(This is a copy of the readme file from the repository.)

Teleward is a Telegram captcha bot written in Clojure and compiled with GraalVM native image. Runs on bare Linux/MacOS with no requirements. Fast and robust.


Telegram chats suffer from spammers who are pretty smart nowadays. They don’t use bots; instead, they register ordinary accounts and automate them with Selenium + web-version of Telegram. Personally, I found Shieldy and other bots useless when dealing with such kind of spammers. This project aims the goal to finish that mess.

Another reason I opened Teleward for is to try my skills in developing Clojure applications with GraalVM. Binary applications are nice: they are fast, and they don’t need installing JDK. At the same time, they’re are still Clojure: REPL is here, and that’s amazing.


  • This is Clojure, so you have REPL! During development, you call Telegram API directly from REPL and see what’s going on.
  • The bot can be delivered either as a Jar file or a binary file (with Graal).
  • When Graal-compiled, needs no requirements (Java SDK, etc). The binary size is about 30 Mb.
  • At the moment, supports only long polling strategy to obtain messages. The webhook is to be done soon.
  • Keeps all the state in memory and thus doesn’t need any kind of a database. The only exception is the current offset value which is tracked in a file.
  • Supports English and Russian languages.
  • Two captcha styles: normal “1 + 2” and Lisp captcha “(+ 1 2)”.
  • The +, -, and * operators are corresponding Unicode characters that prevent captcha from naive evaluation.


The bot listens for all the messages in a group. Once a new pack of messages arrives, the bot applies the following procedure to each message:

  • Mark new members as locked.
  • Send a captcha message to all members.
  • Unless an author of a message is locked, delete that message.
  • If a message is short and matches the captcha’s solution, unlock a user and delete the catpcha message.
  • If a locked user has posted three messages with no solution, ban them.
  • If a locked user hasn’t solved captcha in time, ban them as well.

Please note: the bot processes only messages no older than two minutes from now. In other words, the bot is interested in what is happening now (with a slight delay), but not in the far past. This is to prevent a sutuation what a bot had been inactive and then has started to consume messages. Without this condition, it will send captcha to chat members who have already joined and confuse them.

Java version

To make a Jar artefact, run:

make uberjar

The uberjar target calls lein uberjar and also injects the VERSION file into it. The output file is ./target/teleward.jar.

Binary version, Linux

Linux version is built inside a Docker image, namely the ghcr.io/graalvm/graalvm-ce one with native-image extension preinstalled. Run the following command:

make docker-build

The output binary file appears at ./target/teleward.

Binary version, MacOS

gu install native-image
  • Then make the project:

Setting Up Your Bot

  • To run the bot, first you need a token. Contact @BotFather in Telegram to create a new bot. Copy the token and don’t share it.

  • Add your new bot into a Telegram group. Promote it to admins. At least the bot must be able to 1) send messages, 2) delete messages, and 3) ban users.

  • Run the bot locally:

teleward -t <telegram-token> -l debug

If everything is fine, the bot will start consuming messages and print them in console.


See the version with -v, and help with -h. The bot takes into account plenty of settings, yet not all of them are available for configuration for now. Below, we name the most important parameters you will need.

  • -t, --telegram.token: the Telegram token you obtain from BotFather. Required, can be set via an env variable TELEGRAM_TOKEN.

  • -l, --logging.level: the logging level. Can be debug, info, error. Default is info. In production, most likely you will set error.

  • --telegram.offset-file: where to store offset number for the next getUpdates call. Default is TELEGRAM_OFFSET in the current working directory.

  • --lang: the language for messages. Can be en, ru, default is ru.

  • --captcha.style: a type of captcha. When lisp, the captcha looks like a lisp expression (+ 4 3). Any other value type will produce 4 + 3. The operator is taken randomly.


./target/teleward -t <...> -l debug \
  --lang=en --telegram.offset-file=mybot.offset \

For the rest of the config, see the src/teleward/config.clj file.

Deploying on bare Ubuntu

  • Buy the cheapest VPS machine and SSH to it.

  • Create a user:

sudo useradd -s /bin/bash -d /home/ivan/ -m -G sudo ivan
sudo passwd ivan
mkdir /home/ivan/teleward
  • Compile the file locally and copy it to the machine:
scp ./target/teleward ivan@hostname:/home/ivan/teleward/
  • Create a new systemctl service:
sudo mcedit /etc/systemd/system/teleward.service
  • Paste the following config:
Description = Teleward bot
After = network.target

Type = simple
Restart = always
RestartSec = 1
User = ivan
WorkingDirectory = /home/ivan/teleward/
ExecStart = /home/ivan/teleward/teleward -l debug
Environment = TELEGRAM_TOKEN=xxxxxxxxxxxxxx

WantedBy = multi-user.target
  • Enable autoload:
sudo systemctl enable teleward
  • Manage the service with commands:
sudo systemctl stop teleward
sudo systemctl start teleward
sudo systemctl status teleward

For Jar, the config file would be almost the same except the ExecStart section. There, you specify something like java -jar teleward.jar ....

Health check

The bot accepts the /health command which it replies to “OK”.

Further work

  • Implement webhook.
  • Add tests.
  • Report uptime for /health.
  • More config parameters via CLI args.
  • Widnows build.

© 2022 Ivan Grishaev