GPTeam is an open-source multi-agent simulation project that aims to test the capabilities of Large Language Models (LLMs) to emulate human-like social behavior. The project uses a customizable architecture, allowing users to create their own simulations with unique personalities, memories, and directives. Each agent in the simulation has its own observe, plan, react, act, and reflect processes, which are inspired by the "Generative Agents" paper from Stanford University. The goal of GPTeam is to achieve human-like behavior through a combination of long-term memory systems and self-reflection. The project's results show complex social behavior among agents, coordinating with each other and playing off dialog appropriately. However, there are limitations to the current setup, such as productivity loss due to reliance on dialog, and opportunities for future improvements, including faster language models, more accessible interfaces, and interactive features.