generative_agents

Why we included this project

This is the reference implementation of the Stanford generative agents work, the project that showed how large language models can drive believable, autonomous characters. It ships a full simulation environment, the Smallville town, where each agent keeps a memory stream, reflects on past experiences, and plans daily activities, all coordinated through a backend server. The codebase is a research artifact rather than a turnkey product: you supply an OpenAI API key, run the Django environment server alongside the agent simulation server, and watch the characters interact in the browser. If you are building agent-based simulations or game NPCs, this is the clearest working example of the memory-reflection-planning architecture that many later agent systems borrow from.