and simple

Works only while the computer is on and not asleep

A virtual server (VPS)

Polling or webhook

Runs round the clock, full control

You have to set up the system, updates and autostart

An app platform or serverless functions

Webhook

No server to administer

The platform's own limits; HTTPS required

For a first bot with long polling, a small Linux virtual server is convenient. Next: how to make it run there without you.

Step by step on a VPS

A systemd service: automatic restart

  1. Copy bot.py to the server, for example to /opt/bot, and run it as a separate user without administrator rights.
  2. Put the token in a file /etc/bot/token.env as a single line TELEGRAM_BOT_TOKEN=your_token, and lock it down with chmod 600: only its owner can read it.
  3. Describe the service in the file /etc/systemd/system/bot.service.
[Unit]
Description=Telegram bot
After=network-online.target
Wants=network-online.target

[Service]
User=botuser
WorkingDirectory=/opt/bot
EnvironmentFile=/etc/bot/token.env
ExecStart=/usr/bin/python3 /opt/bot/bot.py
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target
  1. Enable the service and watch the log.
sudo systemctl daemon-reload
sudo systemctl enable --now bot
journalctl -u bot -f

Restart=always restarts the bot after any stop, RestartSec=5 adds a five-second pause between attempts, and enable starts the service at every server boot. The token never ends up in the code or in your command history.

The second mode

Webhook: when you need an address

With a webhook, your server or platform receives Telegram's requests over HTTPS. When setting it up with setWebhook, you pass the address and a secret. Every Telegram request then arrives with the X-Telegram-Bot-Api-Secret-Token header, and anything that comes without it is discarded.

import hmac
import os

WEBHOOK_SECRET = os.environ["WEBHOOK_SECRET"]


def setup_webhook(call, url):
    # call is the Bot API helper from the bot-creation article
    call("setWebhook", url=url, secret_token=WEBHOOK_SECRET, allowed_updates=["message"])


def is_from_telegram(headers):
    received = headers.get("X-Telegram-Bot-Api-Secret-Token", "")
    return hmac.compare_digest(received, WEBHOOK_SECRET)

Keep the secret the same way as the token: in an environment variable, not in the code.

Security

The token: the key to the bot

The token from BotFather is full access to the bot: with it, anyone can control the bot, receive what people send it and reply in its name. So it lives only on the server, in a file with restricted permissions, and never goes into a repository, logs or screenshots.

If the token does leak, open BotFather and issue a new one with the /token command, then replace it in the file on the server and restart the service. What to do after a leak is covered in detail in the "If the token leaks" section of "How to create a Telegram bot: BotFather, your first Python code, and one rule about the token".

Checklist

Before launch

Questions

Can I keep the bot on my home computer?

You can, if the computer is on round the clock and never sleeps. But any power or internet outage stops the bot, and messages older than 24 hours are lost. For constant operation a server is more reliable.

What happens to messages while the bot is down?

Telegram keeps them for up to 24 hours. When the bot starts again, it receives the accumulated updates and replies to them. Anything older than a day the bot will never see.

Why do two copies of the bot get in each other's way?

With long polling, only one program can collect updates. If you start a second copy with the same token, they begin to conflict, and some messages reach the wrong copy. For testing, create a separate bot in BotFather with its own token.


A bot works for as long as its process runs — that is all there is to know about hosting. The rest is choosing a computer that never sleeps and teaching it to bring the bot back after any crash.

A systemd service with automatic restart and the token in a locked file solve both on an ordinary virtual server. A webhook will come in handy later, when the bot moves to a platform that wakes code on request.