Build Your own Telegram Bot with AWS and Node.js

Splitting bills and payments between housemates can be a touchy subject. Debates and disagreements might ensue causing rifts. Faulty manual calculations of who owes what might accidentally have someone paying more or less than what they actually owe. So why not leave the manual and thankless task of tedious calculations and notifying flatmates of what they owe and to whom, to technology. By the end of this article, you will have a telegram bot up and running in an AWS lambda function that does these calculations simply based on telegram messages.

Existing Solutions

Although there are plenty of existing apps and services that have tackled this particular problem, I'm afraid that they've grown to become so complicated that frankly just splitting grocery bills can seem like you're rewiring a supercomputer. So for the sake of simplicity, our solution will only do one thing really well, that is to split dollar amounts equally between group members. It will consist of a telegram chat group, a serverless function, and a database running on AWS that will perform this action.

Architecture

The main parts of the architecture will be:

• AWS Lambda function: To receive, process, and send messages to telegram via the bot's API
• DynamoDB: To store/keep track of the amount owed by each person.
• API Gateway: As the name suggests, an API pathway to interface with the lambda function. Lambda by default does not generate an API that can be invoked.

Defining AWS resources using SAM

We'll deploy our resources to AWS using SAM command-line tool, which lets us build, test, and deploy our AWS resources using either a guided method or by manually specifying the resource template. In practical scenarios, most organizations use a resource template for deploying their resources. So we'll be using this method as this will prove more beneficial for those with already a basic knowledge of the cloud. If you want a beginner-friendly guide to deployments, AWS has an awesome guide: https://docs.aws.amazon.com/serverless-application-model/latest/developerguide/serverless-getting-started-hello-world.html.

First of all, we'll be organizing our project structure like this:

TelegramBotServer
  |- template.yml
  |- BotHelper
      |-index.js
      |-db.js
      |-processMessages.js

Inside the BotHelper, index.js will be the entry point of the function, db.js will handle database CRUD operations and processMesssages.js will use regex to parse the input messages and process them accordingly.

To deploy our AWS cloud infrastructure, we'll be using a YAML file that will contain the AWS Serverless Application Model (SAM) specifications. This template will tell AWS how to deploy our combination of resources including Lambda function, API Gateway, and DynamoDB. This template.yml will look like:

AWSTemplateFormatVersion: "2010-09-09"
Transform: AWS::Serverless-2016-10-31
Description: A serverless function that listens to incoming webhooks from the Telegram Server
Resources:
  TelegramBot:
    Type: AWS::Serverless::Function
    Properties:
      Handler: BotHelper/index.handler
      Runtime: nodejs14.x
      Timeout: 60 
      Environment:
        Variables:
          TABLE_NAME: People
      Policies:
        - DynamoDBCrudPolicy:
            TableName: People
      Events:
        TelegramServiceAPI:
          Type: Api
          Properties:
            Path: /message
            Method: POST
  TelegramBotInvokePermission:
    Type: AWS::Lambda::Permission
    Properties:
      Action: lambda:InvokeFunction
      FunctionName:
        Fn::GetAtt:
          - TelegramBot
          - Arn
      Principal: apigateway.amazonaws.com
      SourceArn:
        Fn::Sub: arn:aws:execute-api:${AWS::Region}:${AWS::AccountId}:*/*/*/*
  PeopleRecords:
    Type: AWS::DynamoDB::Table
    Properties:
      TableName: People
      AttributeDefinitions:
        - AttributeName: id
          AttributeType: S
      KeySchema:
        - AttributeName: id
          KeyType: HASH
      ProvisionedThroughput:
        ReadCapacityUnits: 5
        WriteCapacityUnits: 5

In the YAML file we define these resources:

• TelegramBot: Our Lambda function with Node 14 runtime, the handler file. Also, we specify the timeout as 60 seconds. This is important to note since by default the timeout in serverless functions is 3 seconds which isn't enough for the operations we'll be performing. We also add environment variables like TABLE_NAME which is our DynamoDB table name. We also add policies that allow the function to make updates to the database. Optionally, we can limit the concurrency using the ReservedConcurrentExecutions property which by default is 1000. The function will be triggered by a POST API endpoint /message which we define in the file using the Events property.
• API Gateway: This gateway will trigger the Lambda function so we must explicitly provide a function invoke permission (lambda:InvokeFunction) which in the above file is called TelegramBotInvokePermission
• PeopleRecords: The DynamoDB will act as our store which will have a primary key called id which will be used to uniquely identify our resource. You can also explicitly specify properties like read/write capacities to 5 consistent reads/writes per second.

