Madrook Bot

A Telegram bot for managing group discussions with AI-powered features.

Features

General

• GPT Integration: Mention the bot to get AI-powered text responses with conversation threading
  • Reply to bot's answers to continue conversations with context (5 exchanges history)
  • Multiple users can branch conversations from the same message
  • Database-backed conversation persistence with 7-day retention
• Image Generation: Use image: <prompt> to generate images via Google Gemini (free tier available)
  • Case-insensitive prefix
  • Supports Imagen 3 (default) or configurable models
  • Generates and sends images directly in chat
• Text-to-Speech: Use read: <text> to generate audio via ElevenLabs
• Activity Statistics: /stat command shows group activity charts (admins only)
  • Tracks message activity in hourly buckets
  • Generates visual charts with top 10 contributors
  • GPT-powered analysis of activity patterns
  • 6-month data retention
• Class Scheduler System: Automated system for scheduling and managing online classes.
• RAG/MCP Tool: A Retrieval-Augmented Generation system that gives the bot long-term memory of chat conversations.

Class Scheduler System

The bot features a fully automated system for managing weekly online classes.

• Automatic Scheduling: A new class is automatically scheduled for the next Sunday at 19:00 Berlin time when created.
• GPT-Powered Announcements: The bot generates engaging and friendly class announcements using GPT. These are posted with a random delay (1-6 hours) to feel more natural.
• RSVP Tracking: Users can RSVP by reacting to the announcement message. The bot tracks attendance and can send reminders based on the number of participants.
• Automated Reminders:
  • A 6-hour reminder is sent before the class. If there are fewer than 4 RSVPs, the message has a more urgent tone to encourage sign-ups.
  • A final 1-hour reminder is sent just before the class starts.
• Lifecycle Management: The announcement message is automatically pinned in the chat and then unpinned 2 hours after the class has finished.

RAG/MCP (Memory/Context) System

The bot has a long-term memory of conversations in a dedicated group, implemented as a Retrieval-Augmented Generation (RAG) system. This allows the bot to answer questions about past discussions.

• Message Ingestion: The bot listens for messages in a specified group (MEMORY_GROUP_ID).
• Contextual Message Combining: To create more meaningful entries, consecutive messages from the same user are automatically combined into a single document if they are sent within a 90-second window.
• Vector Embeddings: The text is converted into a vector embedding using an OpenAI model (e.g., text-embedding-3-small).
• Vector Storage: The embedding and the message content (including a link back to the original Telegram message) are stored in a Qdrant vector database.
• Contextual Search: The GPT model can use a search_chat_history tool. When a user asks a question about the past, the model can use this tool to search the Qdrant database for the most relevant messages and use them as context to formulate an answer.
• Privacy: Usernames are hashed before being stored.

Commands

• /help - Shows help message
• /idiom <term> - Shows the definition of an idiom from idioms.thefreedictionary.com
• /stat - Show group activity statistics (admins only, 1/hour rate limit)
• Mention the bot (@bot_name <question>) - Ask GPT, reply to continue conversation
• image: <prompt> - Generate images using Google Gemini AI
• read: <text> - Convert text to speech using ElevenLabs

Environment Variables

Required

• BOT_TOKEN - Telegram bot token
• OWNER_ID - Telegram user ID of the bot owner
• CLASS_GROUP_ID - ID of the Telegram group for class discussions

OpenAI/GPT (optional)

• GPT_TOKEN - OpenAI API key
• GPT_MODEL - OpenAI model to use (e.g., gpt-4, gpt-3.5-turbo)
• GPT_SYSTEM_PROMPT - Custom system prompt for conversations (defaults to helpful assistant)
• OPENAI_TEMPERATURE - Temperature for GPT responses (default: 1.0)
• OPENAI_URL - OpenAI API base URL (default: https://api.openai.com/v1)

Google Gemini Image Generation (optional)

• GEMINI_API_KEY - Google Gemini API key (has free tier limits)
• GEMINI_IMAGE_MODEL - Image model to use (default: gemini-2.5-flash-image-preview)

ElevenLabs TTS (optional)

• ELEVENLABS_API_KEY - Your ElevenLabs API key
• ELEVENLABS_VOICE_NAME - The name of the ElevenLabs voice to use
• ELEVENLABS_MODEL_ID - The ID of the ElevenLabs model to use

RAG/MCP (Memory/Context) System (optional)

• MEMORY_GROUP_ID - The ID of the Telegram group where the bot should listen for messages to be stored in the memory.
• MIN_MESSAGE_LENGTH - The minimum length of a message to be considered for memory storage (default: 20 characters).
• MESSAGE_COMBINE_WINDOW - The time window in seconds within which consecutive messages from the same user are combined into a single entry (default: 90 seconds).
• QDRANT_HOST - The hostname or IP address of the Qdrant service (default: qdrant).
• OPENAI_EMBEDDING_MODEL - The model to use for generating message embeddings (e.g., text-embedding-3-small).
• TOOL_API_URL - The URL for the tool API service (default: http://tool-api:8081).

Running

The recommended way to run the application is with docker-compose, which manages the bot, the Qdrant vector database, and the tool-api service.

With Docker Compose

1. Environment File: You will need to create a .env file in the project root to store your secrets and configuration. You can do this by copying the provided example file:

   cp .env.example .env

   Then, edit the .env file with your specific credentials (like BOT_TOKEN, GPT_TOKEN, etc.).

2. Start Services:

   docker-compose up --build -d

   This command builds the images if they don't exist and starts all services (madrookbot, qdrant, tool-api) in the background.

3. View Logs:

   docker-compose logs -f

With Docker (Manual)

If you prefer to run only the bot container without the associated services:

docker build -t madrookbot .
docker run \
  -e BOT_TOKEN=your_token \
  -e OWNER_ID=your_id \
  -e CLASS_GROUP_ID=group_id \
  # ... other environment variables
  madrookbot

Without Docker

go build -mod=vendor -o madrookbot
./madrookbot

Development

Build & Test

• Build: go build -mod=vendor -o madrookbot
• Run all tests: go test -v ./...
• Run a single test: go test -v -run=TestMyFunction

Code Style

• Standard Go formatting (gofmt)
• Group standard library imports first, then third-party packages.
• Use CamelCase for exported names and lowerCamelCase for unexported names.
• Check errors immediately and use an early return pattern.
• Add comments for all exported functions and types.

Project Structure

• The project follows a single-package architecture.
• Dependencies are vendored.
• Configuration is managed through environment variables.

License

This project is licensed under the MIT License - see the LICENSE file for details.