tsinghua-fib-lab/AgentSociety
AgentSociety 2 is a modern, LLM-native agent simulation platform designed for social science research and experimental design. It provides a flexible framework for creating and managing intelligent agents in simulated environments.
What it solves
AgentSociety provides a framework for simulating complex social interactions and urban environments using LLM-based agents. It allows researchers to create executable social science experiments, simulating how multiple agents interact within a structured environment to understand human behaviors and societal patterns.
How it works
The system uses a modular architecture where agents are treated as stateless records driven by Ray Tasks for scalable execution. In the modern v2 version, it employs a ServiceProxy to manage environment handles, LLM clients, and tracing. It supports various reasoning patterns—such as CodeGen, ReAct, and Plan-Execute—and integrates the Model Context Protocol (MCP) for tool extensibility. The legacy v1 version focused on city-scale simulations with gRPC-based environment integration.
Who it’s for
It is primarily designed for social science researchers and AI developers who want to conduct large-scale agent-based simulations of urban environments or research workflows.
Highlights
- Scalable Execution: Uses Ray for distributed computing to handle city-scale simulations.
- LLM-Native Design: Specifically built for LLM-driven agents with multiple reasoning routers.
- Research-Specific Skills: Includes built-in capabilities for literature search, hypothesis generation, and paper writing.
- Experiment Replay: Features catalog-driven JSONL replay and distributed tracing powered by DuckDB.
- Modular Environments: Supports hot-pluggable tools and flexible environment components.
Related
- Project
- Project
- Project
- Project
- Project