Setting Up a Telegram Bot

To set up a telegram bot, well, there's a bot for that. It's called Botfather (get it?). You can access Botfather via https://telegram.me/BotFather. There you can use the /newbot command to create your bot.

Your interactions should look something like this:

At the end of the process, you should get a unique secret token that you will use to communicate with the bot. A full list of commands accepted by the bot can be found at https://core.telegram.org/bots/api. We'll be communicating via HTTP requests.

Now we'll need to specify the webhook for the bot. This is basically asking Telegram to call our Lambda function with bot-related events. We're more interested in the group chat messages. We can set up our webhook using the following curl command:

curl --request POST \
--url https://api.telegram.org/bot<TELEGRAM_TOKEN>/setWebhook\
--header 'content-type: application/json'\
--data '{"url": "<LINK_TO_YOUR_LAMBDA_API>"}'

Additionally, you might need to update your bot's privacy credentials using /BotFather . This will enable us to read group messages. For this, use the command /setprivacy:

Next, you need to create a telegram group with all the members including your bot. And using our lambda function, we'll receive all the chat messages via the webhook.

Programming the Bot

Now for the fun part, we're going to program the bot so that it reads messages from the chat group where people will share how much they spent and on what. We will then take that chat message, and using regex store the amount text in our Dynamo DB database which will be our source of truth for who spent what. We will then return a message back to the user specifying who owes how much, and to whom.

First we'll install all our dependencies:

npm i axios aws-sdk lodash

Here's our main function index.js

const axios = require('axios');
const { processMessage } = require('./processMessage');
const { get, upsert } = require('./db');

async function addMembers(chatId, message) {
    const newChatMembers = message.new_chat_members;
    const group = await get(Math.abs(chatId).toString());
    const people = group?.people ??
        (message.from.is_bot === true ?
            {} : { [message.from.first_name]: { id: message.from.id, spent: 0, owes: {} } }
        );

    newChatMembers.forEach(member => {
        if (!people[member.first_name]) {
            people[member.first_name] = {
                id: member.id,
                spent: 0,
                owes: {}
            }
        }
    });

    await upsert(Math.abs(chatId).toString(), { people: people });
}

exports.handler = async (event) => {

    const body = JSON.parse(event.body);

    if (body.message?.new_chat_members?.length) {
        await addMembers(body.message.chat.id, body.message);
        return {
            statusCode: 200,
            body: JSON.stringify('OK')
        }
    }

    if (!body.message?.text) {
        return {
            statusCode: 200,
            body: JSON.stringify('Nothing to do')
        }
    }

    const group = await get(Math.abs(body.message.chat.id).toString());

    if(!group) {
        return {
            statusCode: 200,
            body: JSON.stringify('No group found')
        }
    }

    const processedMessage = processMessage(body.message, group.people);

    if(!processedMessage) {
        return {
            statusCode: 200,
            body: JSON.stringify('Nothing to do')
        }
    }

    if(processedMessage?.peopleRecord){
        await upsert(Math.abs(body.message.chat.id).toString(), {people: processedMessage.peopleRecord});
    }

    try {
        if (processedMessage.reply){
            await axios.get(`https://api.telegram.org/bot${process.env.TELEGRAM_TOKEN}/sendMessage?chat_id=${body.message.chat.id}&text=${encodeURI(processedMessage.reply)}`)
        }
        if (processedMessage.gif) {
            const gif = await axios.get(`https://g.tenor.com/v1/random?key=${process.env.TENOR_KEY}&q=${encodeURI(processedMessage.gif)}&limit=1`);
            await axios.get(`https://api.telegram.org/bot${process.env.TELEGRAM_TOKEN}/sendAnimation?chat_id=${body.message.chat.id}&animation=${encodeURI(gif.data.results[0].media[0].gif.url)}`)
        }
    } catch (e) {
        console.error('Error sending message', e);
        await axios.get(`https://api.telegram.org/bot${process.env.TELEGRAM_TOKEN}/sendMessage?chat_id=${body.message.chat.id}&text=${encodeURI('Problems with the server !')}`)
        // Optional: Throw error so Telegram webhook will retry
    }

    return {
        statusCode: 200,
        body: JSON.stringify('OK')
    };

