Exploring agent behavior at scale

A lightweight framework for large-scale social experiments with LLM and RL agents — where agents can talk, decide, and be swapped with humans for RLHF-like setups.

Stanley Milgram
Stanley Milgram, social psychologist known for his obedience experiment. Photo: Wikimedia Commons.

Research focus

  • Scalable simulations of communicating agents.
  • Minimal modules for reproducible results.
  • Effects of social parameters on cooperation.

Modules

  • Agent memory & reflection
  • Simple experiment protocols
  • Visualization via Phaser