Truman Agents
An open-source, multi-LLM-agent social media simulation platform
Truman Agents is an open-source platform for creating realistic social media simulations populated by AI characters. Researchers design social situations, define how characters behave, and observe how multiple human participants respond, all without programming. Developed by the DesignAI Group at Cornell.
How researchers use it
You design a social situation by creating its cast of characters. Each character has a role (such as bully, victim, or bystander), a backstory, personality traits, and behavioral tendencies that determine how they respond to social dynamics. Some characters can be fully scripted for experimental control, while others are driven by large language models and respond dynamically to what participants do. Everything is configured through CSV and JSON files, and you can customize the platform's appearance to support your study's cover story.
When a participant enters, they see a social media feed where your characters are already active, posting, commenting, liking, and reacting to each other (powered by a companion service called TrumanWorld). Participants can also message characters directly and receive in-character responses. You can structure the simulation into stages, each with defined objectives that a built-in grader scores automatically.
Multiple participants can log onto the same social media feed, interacting with each other in addition to the AI characters. The platform records every participant interaction, including time on site, feed actions, and chat messages, exportable for analysis.
Example application: Upstanders' Practicum
The codebase includes a complete set of cyberbullying scenarios and agent configurations from Upstanders' Practicum, a multi-agent simulation designed to help young adults practice public bystander intervention. It covers eight scenarios including hazing, cyberstalking, and doxxing. These serve as working examples that you can run, modify, or use as templates for your own research.
Citation
If you use Truman Agents in your research, please cite:
title={Attention: What Prevents Young Adults from Speaking Up
Against Cyberbullying in an LLM-Powered Social Media
Simulation},
author={Qian Yang and Jessie Jia and Elaine Tsai
and Amy Li and Nader Akoury and Natalie N. Bazarova},
year={2026},
eprint={2605.03287},
archivePrefix={arXiv},
primaryClass={cs.HC},
url={https://arxiv.org/abs/2605.03287}
}