LLM Agent Implementation

This project reproduces the agent code from the Fly.io blog article: You Should Write An Agent.

Caution

The code in this repository is purely vibe-coded and was created just for fun during a late-evening session. Do not use it for anything serious or for real products!

Important

Instead of using OpenAI, this project uses the SAIA API endpoint, which is fully OpenAI-compatible. SAIA is the Scalable Artificial Intelligence (AI) Accelerator developed by the GWDG (where I also work). You can request access via our service catalog if you meet the user requirements for the AI service center KISSKI.

Overview

This repository contains three implementations:

1. agent.py - Basic chat agent (ChatGPT-like)
2. agent_multi.py - Multi-personality agent (Alph tells truth, Ralph tells lies)
3. agent_tools.py - Tool-enabled agent with ping capability

Prerequisites

• Docker and Docker Compose
• API key for the academiccloud endpoint

Setup

The agents are configured to always use the SAIA API endpoint: https://chat-ai.academiccloud.de/v1

1. Set your API key as an environment variable:

   export OPENAI_API_KEY=your-api-key-here

2. Or create a .env file:

   OPENAI_API_KEY=your-api-key-here
   

Environment Variables:

• OPENAI_API_KEY or API_KEY - Your API key (required)
• MODEL - Model name to use (defaults to qwen3-32b)

Troubleshooting 401 Unauthorized Error:

• Make sure your API key is set: echo $OPENAI_API_KEY
• Verify the API key is correct and valid for the academiccloud endpoint
• When using Docker, ensure the environment variable is passed: docker-compose run --rm -e OPENAI_API_KEY=$OPENAI_API_KEY agent

Usage

Basic Agent

Run the basic chat agent (use run for interactive input):

docker-compose run --rm agent

Or build and run manually:

docker build -t aiagent .
docker run -it --rm \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -e MODEL=$MODEL \
  aiagent python agent.py

Multi-Personality Agent

Run the multi-personality agent:

docker-compose run --rm agent-multi

Or manually:

docker run -it --rm \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -e MODEL=$MODEL \
  aiagent python agent_multi.py

Tool-Enabled Agent

Run the agent with ping tool:

docker-compose run --rm agent-tools

Or manually:

docker run -it --rm \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -e MODEL=$MODEL \
  aiagent python agent_tools.py

Note: Use docker-compose run instead of docker-compose up for interactive sessions that require input. The run command properly handles stdin/stdout for interactive terminals.

Examples

Basic Agent

> What is 2+2?
>>> 4

Multi-Personality Agent

> Who are you?
>>> I'm not Ralph.  (or) Yes—I'm Alph. How can I help?

Tool-Enabled Agent

> describe our connectivity to google
>>> [Agent will ping google.com, www.google.com, and 8.8.8.8 automatically]

Notes

• The code uses qwen3-32b model by default
• The OpenAI API structure matches the current Chat Completions API
• Tool calls are handled automatically in a loop until the agent is satisfied
• All agents maintain conversation context throughout the session
• API endpoint is hardcoded to: https://chat-ai.academiccloud.de/v1

Development

To run locally without Docker:

pip install -r requirements.txt
export OPENAI_API_KEY=your-key
python agent.py  # or agent_multi.py or agent_tools.py