Mental BoundMiroFish: predicting the future with swarm intelligence
An open-source multi-agent engine that builds parallel digital worlds — thousands of agents — to forecast policy impact and social dynamics.
MiroFish takes a different approach to forecasting: instead of statistical models over historical data, it constructs a high-fidelity digital twin of a situation and lets it play out. You seed it with real-world information — breaking news, a policy draft, financial signals. It builds a world populated by autonomous agents with independent personalities, long-term memory, and behavioral logic. Thousands of them interact freely, and the system reports what emerges, alongside an interactive world you can explore.
The use cases span further than you'd expect. For decision makers it's a pre-flight simulator — test a policy or a PR campaign before committing resources. For researchers it's a laboratory for emergent social behavior with controllable variables. And for creatives it's a narrative playground: the project's showcase demo feeds in the first 80 chapters of Dream of the Red Chamber and simulates the lost ending of the classic Chinese novel.
Under the hood it's a serious stack: the multi-agent simulation runs on OASIS from the CAMEL-AI team, knowledge graphs are built with GraphRAG, memory runs on Zep, and the whole thing ships as Python backend plus Vue frontend under AGPL-3.0. Traction is real — over 33,000 GitHub stars, backing from Shanda Group, and active hiring, which suggests platform ambitions rather than a research prototype.
The reason we're flagging it: while attention concentrates on chatbots and image generators, this is a different application class — AI for understanding and predicting complex social systems. Testing reality before living it changes how policy, risk, and scenario planning get done. The engine is open source and the live demo takes minutes to try.