aiogram bot hosting: how to deploy an aiogram 3 bot
An aiogram 3 bot needs one process that never stops, the token in an environment variable, state kept outside memory and a clean exit on SIGTERM. On Livops you push a repository with bot.py and requirements.txt, and it runs for 19 credits a month (19,000 so'm), polling or webhook.
By Livops team, the people who build and run LivopsUpdated
aiogram is the asyncio framework most Python Telegram bots are written with, and an aiogram 3 bot is a long-running program: it either asks Telegram for updates in a loop or listens for Telegram’s HTTPS requests. Hosting it well comes down to four things — one process that stays up, the token kept out of the code, state that survives a restart, and an exit that does not drop a reply halfway. This page covers each for aiogram 3.32, the current release, and then the exact files Livops reads to run one.
What does an aiogram bot need in production?
One process, polling or webhook — never both
Telegram has two ways of delivering updates and they are mutually exclusive[6]; aiogram says the same: with a webhook set, long polling cannot run[2]. Pick one per bot. With polling, getUpdates stops working while a webhook is set[6], and aiogram’s start_polling does not remove a webhook for you[8]: a bot moved from a webhook host has to call deleteWebhook once, or it receives nothing.
The right aiogram and the right Python
aiogram 3.32 was released on 15 November 2026 and needs Python 3.11 or newer, below 3.16[5]. aiogram 2 and 3 share a package name and little else: code that imports executor is aiogram 2 and will not start on 3. Pin the major version you wrote for:
aiogram>=3.32,<4
The token from the environment
The token BotFather gave you is the bot. Read it from an environment variable and keep it out of the repository, .env files included.
A clean exit on SIGTERM
A host stops a process with SIGTERM. start_polling and run_polling handle SIGINT and SIGTERM themselves while handle_signals is True, the default[8]. By default the dispatcher does not wait for handlers still running; pass tasks_shutdown_timeout and it stops fetching, waits that long for them, then cancels the rest[1]. Keep the timeout shorter than the host’s own stop window.
State that survives a restart
aiogram’s default FSM storage, MemoryStorage, keeps everything in a dict and loses it on every restart; the docs advise against it in production[3]. RedisStorage.from_url() takes a Redis URL and needs the redis package installed[3].
A production aiogram bot, polling
import asyncio
import os
from aiogram import Bot, Dispatcher
from aiogram.filters import CommandStart
from aiogram.fsm.storage.redis import RedisStorage
from aiogram.types import Message
dp = Dispatcher(storage=RedisStorage.from_url(os.environ["REDIS_URL"]))
@dp.message(CommandStart())
async def start(message: Message) -> None:
await message.answer("Hello!")
async def main() -> None:
bot = Bot(token=os.environ["BOT_TOKEN"])
# Wait up to 5 s for running handlers when SIGTERM arrives.
await dp.start_polling(bot, tasks_shutdown_timeout=5)
if __name__ == "__main__":
asyncio.run(main())
requirements.txt:
aiogram>=3.32,<4
redis
The same bot on a webhook
For a webhook, aiogram ships an aiohttp handler. SimpleRequestHandler compares the X-Telegram-Bot-Api-Secret-Token header with secret_token and answers 401 when they differ[4] — but only when a secret is set, so always set one. It answers Telegram at once and runs the handler in the background by default (handle_in_background)[2].
import os
from aiohttp import web
from aiogram import Bot, Dispatcher
from aiogram.webhook.aiohttp_server import SimpleRequestHandler, setup_application
bot = Bot(token=os.environ["BOT_TOKEN"])
dp = Dispatcher()
# ... handlers as above ...
app = web.Application()
SimpleRequestHandler(dp, bot, secret_token=os.environ["WEBHOOK_SECRET"]).register(
app, path=os.environ["WEBHOOK_PATH"]
)
setup_application(app, dp, bot=bot)
web.run_app(app, host="0.0.0.0", port=int(os.environ["PORT"]), shutdown_timeout=20)
Two differences from aiogram’s own example. Its server binds 127.0.0.1, because it expects a reverse proxy on the same machine[2]; in a container the platform reaches, bind 0.0.0.0. And its on_startup calls bot.set_webhook()[2]; leave that out on a platform that sets the webhook for you. aiohttp’s run_app waits up to shutdown_timeout, 60 seconds by default[7], which is longer than most hosts wait — hence 20 above.
Deploy an aiogram bot on Livops
- Push a repository with
bot.pyandrequirements.txt, or upload the folder as a.zip. In the cabinet, create a project from it. There is also a ready aiogram 3 bot + Redis template. - Under Settings → Variables, add
BOT_TOKENwith the token and mark it as a secret: its value is never shown again. - For FSM state, add a Redis database under Databases;
REDIS_URLis written into the bot’s variables by itself. - Deploy. The log shows the install and the start.
Livops sees aiogram in requirements.txt (or in your imports, when there is no requirements file) and makes the project a Telegram bot (Python). It installs into a virtualenv and starts bot.py; with no bot.py it looks for main.py, app.py, run.py, src/<package>/__main__.py or a file with if __name__ == "__main__". To set it yourself, put livops.json in the repository root:
{
"kind": "bot",
"start": ".venv/bin/python bot.py"
}
A Procfile with worker: .venv/bin/python bot.py does the same. A .python-version file picks the Python version.
