econagents 0.0.7

No project description provided

---

econagents

econagents is a Python library that lets you use LLM agents in economic experiments. The framework connects LLM agents to game servers through WebSockets and provides a flexible architecture for designing, customizing, and running economic simulations.

Installation

# Install from PyPI
pip install econagents

# Or install directly from GitHub
pip install git+https://github.com/IBEX-TUDelft/econagents.git

Framework Components

econagents consists of four key components:

1. Agent Roles: Define player roles with customizable behaviors using a flexible prompt system.
2. Game State: Hierarchical state management with automatic event-driven updates.
3. Agent Managers: Manage agent connections to game servers and handle event processing.
4. Game Runner: Orchestrates experiments by gluing together the other components.

Example Experiments

The repository includes three example games:

1. prisoner: An iterated Prisoner's Dilemma game with 5 rounds and 2 LLM agents.
2. ibex_tudelft/harberger: A Harberger Tax simulation with LLM agents.
3. ibex_tudelft/futarchy: A Futarchy simulation with LLM agents.

Running the Prisoner's Dilemma game

The simplest game to run is a version of the repeated prisoner's dilemma game that runs on your local machine.

# Run the server
python examples/prisoner/server/server.py

# Run the experiment (in a separate terminal)
python examples/prisoner/run_game.py

Note: you still have to set up the connection to the agents.

Key Features

• Flexible Agent Customization: Customize agent behavior with Jinja templates or custom Python methods
• Event-Driven State Management: Automatically update game state based on server events
• Turn-Based and Continuous Action Support: Handle both turn-based games and continuous action phases
• LangChain Integration: Built-in support for LangChain's agent capabilities

Documentation

For detailed guides and API reference, visit the documentation.

You should also check out the econagents cookbook for more examples.