Generative Agents:
Interactive Simulacra
of Human Behavior

Generative agents proposed by this paper extend LLMs (like ChatGPT) with memory, reflection, planing and a sandbox environment.

We're building our Python implementation that allows you to add Generative Agents to your own worlds.

TODO and contributing

We have a #dev chanel on Discord.

Answer questions.

Check out the issues and projects tabs, there are also # TODOs scatered around the code.

• Python 3.10+
• Black
• isort

Parts

• 4.1 Memory represents the most basic type of memory, an observation. Importance calculation is done here.
• 4.2 Reflection a reflection.
• 4.3 Plan a plan.
• 4.1 MemoryStream a stream of memories. Memory retrival is done here.
• 4 Agent
• 5 Sandbox tick, time,
• Figure 2 World
• Figure 2 Area
• Figure 2 Object

Usage

pip install openai

Authentication

Create a openai_secrets.py file and set your key there.

Note https://platform.openai.com/docs/api-reference/authentication

import openai

openai.api_key = 'Your OpenAI API key'