AI Telegram Bot with Local LLM

A secure, private Telegram bot powered by a local large language model (Qwen3:4b).

Overview

This project runs a Telegram bot that connects to a local Ollama instance for private, offline AI conversations. No data leaves your machine.

Hardware: Beelink SER8 (AMD Ryzen 7 8745HS, Radeon 780M, 64GB RAM) OS: Omarchy Linux (Arch-based)

What's Included

• bot.py - Telegram bot written in Python using python-telegram-bot
• Dockerfile - Containerized Python environment for the bot
• docker-compose.yml - Orchestrates Ollama + bot services
• .env - Environment variables (tokens, model settings)
• .gitignore - Excludes sensitive files

Quick Start

# Start the stack
docker-compose up -d

# Check status
docker-compose ps

# View logs
docker-compose logs -f bot

# Stop
docker-compose down

Configuration

Edit .env to customize:

TELEGRAM_TOKEN="your_token_here"
OLLAMA_MODEL=qwen3:8b

To change models:

docker exec ollama ollama pull <model-name>
# Example: ollama pull gemma3:12b

Then update OLLAMA_MODEL in .env and restart:

docker-compose restart bot

Development

When editing bot.py:

docker-compose up -d --build

Models Tested

Model	Size	Notes
qwen3:8b	5.2 GB	Current - balanced, efficient
phi3:mini	~3 GB	Smaller, faster but less capable

Security

• Token stored in .env (gitignored)
• Docker provides container isolation
• All processing happens locally
• No cloud dependencies

Architecture

Telegram → ai-bot container → ollama container → Local LLM
                    ↓
             Docker network

Both services communicate via isolated Docker bridge network (ollama-network).

Troubleshooting

Bot not responding:

docker-compose logs bot

Ollama issues:

docker-compose logs ollama

Restart everything:

docker-compose down && docker-compose up -d

Check running models:

docker exec ollama ollama list

References

• Ollama: https://ollama.com
• python-telegram-bot: https://python-telegram-bot.org
• ROCm (AMD GPU): https://rocm.docs.amd.com

License

MIT