# WorldSeed

A multi-agent world engine where defined roles, rules and consequences let AI agents interact until useful behavior emerges, usable for research simulations, games and fictional scenarios.

WorldSeed is a scene-agnostic engine for running multi-agent simulations: a user defines roles, rules, private information, actions and consequences, and agents interact until useful artifacts emerge, with the option to watch from above, intervene, or step into a character directly. The same engine runs research simulations, production rooms, games and fictional scenarios without changing its core.

Example scenes include an autoresearch scenario where a cohort of specialist agents proposes hypotheses, runs experiments and peer-reviews each other's results toward a machine-learning objective; an AI tool pilot lab where builder agents create demos that critic and audience agents judge; and a workplace scenario exploring how agents respond under pressure. The project reports emergent behaviors, such as an agent drifting into a teammate's area of responsibility, that were not specified in its configuration.

Website: https://worldseed.morphmind.ai/demo