Steve HarvAI

This is a simple implementation of a "20 Questions" game that I built to showcase the capabilities of multi-agent interaction. The name "Steve HarvAI" is an homage to the famous (and hilarious) Steve Harvey, who hosted the popular game show "Let's Make a Deal."

Project Structure

.
├── agents.py          # Contains the agent definitions
├── game.py            # Contains the game logic (main script)
├── messages.py        # Contains the message objects
├── prompts.py         # Contains the system prompts for the agents
├── tools.py           # Contains the tools that the agents can use
├── utils.py           # Contains utility functions
├── README.md          # This file
├── assets             # Contains assets
├── pyproject.toml     # Contains project metadata and dependencies
└── parallel_game.py   # Contains the script for running multiple games in parallel (see game.py for details)

Setup

# Create a virtual environment and install dependencies
uv venv && source .venv/bin/activate  
uv sync

Also, do not forget to set the OPENAI_API_KEY environment variable to your API key.

Running

Whenever you run the game, it will create a new log file in the game_logs directory. This way, you can either inspect the game progress in your terminal (with pretty coloring for easier reading) or you can open the log file in a text editor to inspect the full run.

Run One Game

# Run the game
python game.py
# You can also run the game with a specific topic for the Guesser to guess
python game.py --topic "Sam Altman"

Run Multiple Games in Parallel

# Run multiple games in parallel
# This will clear the logs directory before running (to avoid clutter) and run 20 games in parallel
python parallel_game.py \
--num_games 20 \ # Number of games to run in parallel
--clear_logs \ # Clear the logs directory before running; just for convenience (default: True)

The output will be a summary of all the games: