TelegramGPT

A ChatGPT Bot for Telegram.

Features

• Chat with GPTs via Telegram UI.
• Include chat message history.
• Customisable GPT system prompts.
• Message moderation.
• Specify allowed chat ids.
• Configurable serverless deployment.
• Image recognition with GPT-4 Vision.
• Image generation with DALL-E 3.
• Audio recognition with Whisper.
• Audio generation.

Get Started

This guide will walk you through deploying TelegramGPT on AWS. Ensure you have the AWS CLI and Terraform CLI setup, and Rust installed on your machine.

Prepare the Project

1. Clone the repository:

   git clone https://github.com/rossheat/telegramgpt.git
   

2. Navigate to the project directory:

   cd telegramgpt/
   

3. Open the project in your preferred editor:

   code .  
   # or
   vim .   
   

4. Prepare configuration file: Rename infra/terraform.tfvars.example to infra/terraform.tfvars.

Configure a Telegram Bot

1. Create a new bot on Telegram via @BotFather. Follow the instructions to receive your bot token.

2. Set your bot token: Replace the example telegram_bot_token in infra/terraform.tfvars with your new bot token.

3. Set your bot name: Replace the example bot_name with your new bot name.

4. Adjust bot settings:

   • Access your bot settings in BotFather.
   • Navigate to ‘Bot Settings’ > ‘Group Privacy’ > ‘Turn off’. This allows your bot to access messages in group chats.

5. Set a webhook secret token: Generate a token using the command:

   openssl rand -hex 32
   

   Assign this token to telegram_webhook_secret_token in infra/terraform.tfvars.

Configure GPT

1. Generate an OpenAI API key at OpenAI Platform.

2. Set your OpenAI API key: Replace the example openai_api_key in infra/terraform.tfvars with your new key.

3. Customise the system prompt: The system_prompt guides how the GPT model will interact in the chat. When setting this prompt, follow these rules:

   • The prompt should be clear and specific to direct the model's responses.
   • It must instruct the model to never output timestamps.
   • Responses should be in the specified JSON format: {"has_response" : <boolean>, "message" : <string>}.
   • If the model has no response, set has_response to false and message to an empty string.

   Example for Group Chat:

   "You are GPT, a helpful assistant in a Telegram group chat. Respond only to messages directed at you. Do not respond to general conversation. Never output timestamps. Respond in the following JSON format: {\"has_response\" : , \"message\" : }"

   Update system_prompt in infra/terraform.tfvars as per these guidelines.

Set Up MongoDB

1. Create a MongoDB deployment at MongoDB Cloud.
2. Configure MongoDB:
   • Create a new project and a free M0 deployment.
   • Set up a user for database access.
   • Allow access from all IPs (0.0.0.0/0) for Lambda function connectivity.
3. Retrieve and set the connection string: Update mongodb_uri in infra/terraform.tfvars with your MongoDB connection string.

Configure Allowed Chat IDs

1. Start a chat with your bot in Telegram and send a couple of messages. You will not get a response.
2. Retrieve your chat ID: Run ./scripts/get_private_chat_id.sh from the project root. Locate the chat ID in the JSON output, and add it to allowed_chat_ids in infra/terraform.tfvars.

Test the Setup

1. Perform local CI checks:

   ./scripts/ci.sh

   Ensure that this script passes without errors before contributing.

2. Start the Rust AWS Lambda development server:

   ./scripts/watch.sh

3. Send a test webhook payload:

   ./scripts/local_lambda_test.sh

Deploy to AWS Lambda

1. Initiate deployment:

   ./scripts/deploy.sh
   

   Use the --delete-messages option to delete any existing messages in the MongoDB messages collection.

2. Test your deployment:

   ./scripts/prod_lambda_test.sh

Interact with Your Bot

After deployment, interact with your bot through the Telegram UI. Responses will come from your chosen OpenAI GPT model.

Monitoring and Logs

Lambda function logs are available in AWS CloudWatch.

Clean Up

To destroy the AWS infrastructure created by this project:

./scripts/destroy_infra.sh

Contributing

We welcome contributions to the TelegramGPT project. Please see CONTRIBUTING.md for guidelines.

Code of Conduct

To ensure a welcoming and positive environment for everyone, all contributors and participants are expected to adhere to our Code of Conduct. Please read CODE_OF_CONDUCT.md for details.

License

This project is open-source and available under the MIT License. By contributing, you agree that your contributions will be licensed under this license.

Community and Support

Join our community to discuss features, ideas, and get help:

• Telegram Group: Join our Telegram group here for discussions and support.
• GitHub Issues: For bug reports and feature requests, use the GitHub issues page.