Exploring Simulacra: A Series on Generative Agents

This is the repository accompanying the tutorial series dedicated to exploring the generative agents of Simulacra. This repository offers a simplified version of the original Simulacra code. For an introduction and overview of the tutorials and simulation design, consider starting with this blog post.

Getting Started

Prerequisites

• OpenAI Python Library: Required to interact with LLMs. This tutorial uses GPT via AzureOpenAI Studio, though the prompt function is adaptable for use with other LLMs.

• Access to the Original Repository: Necessary for the environment and agent persona descriptions. If not already cloned, use:

  git clone https://github.com/joonspk-research/generative_agents.git

  Then, copy the required files into your working directory:

  cp ./generative_agents/reverie/backend_server/maze.py .
  cp ./generative_agents/reverie/backend_server/global_methods.py .
  cp ./generative_agents/reverie/backend_server/utils.py .

Running the Simulation

Execute the simulation with the following command:

python main.py

Understanding the Output

The simulation generates various logs for a comprehensive analysis:

• prompts_log.txt: Chronicles all simulation prompts and their responses.
• failsafe_logs.txt: Details instances of prompt failures.
• schedule_logfile.txt: Offers insights into agent scheduling and planning.
• sim_logs.txt: Tracks agent movements within the simulation.