<- all tokdocs

MiroFish: Is the Swarm Prediction Engine Free, and Who Uses It?

By , founder of Mavric and Apso

View on TikTok ->

MiroFish is an open-source "swarm intelligence" engine that simulates thousands of AI agents reacting to an event, a policy or a product. The code is free under AGPL-3.0, but running it requires a per-token LLM API key, and the hosted product was a waitlist at the time of the video. The "72,000 people" using it is a GitHub star count.

What is MiroFish?

MiroFish lives at github.com/666ghj/MiroFish. Its demo site headlines "Predict Anything" over "The World's First Simple and Universal Open Source Swarm Intelligence Engine." You feed it seed material, it spins up agents with long-term memory, runs them through rounds of interaction and generates a report on how opinion moved.

The simulation core comes from elsewhere. The README credits OASIS (Open Agent Social Interaction Simulations) from CAMEL-AI as the engine. OASIS is Apache-2.0 licensed, was created 2024-11-13 and has 5,240 stars. MiroFish adds GraphRAG seed extraction, a web frontend and report generation on top.

Do 72,000 people use MiroFish?

A TikTok from @ai.amirtech says "72,000 people have already started using it." That number tracks GitHub stars. As of 2026-10-08 the GitHub API reported 77,118 stars, 11,788 forks, 460 watchers and 136 open issues.

A star is a bookmark. It records a click on a button. It says nothing about who cloned the project, installed Python 3.11, supplied two API keys and ran a simulation. The project's growth has been steep: panews reported 18,000 stars within days of 7 March 2026, when MiroFish hit number one on GitHub trending. That pace makes stars a weak stand-in for adoption.

Is MiroFish free?

The source is free. The compute is not. The README's Quick Start requires an LLM_API_KEY for any OpenAI-SDK-compatible endpoint and recommends qwen-plus through Alibaba's Bailian console.

Alibaba's Model Studio page for qwen-plus lists a 1,000,000-token context window and per-token pricing: "Input: 0.115 Per 1M tokens, Output: 0.287 Per 1M tokens" for inputs under 128k in the China (Beijing) region, rising to 0.689 per 1M input tokens between 256k and 1M. A simulation that runs thousands of agents through multiple rounds uses a lot of tokens by design.

The README also requires a Zep Cloud key for memory and says Zep's "free monthly quota is sufficient for simple usage." That is the project's own description, not a measured figure for a large run.

The AGPL-3.0 license adds an obligation the video skips. If you modify MiroFish and offer it over a network, you must make your modified source available to your users.

Is there a hosted version of MiroFish?

Not yet, at the time of the video. The project's demo site reads "Join the waitlist to get notified at launch!" and its code includes a full waitlist form. The repository's homepage field points to mirofish.ai, and the only web destination linked in the README is the github.io demo. Other MiroFish-branded sites selling monthly plans are not linked from the project.

That leaves two options: self-host the code with your own keys, or join the list.

What is MiroFish used for?

The video pitches it for "new product launches, pricing tests, and ad campaigns." The demo site lists six use cases, each with an example:

  • Financial Decision Support: "Market sentiment simulation after central bank rate hikes"
  • Policy Opinion Forecasting: "Carbon tax public acceptance simulation"
  • Crisis PR Simulation: "Product recall sentiment simulation"
  • Marketing Strategy Testing: "New product launch social media reaction prediction"
  • Story and Fiction Deduction: "Predicting the lost ending of Dream of the Red Chamber"
  • Academic Research Support

Only one of the six matches the video's pitch. The README's two featured demos are a Wuhan University public opinion forecast and a prediction of the lost ending of Dream of the Red Chamber. "Pricing tests" does not appear in the README or on the site.

Who built MiroFish?

The video says a "20-year-old student in Beijing built this in just 10 days," and that holds up. panews names Guo Hangjiang, 20, a senior at Beijing University of Posts and Telecommunications, who built the first version in 10 days.

36Kr reports that Chen Tianqiao of Shanda Group committed 30 million RMB, about 4.1 million USD, within 24 hours of seeing a demo video. The 666ghj GitHub profile lists the name "BaiFu," company "USTC, BUPT," and location "Shanghai, China." The README's recruiting address is mirofish@shanda.com. The video does not mention that a funded company sits behind the free repository.

The 10 days describe the first push. The repo was created on 2025-11-26, took its AGPL license on 2025-12-19, and shipped v0.1.0 (2025-12-22), v0.1.1 (2026-01-22) and v0.1.2 (2026-03-07), with commits through 2026-10-01.

Bottom line

MiroFish is a real, actively developed open-source project with a funded company behind it. The 72,000 figure counts stars, not users. Running it costs per-token LLM fees, the hosted version was a waitlist, and marketing is one of six intended uses.

Sources

Published October 7, 2026. Based on a TikTok by @ai.amirtech. Full video notes (markdown)