Polis-Darwin: Emergent Governance Through Multi-Agent Evolution

Polis-Darwin functions as a living laboratory for multi-agent evolution, not a scripted simulation. It leverages LangGraph to create a customizable micro-society with five distinct Large Language Model agents that freely communicate, cooperate, or compete, allowing users to watch how governance rules and social norms emerge from the bottom up.

Its primary use case is rapid experimentation: researchers can modify agent personalities, memory modules, and interaction protocols to run controlled experiments on collective decision-making. Built in TypeScript, the codebase is accessible for further development and integration with various LLM backends.

The key advantage is its ability to produce complex, unpredictable emergent behavior by combining generative agents with long-term memory, offering a unique window into multi-agent alignment and consensus formation. However, as an early-stage demo with only 152 GitHub stars and limited documentation, the simulation is confined to five agents in a small-scale environment; result stability and reproducibility are not yet proven, and significant customization is necessary before any research-grade or production use.