OASIS is an open-source framework from the CAMEL-AI community for simulating social media with large language model agents โ at a scale of up to one million agents. Short for Open Agent Social Interaction Simulations, it recreates platforms resembling Twitter/X and Reddit, complete with actions such as posting, following, commenting and recommending, so researchers can study how information spreads, how opinions polarize and how groups behave online. The project has about 5,300 stars on GitHub and is actively developed, with updates in 2026.
Understanding social dynamics usually means either small lab experiments or observational data you cannot experiment on. OASIS offers a third option: a controllable, large-scale simulation where LLM-powered agents with profiles and memories interact on a simulated platform with a real recommendation engine in the loop. Built on the CAMEL multi-agent framework, it models agent behaviors including posting, liking, sharing, following and reporting, and it supports experiments such as information-spread tracing, group polarization and herd effects. The accompanying paper and documentation describe the architecture and example studies, and a dataset of simulation traces is published on Hugging Face. Because the environment is synthetic, researchers can rerun scenarios, vary parameters and test interventions in ways that real platforms never allow.
OASIS is free and open source under the Apache-2.0 license. Your main cost is the LLM inference needed to run large numbers of agents, which scales with the size of your simulation.
OASIS is for computational social scientists, AI researchers and platform-policy teams who want to experiment with social dynamics at realistic scale, and for multi-agent researchers who need a demanding, realistic testbed for agent societies.