Skip to content

AgentSociety

Simulate human behaviour and social interaction with LLM agents

Visit official site
Category Agents

What AgentSociety is

The framework is described as a project of the FIB lab at Tsinghua University that uses LLM-driven agents to simulate human behaviour and social interaction. The material states it draws on the planning, memory and reasoning abilities of large language models to produce behaviour patterns, with environments designed from datasets, text and rules.

What you can do with it

  • Simulate agent behaviour with classic theories such as Maslow's hierarchy of needs
  • Design environments from datasets, text and rules
  • Monitor and interact with agents through a real-time visual interface
  • Run social experiments with interview, survey, intervention and metric recording tools
  • Use point-to-point and group communication between agents

Application scenarios

  • Study how urban traffic congestion affects resident behaviour
  • Simulate consumption and employment behaviour under different policies
  • Simulate student learning behaviour and social interaction in an education setting

Getting started

  • Install the package with pip install agentsociety
  • Configure agent and environment parameters and define the task goal
  • Launch the framework and run the agent simulation
  • Monitor agent behaviour and environment state in the visual interface

Pros & cons

✓ What we like

  • Point-to-point and group communication between multiple agents
  • Compatibility with several LLM models such as OpenAI and Qwen

! What to watch out for

  • Simulation is driven by the planning, memory and reasoning of LLMs
  • Environment design is built from datasets, text and rules

FAQ

Who develops it?

The material names the FIB lab at Tsinghua University as the developer.

What models are compatible?

The feature list names OpenAI and Qwen among the compatible LLM models.

Last reviewed: 2026-11-15

More AI office assistant tools

View all →

How we review