This repository provides the core simulation module for Generative Agents, computational entities that exhibit believable human behaviors within an interactive game environment. It is designed for researchers and developers interested in simulating complex social dynamics and emergent behaviors driven by large language models.

How It Works

The system simulates agents powered by large language models (LLMs) that generate behaviors based on their internal memory and environmental context. Agents interact within a simulated environment, making decisions and taking actions that influence their state and the state of the world. The approach leverages LLMs to create emergent, human-like behaviors, moving beyond scripted interactions to dynamic, context-aware simulations.

Quick Start & Requirements

• Installation: Requires generating a utils.py file with your OpenAI API key in reverie/backend_server/. Install dependencies via requirements.txt.
• Prerequisites: Python 3.9.12 tested. Requires an OpenAI API key.
• Setup: Two servers (environment and simulation) must be run concurrently. The environment server is a Django project.
• Links: Paper, Demo (local setup required).

Highlighted Details

• Simulates emergent human behaviors using LLMs.
• Supports agent memory, planning, and interaction.
• Allows customization of agent initial states via history files.
• Includes replay and demo functionalities for visualization.

Maintenance & Community

The project is associated with research from Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Citation is requested via the provided BibTeX entry.

Licensing & Compatibility

The repository's license is not explicitly stated in the README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The simulation can be costly due to OpenAI API usage. OpenAI API rate limits may cause hangs, requiring simulation restarts. The replay function is primarily for debugging and does not optimize visuals. Browser compatibility notes suggest Chrome or Safari are preferred over Firefox for the frontend.