Generative Agents

Generative Agents is the research implementation behind a simulation of computational agents that exhibit believable human-like behavior. It combines a Python-based agent simulation with a browser-accessible game environment representing a small virtual town. Agents operate through stored memories and language-model-driven behavior while moving and interacting inside the environment. Users can start new simulations, advance them by configurable time steps, and save their progress for later use. Completed simulations can be replayed or prepared as visual demonstrations at different playback speeds. The framework also supports custom agent histories and modified base simulations for experimenting with different populations and scenarios.