AutoRAG

AutoRAG is a Retrieval-Augmented Generation (RAG) system that creates a Telegram bot capable of mimicking a specific persona or original author you want. By indexing a message history file, the bot adopts the slang, tone, and writing style of the original author.

Features

• Persona Mimicry: Uses RAG to ground LLM responses in a provided message history.
• Dynamic Model Switching: Supports multiple Ollama models (e.g., Qwen, Gemma).
• User Management API: Built-in endpoints to dynamically add or remove authorized Telegram users.
• Security: Access is restricted to a configurable list of Telegram User IDs stored in Redis.
• Production Ready: Includes Docker Compose configurations for both GPU (NVIDIA) and CPU-only environments.
• GitLab CI/CD: Pre-configured pipeline for building and pushing Docker images.

Architecture

• Backend: FastAPI providing the chat logic and user management.
• Vector Database: Redis (Redis Stack) for high-performance vector similarity search.
• LLM Engine: Ollama (running locally or in a container).
• Telegram Bot: Python-based bot using python-telegram-bot.

Installation & Setup

Prerequisites

• Docker and Docker Compose
• (Optional) NVIDIA Container Toolkit for GPU acceleration
• A Telegram Bot Token (from @BotFather)

1. Prepare Data

Place your persona's message history in a file named messages.txt in the root directory.

2. Run with Docker Compose

For NVIDIA GPU Acceleration:

export TELEGRAM_TOKEN="your_telegram_bot_token"
docker-compose up -d --build

For CPU Only:

export TELEGRAM_TOKEN="your_telegram_bot_token"
docker-compose -f docker-compose-noGPU.yml up -d --build

3. Initialize Models

Once the containers are running, you must pull the required models into the Ollama container:

docker-compose exec ollama ollama pull bge-m3
docker-compose exec ollama ollama pull qwen3:4b

User Management API

You can manage authorized users via the FastAPI backend (default port 8000):

• List Users: GET /users
• Add User: POST /users (Body: {"user_id": 12345678})
• Remove User: DELETE /users/{user_id}

Configuration

Environment variables can be adjusted in the docker-compose files:

• TELEGRAM_TOKEN: Your bot token.
• EMBEDDING_MODEL: Defaults to bge-m3.
• OLLAMA_BASE_URL: Connection string for the Ollama service.
• REDIS_URL: Connection string for the Redis service.

Author

Giuseppe Marco Randazzo gmrandazzo@gmail.com