};

Let's go through what happens here one step at a time:

• The main handler gets the event from the Telegram servers.
• The program checks if the event has any new chat members, if so it will add any new members to the existing group including the invitee if they're not already added yet in the addMembers function.
• If the event has an incoming message, then the program processes the message accordingly and returns a reply, updated records, and optionally a gif. We can update the existing records, and reply to the chat using a simple message and a gif (which can be done by a simple call to the TENOR API).

Note that we're using a positive chatId as the key. This is because we want to organize people by chat groups and since Telegram uses negative values for its chat groups we have to convert them into a positive values.

Since we're sending multiple requests with awaits, it is important that we change the default timeouts that Lambda provides by default (3 seconds) to something of a higher value like we did in our SAM template.

Our db.js will handle our CRUD operations (in this case CRU operation) by using the aws-sdk library.

const AWS = require('aws-sdk');
const docClient = new AWS.DynamoDB.DocumentClient();
const tableName = process.env.TABLE_NAME;

async function get(id) {
    const data = await docClient.get({
        TableName: tableName,
        Key: {
            'id': id
        }
    }).promise();

    return data.Item;
}

async function upsert(id, item) {
    await docClient.put({
        TableName: tableName,
        Item: {
            ...item,
            id: id,
        }
    }).promise();
}

module.exports = {
    get,
    upsert
}

Although omitted for the sake of brevity, it is crucial that there is proper error handling in cases of database operations.

Now the main part of our application is the processMessage function which will do the actual bill dividing tasks.

const _ = require('lodash');

function divideBills(purchase, peopleRecord) {
    let people = { ...peopleRecord };

    if(!people[purchase.spender]) {
        people[purchase.spender] = {spent: 0, owes: {}};
    }

    let peopleNames = Object.keys(people);
    peopleNames.forEach(person => {
        if (person === purchase.spender) {
            people[person].spent = people[person].spent + purchase.amount;
        } else {
            people[person].owes[purchase.spender] = (people[person].owes[purchase.spender] ? people[person].owes[purchase.spender] : 0) + _.round((purchase.amount / peopleNames.length), 2);
            people[person].owes = people[person].owes ? people[person].owes : {};
        }
    })
    return people;
}

function balanceBills(peopleRecord) {
    let people = { ...peopleRecord };
    let peopleNames = Object.keys(people);
    peopleNames.forEach(person => {
        const personOwes = Object.keys(people[person].owes);
        personOwes.forEach(personOwed => {
            if (people[person].owes[personOwed] === 0) {
                return;
            }
            if (people[person].owes[personOwed] >= people[personOwed].owes[person]) {
                people[person].owes[personOwed] = people[person].owes[personOwed] - people[personOwed].owes[person];
                people[personOwed].owes[person] = 0;
            } else if (people[personOwed].owes[person] > people[person].owes[personOwed]) {
                people[personOwed].owes[person] = people[personOwed].owes[person] - people[person].owes[personOwed];
                people[person].owes[personOwed] = 0;
            }
        })
    })
    return people;
}

function clearBills(peopleRecord, person) {
    let people = { ...peopleRecord };
    people[person].owes = {};
    return people;
}

function getOwed(people, person) {
    if(Object.keys(people[person].owes).length === 0) {
        return `You owe nobody. ${person} is free.`;
    }

    let msg = `${person} owes:`;
    Object.keys(people[person].owes).forEach(personOwed => {
        msg += `${personOwed} ${people[person].owes[personOwed]} dollars,`;
    });
    return msg;
}

function processMessage(message, people) {
    if(!message.from.first_name) throw new Error('no user');
    if(!message.text) throw new Error('no text message');
    const messageText = message.text;

