MiroFish The Engine That Simulates Tomorrow
MiroFish — The Engine That Simulates Tomorrow
A 20-year-old student built an open-source swarm intelligence engine in ten days. Billionaires came knocking within 24 hours. Here is the full story of MiroFish — the AI that builds a living digital world to predict anything.
Born from a Graduation Project
Every decade or so, a piece of software appears that feels genuinely out of its time — too ambitious, too strange, too fast to be real. MiroFish is that kind of project. It did not emerge from a well-funded lab or a team of seasoned engineers. It was built by Guo Hangjiang, a senior undergraduate at Beijing University of Posts and Telecommunications, over the course of a single intense week and a half in early 2026.
The story begins with a predecessor called BettaFish — a multi-agent public opinion analysis tool that Guo built for his graduation project. BettaFish looked backward, mining past data to understand how public sentiment had moved. It climbed to the top of GitHub's trending charts almost immediately upon launch, but the achievement left Guo oddly cold. "After reaching 10k stars," he later said, "I kind of lost the feeling."
The end-point of one fish became the starting-point of the other. If BettaFish could explain the past, what would it take to simulate the future? That question led directly to MiroFish — a tool that does not dig through history but instead constructs a miniature society and watches what unfolds inside it.
A Miniature Society, Not a Statistical Model
Most prediction tools work by feeding numbers into a statistical model and receiving a probability in return. MiroFish takes a fundamentally different approach. Feed it a news article, a financial report, a policy draft — even a work of fiction — and it builds a parallel digital world populated by thousands of autonomous AI agents, each with a distinct personality, memory, and social network.
These agents do not sit still. They post, reply, argue, form coalitions, change their minds, and follow each other — much like real people on social media platforms. Their collective behavior produces emergent patterns that the system then translates into a structured prediction report with confidence scores and signals. The whole process is driven by OASIS, an open-source simulation framework developed by the CAMEL-AI research community that supports up to one million agent interactions.
"You no longer need a large dataset of historical outcomes to build a prediction tool. You can simulate the crowd, inject the variable, and observe the emergent behavior."
The technical ambition of this is significant. MiroFish is not a polished commercial product — it is a proof of concept that demonstrates what becomes possible when multi-agent orchestration, knowledge graph construction, and dynamic world simulation are combined in a single open pipeline. A system of this architectural complexity would have required a funded team and months of engineering just a few years ago.
The Five-Stage Prediction Pipeline
From Intern to CEO in One Night
The project caught the eye of Chen Tianqiao — founder of Shanda Group, one of China's pioneering internet companies that at its peak in 2004 was the country's largest by market capitalisation. Chen was struck by Guo's ability to identify a real problem and build a working solution around it using AI tools. After watching a rough demo video of MiroFish in action, Chen made a decision.
Within 24 hours of submitting his demo, Guo received a commitment of 30 million yuan — approximately $4.1 million USD — from Chen Tianqiao to incubate MiroFish into a fully developed product. The undergraduate student became a CEO before his graduation project was even submitted.
On March 7, 2026, MiroFish reached the top of GitHub's global trending list, accumulating 18,000 stars and nearly 1,900 forks within days. By mid-April, the repository had crossed 53,000 stars and over 4,100 forks — placing it among the most watched AI projects of the year.
What MiroFish Cannot Yet Do
MiroFish is at version 0.1.2 as of March 2026 — early by any measure. For all its technical ambition, it carries limitations that matter and deserve to be stated plainly.
Where MiroFish genuinely shines is in structured qualitative prediction — public opinion analysis, PR modelling, and narrative extrapolation. The automated persona creation and knowledge graph construction from raw text is technically impressive for a project at this stage. Post-simulation querying — talking directly with individual simulated agents after a run — provides a qualitative research tool that no conventional forecasting method can replicate.
Predicting Markets, Elections, and Novels
The use cases that MiroFish has already demonstrated stretch across wildly different domains. In one demo, the system performed a financial market prediction by simulating thousands of investor agents reacting to a macroeconomic report. In another, it produced a political news forecast — modelling how media and public sentiment might shift in response to a policy announcement.
Perhaps the most striking demonstration was literary: MiroFish was fed the first 80 chapters of Dream of the Red Chamber — a sprawling 18th-century Chinese novel — and asked to simulate a possible ending for the sections that history lost. The agents, embodying characters from the novel, played out social dynamics consistent with the book's internal logic and produced a plausible narrative conclusion.
This breadth is intentional. Guo's stated goal is a "simple and universal swarm intelligence engine, predicting anything." The architecture is domain-agnostic — whatever you feed in becomes the world; whatever you ask becomes the experiment. The question of whether those simulations are accurate enough to be reliably useful is one the next twelve months of real-world deployment will begin to answer.
How to Run MiroFish
MiroFish is fully open source and available on GitHub at github.com/666ghj/MiroFish. Getting started requires a Python environment, an LLM API key (compatible with OpenAI, Claude, or similar endpoints), and Docker for the simplest setup path. A single start command brings up the system. The project has Discord support and is actively maintained — commits as recent as April 2026 show ongoing security patches and internationalization improvements.
For developers building in 2026, MiroFish represents more than a single tool. It is evidence that the playbook for AI applications has genuinely expanded. Guo Hangjiang uploaded an engine. What the developer community builds on top of it is the next chapter — and that chapter is just beginning.
"A system of this architectural complexity would have required a funded team and months of engineering a few years ago. Today, one undergraduate built it in ten days."
Comments
Post a Comment