OASIS is a scalable, open-source framework for simulating social media platforms (Twitter, Reddit) using LLM-powered agents. The system models up to one million agents with realistic actions (post, like, follow, comment, report, etc.) and supports configurable recommendation systems, agent graphs, and time scaling. It is designed for research into information spread, misinformation, group polarization, and herd behavior, as evidenced by extensive experiment scripts and real-world datasets included in the repository. The architecture follows a PettingZoo-style environment loop with a Platform, SocialAgent, AgentGraph, and RecSys modules, all built on top of the CAMEL framework. While the core is a reusable simulation library, the repo also contains research data, visualization scripts, and one-off experiment configurations, making it a hybrid of software application and research project.