Polling is the default and needs nothing else: no port, no domain. For a webhook, open the project’s Settings → Connection to Telegram, choose Webhook and press Save and redeploy. If your code already uses SimpleRequestHandler, the new-project wizard picks webhook mode on its own. The platform generates WEBHOOK_SECRET, tells Telegram the address with that secret once the bot answers, and refuses requests to the path that do not carry it. Your code listens on PORT, receives on WEBHOOK_PATH (/webhook unless you change it) and checks WEBHOOK_SECRET — exactly the webhook example above. In webhook mode the token must be in BOT_TOKEN.
What happens after that:
- On a new push, a polling bot is stopped first — SIGTERM, then ten seconds — and the new version starts, so two copies never poll with one token. A webhook bot’s new copy must answer on
PORTwithin a minute; it takes the updates before the old one is stopped. - If the bot exits in its first 30 seconds, the deploy is marked failed and the log shows its last lines. If the build fails, the running version is left alone.
- If it crashes later, it is started again. Ten crashes inside 15 minutes and it is stopped, and you are notified with the last lines it printed — in Telegram, if your account is linked.
- Files written to
/app/data(theDATA_DIRvariable) survive restarts and deploys;/tmpand the code folder do not. Anaiosqlitefile belongs there. For orders or payments, use PostgreSQL, which is backed up. - The clock is UTC. For schedules in Tashkent time, add
TZ=Asia/Tashkentunder Variables.
A bot costs 19 credits a month (19,000 so’m), polling or webhook. One credit is 1,000 so’m; see Pricing.
Livops or a VPS for an aiogram bot?
A VPS gives you root and room for several bots on one machine, which is the cheaper path once you have many. In exchange you install Python, write a systemd unit that restarts the bot, deploy by hand, and keep the system updated. On Livops a git push is the deploy and restarts are the platform’s job, but you get no root: a bot that needs system packages of its own is better on a VPS or in its own Dockerfile.
Questions
Does Livops run aiogram 2 bots?
Yes, as long as the version is pinned. If requirements.txt says aiogram with no version and the code imports executor, Livops installs aiogram below 3 so the old code still starts. New bots should be written for aiogram 3.
Which file does Livops start?
bot.py by default. If there is none, it looks for main.py, app.py, run.py, src/<package>/__main__.py or a file with if __name__ == "__main__". You can name the command yourself in livops.json or in Settings.
Polling or webhook for an aiogram bot?
Polling unless you have a reason. It needs no address and no port. If your code uses SimpleRequestHandler, the wizard picks webhook mode for you, and the platform tells Telegram the address; your code must not call set_webhook.
Where should an aiogram bot keep FSM state?
Not in MemoryStorage, which is lost on every restart and deploy. Add a Redis database to the project and pass REDIS_URL to RedisStorage.from_url. A SQLite file belongs in /app/data, the DATA_DIR variable.
What happens when I push a new version?
A polling bot is stopped first, with SIGTERM and ten seconds to finish, and the new one starts after it, so two copies never poll at once. Messages sent in between wait at Telegram. A webhook bot's new copy starts and takes updates before the old one is stopped.
How much does it cost?
19 credits a month, 19,000 so'm at 1,000 so'm a credit, the same for polling and webhook. There is no free tier for bots; the first bot on an account that has not paid runs for 15 minutes on a zero balance.
Sources
- aiogram — Long-polling (docs.aiogram.dev). Checked .
- aiogram — Webhook (docs.aiogram.dev). Checked .
- aiogram — FSM storages (docs.aiogram.dev). Checked .
- aiogram source — aiogram/webhook/aiohttp_server.py (raw.githubusercontent.com). Checked .
- aiogram on PyPI (pypi.org). Checked .
- Telegram Bot API — Getting updates (core.telegram.org). Checked .
- aiohttp — Server reference (run_app) (docs.aiohttp.org). Checked .
- aiogram source — aiogram/dispatcher/dispatcher.py (raw.githubusercontent.com). Checked .
Put your project online today
Connect GitHub or drop a folder. No card is asked for.
Deploy your aiogram botRelated
- Telegram bot webhook vs polling: which one to usePolling means your bot asks Telegram for updates in a loop; a webhook means Telegram POSTs each update to your HTTPS address. Polling is simpler and needs no address; a webhook needs HTTPS but no always-running loop. A bot uses one or the other, never both. On Livops both run for the same 19 credits a month.
- python-telegram-bot hosting: deploying a v22 botA python-telegram-bot bot runs as one Application: run_polling() for polling, run_webhook() with the webhooks extra for a webhook. Both stop cleanly on SIGTERM. On Livops it runs from a push with bot.py and requirements.txt for 19 credits a month (19,000 so'm).
- Telegraf bot hosting: deploying a Telegraf 4 bot on Node.jsA Telegraf bot is one Node.js process: bot.launch() for polling, or bot.webhookCallback() behind an HTTP server for a webhook, with bot.stop() on SIGTERM. On Livops a repository with telegraf in package.json is run as a bot from every push, for 19 credits a month (19,000 so'm).
- grammY bot deploy: polling, webhooks and hostingDeploy a grammY bot as one always-on Node.js process: bot.start() for polling, or webhookCallback() behind an HTTP server, with bot.stop() on SIGTERM. On Livops a repository with grammy in package.json runs as a bot from every push, for 19 credits a month (19,000 so'm).