MiroFish: The FREE AI Prediction Machine

MyraFish, the free AI prediction machine. A 20-year-old student in China just built an AI tool in 10 days. He uploaded it to GitHub. Within days, it was the number one trending repository on the entire platform.

Above OpenAI, above Google, above Microsoft, within 24 hours of a rough demo, a billionaire actually wrote a check for 4.1 million dollars to fund it. The tool is called MyraFish. And once you understand what it actually does, the reaction makes complete sense. So here's what makes it different from anything else you've seen, right?

MyraFish doesn't predict the number or the future by crunching numbers. It doesn't run a statistical model. It doesn't ask an AI chatbot, what do you think will happen? It builds a working digital copy of the world relevant to your question.

So thousands of AI agents, each with their own personality, their own memory, their own opinions, and then it runs them forward. It watches what emerges and then it tells you what's coming. Think about what that means. You want to know how customers will react to a price increase.

Maybe you want to know if a PR crisis is going to blow over or blow up. Maybe you want to know how a new product launch lands before you spend the budget on it. With every other tool, you're guessing. With MyraFish, you're running a rehearsal.

The creator's name is Guo Hangjiang. He's a senior undergraduate student at Beijing University of Posts and Telecommunications. He'd already done this once before. His previous project, Better Fish, hit number one on GitHub trending in late 2024 and collected 20,000 stars in a single week.

So when he dropped MyraFish in March 2026, people paid attention fast. By March 7th, it had 18,000 stars and nearly 1,900 forks within days. It now sits at over 53,000 stars and is still climbing on GitHub. Chen Tianchao, the founder of Shanda Group and formerly the richest man in China, saw the demo and actually committed 30 million yuan, around 4.1 million dollars, in under 24 hours to this.

He promotes a concept he calls the super individual theory. The idea that in the AI era, a single person can create the equivalent of an entire company. Guo went from student intern to CEO overnight, and that's the story here. But it's not the interesting part.

The interesting part is what this tool can actually do for your business right now. Let me explain how it works and I'm going to go deep on this, but I'll try and keep it as simple as it sounds, right? So you basically feed MyraFish a document. It could be like a news article, could be a financial report, could be a competitor's press release.

It could even be a piece of fiction if you're trying to simulate something creative, right? Now MyraFish actually reads it and does something most tools skip entirely. It doesn't just scan the words, it actually builds a knowledge graph from it. A map of who the players are, how they're connected, what pressures exist, what institutions are involved.

And it's a difference between, say, reading a novel and actually understanding the relationships between every character. From that map, it generates thousands of AI agent personas. And each agent gets a unique personality, a backstory, an initial opinion on the topic, and social connections to other AI agents. And then it drops them all into two simulated environments.

One that works like Twitter, you know, fast and loud. And one that works like Reddit, which is slower and more reasoned. The agents start talking to each other. They argue.

They persuade each other. Opinions shift. Coalitions form. Some agents become influential.

Some get ignored. The simulation runs, and what comes out the other side isn't a single prediction. It's a pattern of emergent behavior that maps onto how real groups of real humans actually move. One developer plugged Mirafish into a trading strategy and ran simulations with 2,847 digital humans before every single trade.

Brian Rommel, a researcher known for building extreme-scale AI systems, ran a Mirafish simulation with 500,000 agents. That last number matters, right? Because this isn't capped at a few dozen AI agents in a chatroom. The simulation engine powering Mirafish is called Oasis.

It's built by the Camel AI research team, and it scales to 1 million agents. Peer-reviewed, open-source, right? The infrastructure is real. Now, if you're running a business, an agency, a brand, a content operation, an e-commerce store, here's where this gets practical.

You want to know what happens if you raise your prices, for example, right? You feed in your positioning, your audience data, your competitor landscape. You run the simulation. Mirafish will show you how different customer segments react, which groups push back hardest, and where sentiment stabilizes.

So let's say, for example, you want to know how a piece of content will land before you publish it. Well, you can feed in the topic, feed in the angle and the audience, and then simulate the conversation and see where the objections come in before you go live. Let's say, for example, you want to test a new offer or a new service line. You could simulate the market response before you actually spend anything.

This is a loop businesses never get access to, right? You can test before you act. Mirafish is, at its core, a risk reduction engine for decisions that used to require gut feeling. Now, if you want a 30-day roadmap for implementing tools like Mirafish inside your business, how to set up the simulations, what questions to run first, and how to use the outputs to make better decisions, come join us in the AI Profit Board.

And we've already got 2,800 business owners in there working through exactly this kind of AI automation. We run four live coaching calls every week, daily step-by-step tutorials, and a prompt library for people who want to use AI to get more leads, more customers, and more time back. Specifically around these sort of tools and AI agents, we've built out playbooks for how to use these kinds of systems to test your offers, your pricing, and your content before you put real money behind them. Link in the comments description or go to the AIprofitboard.com to get access.

Now, let's talk about what makes Mirafish technically different from just asking Claude or Chachibity the same question. When you ask a single AI model, how will my customers react to this, you get one answer, one perspective, right? Whatever biases the model was trained with baked into a single output, Mirafish doesn't do that. So it runs thousands of independent agents, each reasoning separately, each influencing each other, each surfacing what emerges from that collective process, right?

