MiroFish Brings Swarm Intelligence to Forecasting, Backed by Shanda
# MiroFish Brings Swarm Intelligence to Forecasting, Backed by Shanda *A student-built project leveraging CAMEL-AI's OASIS simulation engine is gaining traction for predicting everything from public opinion shifts to novel endings.* MiroFish, an open-source prediction engine built on multi-agen...
MiroFish Brings Swarm Intelligence to Forecasting, Backed by Shanda
A student-built project leveraging CAMEL-AI's OASIS simulation engine is gaining traction for predicting everything from public opinion shifts to novel endings.
MiroFish, an open-source prediction engine built on multi-agent swarm intelligence, has climbed GitHub's global trending list this month. The project—built by Guo Hangjiang, a senior undergraduate in China—uses thousands of AI agents with distinct personalities and memories to simulate how real-world scenarios might unfold.
The simulation engine powering MiroFish is OASIS (Open Agent Social Interaction Simulations), developed by the CAMEL-AI team. OASIS supports up to one million agents performing 23 social actions—posting, commenting, following, reposting—across platforms modeled on Twitter and Reddit. It's real infrastructure that's been in active development since late 2024.
"Instead of crunching numbers, it simulates the messy, social dynamics of the real world using thousands of AI agents that talk, argue, persuade, and evolve," the MiroFish documentation explains.
The project has attracted investment from Shanda Group founder Chen Tianqiao, according to coverage on Dev.to. MiroFish's team is actively recruiting, with contact email listed at mirofish@shanda.com.
What's actually been demo'd:
Users upload seed material—news articles, reports, or narrative texts—and describe what they want to predict. The system builds a knowledge graph, generates agent personas with unique backgrounds and stances, runs parallel simulations on dual platforms, and produces a prediction report. Users can then chat with individual agents to understand their reasoning or inject new variables to test alternative scenarios.
The tech stack: Python backend, Vue.js frontend, OASIS for simulation, GraphRAG for knowledge graphs, Zep Cloud for agent memory, and support for any OpenAI SDK-compatible LLM (Qwen-plus via Alibaba's Bailian platform recommended).
MiroFish joins a growing set of tools applying large-scale agent simulations to forecasting—a space that includes OASIS itself, Matrix (a social media simulation platform built on OASIS), and various academic research projects using agent-based modeling for predicting social phenomena.
The project is licensed under AGPL-3.0 and available on GitHub.