    //Example message: "tbot is spent 100 dollars on food"
    if(/tbot\si\sspent\s(\d+((.)|(.\d{0,2})?))\sdollars\son.+/gmi.exec(messageText)?.length) {
        const amountSpent = /jesus\si\sspent\s(\d+((.)|(.\d{0,2})?))\sdollars\son.+/gmi.exec(messageText)[1];
        const peopleRecord = balanceBills(divideBills({ spender: message.from.first_name, amount: Number(amountSpent) }, people));
        return {
            peopleRecord,
            reply: `${message.from.first_name} spent ${amountSpent} dollars, and everyone else owes ${message.from.first_name}, ${_.round((Number(amountSpent) / Object.keys(peopleRecord).length), 2)} dollars.`,
        }
    }

    //Example message: "tbot i cleared my bills"
    if(/tbot\si\scleared\smy\sbills/gmi.exec(messageText)?.length) {
        const peopleRecord = clearBills(people, message.from.first_name);
        return {
            peopleRecord,
            reply: `${message.from.first_name} cleared their bills.`,
            gif: 'hurray'
        }
    }

    //Example message: "tbot how much do i owe"
    if(/tbot\show\smuch\sdo\si\sowe?.+/gmi.exec(messageText)?.length) {
        const msg = getOwed(people, message.from.first_name);
        return {
            people,
            reply: msg,
        }
    }
}

module.exports = {
    processMessage,
}

This function uses regex to check for key phrases like tbot i spent 100 dollars on food this will automatically divide the 100 dollars between members of the group using the divideBills() function which divides the 100 dollars equally between the members and the balanceBills() function which will then rebalance balance everything so that person A does not need to pay person B if person B owes more money.

Other phrases available are tbot i cleared my bills which will clear the existing owed balance and tbot how much do i owe which show the message sender's owing balance.

Finally, some unit tests are never a bad idea.

Deploying to AWS

Before we start using deploy commands, we'll need to set up some configurations in our local environment. The following environment variables should be present before using the deploy commands:

• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY
• AWS_REGION

The access keys should be generated from a console which ideally should be associated with an IAM role. However, keys from a root user also work but it is not recommended.

Alternatively, you can use the aws configure command to set up these environment variables.

You can use these commands to set it up in your Mac/Linux:

export AWS_ACCESS_KEY_ID=your_access_key_id
export AWS_SECRET_ACCESS_KEY=your_secret_access_key

For windows, you can replace export with set.

You can now validate your template.yml file using the command sam validate --region ap-southeast-2 which should hopefully tell you that you have a valid template. If not, investigate the details of the error. For the deployment operations, we'll need to create an s3 bucket using the command:

aws s3 mb s3://telegram-bot-deployement-bucket --region ap-southeast-2

To verify it's been created, you can list all s3 buckets using aws s3 ls command.

After that, we need to package our code using:

sam package \                                                  
--template-file template.yml \     
--output-template-file package.yml \
--s3-bucket telegram-bot-deployement-bucket \                                 
--region ap-southeast-2

Then we can deploy using:

sam deploy \                                                                                                                   
--template-file package.yml \
--stack-name telegram-service-stack \    
--capabilities CAPABILITY_IAM \
--region ap-southeast-2

We have to explicitly specify that we are okay with creating an IAM-related resource using the --capabilities argument.

Since you'll probably be packaging and deploying it together, to save us some time, we can use an alias for deployments using:

alias deployx='sam package --template-file template.yml --output-template-file package.yml --s3-bucket telegram-bot-deployement-bucket --region ap-southeast-2'

So next time you can deploy using deployx

Testing the Bot

To test the bot, simply create a group with your bot and start adding members to the group.

Then try sending a message like this:
Try sending other messages like <botname> i cleared my bills, or <botname> how much do i owe . If everything went right then congratulations, you now have a bot that can divide the bills for you.

Source Code

The source code for this tutorial is available at https://github.com/nipeshkc7/jesus-telegram-bot. Feel free to clone the repository and deploy your own instance to AWS.