OASIS is a scalable agent-based simulation framework that enables researchers to model social media dynamics with up to one million large language model agents.
The tool addresses the challenge of studying complex social phenomena at scale by simulating realistic user behavior on platforms like Twitter and Reddit. It incorporates LLM agents with diverse action spaces—supporting 23 distinct actions including following, commenting, and reposting—to create authentic social interactions. The framework includes built-in recommendation systems using both interest-based and hot-score-based algorithms to mirror how users discover and engage with content in real social networks.
OASIS suits researchers investigating information spread, group polarization, herd behavior, and other emergent social dynamics. It is designed for those who need to study social media phenomena at scales impractical to observe directly in production systems. The tool's dynamic environment adaptation allows simulations to reflect real-time changes in social networks and content, making it appropriate for exploring how platform mechanics influence user behavior and collective outcomes.
The project maintains active development with regular updates to its simulation capabilities and agent behavior models. Documentation is comprehensive and accessible through an official site, supporting users from initial setup through advanced configuration. The team provides example implementations and maintains a public dataset derived from simulations, enabling reproducibility and comparative research across different experimental designs.
