AboBot

Русский · English

A Telegram bot for group chats. It:

• pings a member when their name comes up in a message or a voice note;
• keeps statistics for the chat and every member;
• fixes text typed in the wrong keyboard layout;
• transcribes and voices messages;
• runs playful events.

The bot speaks Russian.

Status: personal project, running since 2021 · bot @chat_abobot · over 20,000 users from 45 chats in the bot's database (October 2026)

Stack: Python 3.10+ · aiogram 3 · PostgreSQL (asyncpg) · pymorphy3 · pyenchant · SpeechRecognition · gTTS · Docker

Features

• Name mentions. Words of a message are reduced to their base form (pymorphy3) and compared with members' names, so any grammatical case of a name mentions the right person. A custom name can be set with /edit. Names are also found in voice notes up to a minute long.

• /all mentions every member. A long list is split into several messages.

• Big chats (over 100 members).

  • /all is for admins only, at most once in 5 minutes, and only for people who wrote in the last 30 days.
  • The same person is mentioned by name at most once a minute.

• All-time and monthly statistics (/stat_group, /stat_user): messages, replies, commands, links, media, stickers, voice and video notes. Months are kept, so /month 2026-09 shows any month's summary and /top shows the most active members for the month or all time (/top all).

• /settings — per-chat switches: name mentions, names in voice notes (audio goes to Google), text events, deleting service messages. Admins can change them.

• Wrong keyboard layout. A message like ghbdtn is repeated as «привет». English words are recognized with a dictionary (pyenchant) and left alone.

• Voice. /recognize as a reply to a voice note transcribes it; «Озвучь - текст» turns text into a voice note.

• Events:

  • random numbers;
  • reversing text;
  • a language game;
  • mock fights;
  • picture reactions.

  Turned off with /stop_bot or in /settings.

• Deleting service messages (joins, leaves, title and photo changes) when the bot is an admin.

• Tracking the bot in groups. The bot records which groups it is in and with which rights: Telegram events plus a check on start and every 6 hours. Changes are kept as history. If the bot is removed, the group's statistics stay; only the mark that the bot is no longer there changes.

Running

git clone https://github.com/EDeev/chatping_abobot.git && cd chatping_abobot
cp .env.example .env      # BOT_TOKEN from @BotFather
docker compose up -d      # the bot and PostgreSQL

Prebuilt image: docker pull ghcr.io/edeev/chatping_abobot or docker pull git.deev.su/edeev/chatping_abobot. Tables are created on first start (code/schema.sql).

Without Docker you need:

• Python 3.10+;
• PostgreSQL;
• the enchant system library with an English dictionary (apt install libenchant-2-2 hunspell-en-us).

Then: pip install -r requirements.txt and cd code && BOT_TOKEN=… DATABASE_URL=postgresql://… python bot.py.

Migrating data from the old version (four SQLite databases): python scripts/migrate_sqlite.py --sqlite-dir path/to/db --dsn postgresql://…. The script checks the totals after the transfer.

Deployment

@chat_abobot runs on a home server as a systemd service: its own venv, settings in /etc/abobot.env, PostgreSQL on the same server. The nightly backup dumps the database, and the statistics feed a Metabase dashboard.

How it works

code/bot.py          entry point, error reports to a tech chat, checking the bot's status in groups
code/handlers/       routers: help, settings, stats, mentions, voice, events, chat (bot in groups, service messages)
code/db.py           PostgreSQL queries
code/schema.sql      schema: chats, users, members, chat_stats and member_stats by period, bot_status_history
code/nlp.py          base forms of names, keyboard layout, reversal, escaping
scripts/             migration from SQLite
data/                event pictures and the greeting

All-time and current-month counters grow in a single query (INSERT … ON CONFLICT DO UPDATE), so simultaneous messages are not lost. Database calls are asynchronous and speech recognition and synthesis (Google Web Speech and gTTS) run in a separate thread, so the bot never freezes. Messages use HTML markup and user text is escaped.

Development

pip install -r requirements-dev.txt
ruff check --select E9,F,B code tests && pytest

The tests need PostgreSQL (TEST_DATABASE_URL). What they cover:

• counters by period;
• mentions and big-chat limits;
• splitting /all;
• the bot's status in groups and its history;
• migration from SQLite;
• keyboard layout and escaping.

CI runs them on Python 3.10 and 3.12.

The Docker image is built on v* tags and published to GitHub Packages and git.deev.su.

License

MIT — see LICENSE.

Author

Egor Deev — GitHub · Telegram · egor@deev.space

---

_{⭐ If you find this project useful, give it a star on GitHub!}

_{Made with ❤️ — deev.space}