MiroFish: Predicting the future through swarm intelligence
A groundbreaking multi-agent simulation engine that builds parallel digital worlds to forecast social dynamics, policy impacts, and complex system behavior.
What if you could see the future before it happens? Not through crystal balls or fortune telling, but by creating a digital twin of reality where thousands of AI agents interact, evolve, and reveal how complex systems actually behave.
That's exactly what MiroFish does.
A new approach to prediction
MiroFish is an open-source swarm intelligence engine that takes a radically different approach to forecasting. Instead of relying on statistical models or historical data alone, it constructs high-fidelity parallel digital worlds populated by autonomous agents.
Here's how it works:
- Seed extraction — You provide real-world information: breaking news, policy drafts, financial signals, or even novel storylines
- World building — The system automatically constructs a digital environment with agents that have independent personalities, long-term memory, and behavioral logic
- Simulation — Thousands of agents interact freely, their collective behavior emerging from individual decisions
- Prediction — The system generates detailed reports and an interactive digital world you can explore
Think of it as a sandbox where you can test "what if" scenarios without real-world consequences.
From macro policy to micro creativity
What makes MiroFish particularly interesting is its versatility. The same technology serves wildly different use cases:
For decision makers, it's a pre-flight simulator for policies and public relations campaigns. Test your strategy in a risk-free environment before committing resources.
For researchers, it's a laboratory for studying emergent social behavior and complex system dynamics.
For creatives, it's a narrative playground. The project showcases this with a fascinating example: using the first 80 chapters of Dream of the Red Chamber to predict the lost ending of this classic Chinese novel.
The project includes live demos of real-world applications, including public opinion analysis of the Wuhan University incident and literary prediction using classical literature.
Technical foundation
MiroFish is built on solid technical foundations:
- Multi-agent architecture powered by the OASIS simulation engine from CAMEL-AI
- Knowledge graph construction using GraphRAG for structured memory
- Dual-platform simulation for parallel processing and validation
- Tool-rich ReportAgent that can deeply interact with the simulated environment
- Temporal memory that updates dynamically as the simulation progresses
The stack combines Python (57.8%) for the backend simulation engine and Vue (41.1%) for the interactive frontend interface.
Open source and accessible
The project is released under AGPL-3.0 license and has already gained significant traction:
- 33.3k stars on GitHub
- 4.2k forks showing developer interest
- Active development with 219 commits and recent updates
- Comprehensive deployment options including source code and Docker
Getting started requires:
- Node.js 18+ for the frontend
- Python 3.11-3.12 for the backend
- Access to OpenAI-compatible LLM APIs (they recommend Alibaba's Qwen-plus)
- Zep Cloud for agent memory management (free tier available)
The bigger picture
MiroFish represents something important in the evolution of AI applications. While much attention goes to chatbots and image generators, this project explores a different frontier: using AI to understand and predict complex social systems.
The implications are significant:
- Policy testing before implementation
- Risk assessment through simulation rather than trial-and-error
- Scenario planning with unprecedented fidelity
- Social dynamics research with controllable variables
- Creative exploration of narrative possibilities
Backed by serious players
MiroFish has strategic support from Shanda Group, a major Chinese technology conglomerate. The project is actively hiring for full-time and internship positions, suggesting this is more than a research prototype — it's a platform with commercial ambitions.
The simulation engine is powered by OASIS from the CAMEL-AI team, demonstrating how open-source AI research can enable entirely new categories of applications.
Try it yourself
The project offers an online demo where you can experience a simulation of public opinion dynamics around a real news event. The full source code is available on GitHub with detailed documentation in both English and Chinese.
Whether you're interested in policy simulation, social dynamics, financial forecasting, or just exploring what's possible with multi-agent systems, MiroFish offers a glimpse into a future where we can test reality before we live it.
The question isn't whether we can predict the future. It's whether we're ready to simulate it.
Explore MiroFish: GitHub Repository | Official Website | Live Demo


