ChatArena: Build Multi-Agent Language Game Environments

Overview

ChatArena provides environments where multiple language-model agents interact through games and other structured scenarios. It addresses the need for a shared framework to define game rules, observations, players, and interaction loops when studying how LLM agents communicate and collaborate.

It is aimed at researchers and developers who want to prototype or examine agent behavior, rather than people looking for a maintained, general-purpose chat application. The repository states that the project was deprecated on August 11, 2025, with no further updates or support planned.

Key Features

• Models games as environments with state transitions and natural-language observations for players.
• Provides an Arena abstraction to run an environment with a collection of players.
• Supports language backends that turn text inputs into model responses.
• Includes conversation, moderator-led games, Chameleon social deduction, and PettingZoo-based chess and tic-tac-toe environments.
• Offers CLI and Web UI interfaces for interacting with games and testing agents.
• Allows custom environments, players, backends, and Arena interaction loops.

Use Cases

• LLM researchers can compare agent behavior in structured multi-player interactions.
• Agent developers can prototype prompts and roles in games such as social deduction or turn-based conversation.
• Educators can use language-game examples to explore communication and collaboration among AI agents.
• Developers porting an existing game environment can use the included PettingZoo integrations as reference examples.

Project Facts

• Language: Python
• License: Apache-2.0
• Stars: 1.6k
• Forks: 151
• Topics: ai, artificial-intelligence, chatgpt, gpt-4, large-language-models, multi-agent, multi-agent-reinforcement-learning, multi-agent-simulation, natural-language-processing, python
• Archived: No

Getting Started

Install the package with pip:

pip install chatarena

See the repository README for optional dependencies, configuration examples, and usage details.

Alternatives

• llm-consortium: It coordinates multiple LLMs to solve problems through consensus, rather than running language games to study agent interaction and social behavior.
• Strands Agents SDK-Python: It is a general-purpose SDK for building and deploying AI agents, rather than a research framework centered on language games.

Considerations

• The project was deprecated on August 11, 2025, and is no longer planned to receive updates or support.
• Using hosted LLM backends such as GPT-3.5-turbo or GPT-4 requires an OpenAI API key; other backends and environments may require optional dependencies.
• The repository describes a framework for research and prototyping. Evaluate whether its environment coverage and maintenance status fit your needs before adopting it for a new project.