The output isn't one opinion, it's basically a pattern. There's a concept researchers call swarm intelligence, and it's the same reason a flock of birds moves the way it does, or, you know, why financial markets sometimes move the way they do. No single bird or trader is coordinating the whole. The pattern comes from the interaction.

Mirafish is an attempt to harness that same emergent logic for prediction. The technical setup is pretty straightforward. You run it locally with Docker. It's a single command.

And then Mirafish recommends starting with fewer than 40 agents and fewer than 50 simulation cycles before scaling up. Both to keep the cost scalable and manageable, but also to validate that your knowledge graph is being built correctly. It runs on any LLM API with an OpenAI compatible format. But it's specifically recommended with Alibaba's Quen Plus model, which is significantly cheaper than GPT-4 and still produces strong outputs, right?

That's worth noting. You can run real meaningful simulations without enormous API costs. And the workflow, once you're set up, is five steps. So you feed in your seed material.

And then Mirafish will build a knowledge graph from it. It generates agent personas from that graph. And then it runs the dual platform simulation, the Twitter-style fast environment and the Reddit-style threaded one running simultaneously. So you capture both fast emotional reactions and slower reasoned debate.

Then a specialized report agent analyzes what happened and writes a structured report based on that. And here's where it gets genuinely useful for ongoing decision making. After the simulation runs, you don't just get a report and walk away. You can interact directly with any individual agent inside the simulation.

So you can ask one of them why they changed their opinion. You can ask the report agent, what would have happened if we'd announced this differently? You can inject a new variable, a competitive move, a price change, an external news event, and rerun to see how the simulation adapts. Mirafish calls this the God's Eye View.

Change the conditions. Watch the world reorganize. Think about how powerful that is for an agency. Let's say, for example, a client wants to launch a new campaign.

Instead of presenting one strategy based on gut and experience, you run three versions through Mirafish. You show the simulated responses side by side and you present the one with the strongest predicted outcome. The work becomes demonstrably more rigorous. The client conversation changes entirely.

Or for e-commerce, for example, let's say you're about to run a sale. How deep should the discount go? Well, what does a 10% discount do to perceived brand value versus a 20% one? Which segment of your audience responds more to scarcity messaging versus value messaging?

These are questions where most brands are flying blind. But Mirafish gives you a way to test them before you commit. Now, what are the honest limitations here? Well, Mirafish is at version 0.1.2.

It dropped in December 2025. So this is early software. The architecture itself is strong and the simulation engine underneath it, Oasis, is peer-reviewed research. But the English documentation is still maturing.

The community is still forming and it's mostly concentrated in China right now. There are no published benchmarks comparing Mirafish's predictions against historical outcomes at scale. And the demos are compelling demonstrations of the approach, but not evidence of accuracy across all use cases. So where does it perform best right now?

Public opinion and sentiment analysis. So PR and crisis scenario planning, offers and pricing testing, content strategy validation, policy impact forecasting. Anywhere where the outcome depends heavily on social dynamics and human behavior, rather than pure quantitative modeling. Mirafish is genuinely useful in ways that single model AI tools aren't.

The founder hasn't abandoned the project, so it's actively maintained. It commits as recent as April this month. And the roadmap includes expanded LLM provider support, better visualization of agent interactions, and performance improvements to support simulations with over 100,000 agents. This thing is moving fast, right?

But here's a wider pattern worth understanding. A year ago, running a simulation with thousands of autonomous agents required a research team, serious compute infrastructure, and months of setup. Today, a student built it in 10 days using off-the-shelf components. LLM for agent reasoning, GraphRag for knowledge grounding, CloudMemory for persistence.

And the result is something functionally novel that a business can deploy locally in an afternoon. The components matured to a point where something like Mirafish became buildable. And once it becomes buildable, someone built it. That's the trajectory we want.

The tools that used to require enterprise infrastructure are becoming things that one person can set up and run. The ceiling on what a small team can test, model, and predict is rising. And the businesses that learn how to use these tools early are going to make better decisions than the ones still relying on intuition and A-B tests. The question isn't whether prediction tools like Mirafish are going to become a normal part of how businesses operate.

They are. The question is whether you're learning how to use them before your competitors do or catching up later. If you want to get ahead of this, inside the AR Profitable Boardroom, we've put together daily tutorials on exactly how to use tools like Mirafish and other AI agent platforms to grow your business. Four weekly live coaching calls where you can ask questions, get your specific setup reviewed, and learn from 2,800 other business owners who are actively automating with AI.

We've built out 30 day roadmaps specifically for agency owners, e-commerce operators, and freelancers, covering how to use AI prediction and simulation tools to test offers, validate content, and get more customers before spending budget on anything. Plus, a prompt library and a local member map so you can connect with people near you who are running the same tools. Link in the comments description or go to the arprofitboardroom.com to check that out. Mirafish is open source, free to run.

The GitHub repo is available at Mirafish. Start small, fewer than 40 agents, a simple scenario you already know the answer to. For example, like a past campaign, a product launch that you already ran, that could be fun. And then you use it to calibrate, see how close the simulation gets to what actually happened.

And once you trust the output, you can scale it up. A 20 year old student built a prediction engine in 10 days. A billionaire funded it in 24 hours. It's already got 53,000 stars on GitHub.

The people paying attention to this space are moving. The only question left is whether you're one of them.