Can I run the simulation without an OpenAI key?

No, the agents rely on the OpenAI API for language generation.

How long does a typical simulation take?

Each game step represents ten seconds of simulated time; longer runs require more API calls and compute.

Is the project actively maintained?

It is released as accompanying code for a research paper, so updates depend on the authors.

generativeagents

Verified

Simulate believable human behaviors using generative agents in a virtual town.

Autonomous AgentsResearch 21.5kOpen source

View on GitHub

Updated 2026-06-15

What is generativeagents?

Generative Agents creates computational characters that exhibit human-like routines, conversations, and decision-making within a shared sandbox world. The system combines language models with memory and planning modules so agents can reflect on past events and coordinate actions without direct scripting.

It operates by running two concurrent servers: one for the visual environment and another for the agent logic. Agents update their states in discrete time steps, allowing emergent social patterns to appear as the simulation progresses.

The toolkit targets researchers and developers exploring social simulation, AI behavior modeling, or interactive storytelling who need a reproducible starting point for multi-agent experiments.

Capabilities

simulate believable human behaviors

run interactive multi-agent simulations

integrate with openai api

manage game environments

replay simulation animations

What you can build with generativeagents

Social Behavior Research

Run controlled experiments to study how agents form relationships or spread information across a community.

Interactive Storytelling

Generate dynamic narratives by letting agents pursue individual goals that intersect in unexpected ways.

Agent Architecture Testing

Prototype new memory or planning components by extending the existing simulation loop with custom logic.

Install generativeagents

1Create a utils.py file in reverie/backend_server with your OpenAI API key and required paths.
2Install packages from the requirements.txt file, preferably inside a virtual environment.
3Start the Django environment server with python manage.py runserver in the frontend_server directory.
4Launch the simulation server by running python reverie.py and enter the base simulation name followed by a new simulation name.
5Open the simulator interface in a browser and issue run commands with desired step counts to advance the agents.

Works with

OpenAI APIPythonDjango