Mirofish

Open source swarm intelligence engine for simulating and predicting outcomes

Advanced
Screenshot of Mirofish, Open source swarm intelligence engine for simulating and predicting outcomes

What is Mirofish?

Mirofish is an academic research tool. An open source multi-agent simulation engine that builds a parallel digital world from seed data, lets thousands of persona-driven agents interact, and returns a prediction report plus a world you can question directly. Positioned for finance, policy, PR, marketing, research and fiction deduction, with a hosted online edition still on the waitlist.

MiroFish is an open source swarm intelligence engine built around multi-agent simulation. Rather than extrapolating trends from historical data alone, it takes seed materials from the real world (breaking news, draft policies, financial signals, novel chapters) and builds a parallel digital world inside which thousands of agents with independent personas, long term memory and behavioral logic interact and evolve socially. The premise is that group behavior emerges from individual interaction, and that rehearsing that emergence in a sandbox gives a better read on what happens next than a single statistical model can. In practice the workflow is a five stage pipeline. Graph construction extracts the reality seed and injects individual and collective memory into a GraphRAG structure. Environment setup pulls out entity relationships, generates personas and configures simulation parameters. The simulation then runs on two platforms in parallel, auto-parsing your prediction request while temporal memory updates on the fly. A ReportAgent equipped with a toolset examines the post-simulation environment and produces a prediction report. Finally you can interrogate the result directly by chatting with any individual agent in the simulated world or with the ReportAgent itself. The intended audience is broad by design. The site positions it for financial decision support and investor sentiment rehearsal, policy opinion forecasting for government and think tank users, crisis PR simulation for enterprises, marketing strategy testing, academic research on information propagation, and story or fiction deduction, including a demo that predicts a lost ending for Dream of the Red Chamber from its first eighty chapters. That spread is the product's stated thesis: the same simulation machinery serves serious forecasting and playful what-if exploration. Where it fits among alternatives is less clear from the homepage. MiroFish is not an LLM wrapper or a chatbot with search; it sits closer to agent based modeling tools, which usually demand programming and domain modeling expertise. MiroFish claims end-to-end automation and natural language input for that class of work. The catch is maturity: the project is open source and self described as heading toward an online edition that has not launched, with a waitlist in place, so most users will meet it as a repository rather than a hosted service.

How do you use Mirofish?

  1. 1Check the live demo and repository
    Start on the homepage to try the live demo and follow the GitHub link. The repository is where the open source engine actually lives today, since the hosted edition is still gathering waitlist signups.
  2. 2Watch the demo videos
    Two walkthroughs are embedded: Wuhan University opinion prediction, which also covers the technical architecture, and a Dream of the Red Chamber ending prediction built from the first eighty chapters.
  3. 3Prepare seed material and a prediction request
    Gather the inputs the simulation will be built from, such as a data analysis report, a news story, a draft policy or a set of novel chapters. Then write your prediction need in plain natural language.
  4. 4Build the graph and configure the environment
    Run graph construction to extract the reality seed and inject memory, then let environment setup extract entity relationships, generate personas and set the simulation parameters.
  5. 5Launch the simulation and read the report
    Start the parallel simulation, which parses the prediction request and updates temporal memory as it runs. The ReportAgent then works through the resulting environment and returns the prediction report.
  6. 6Interrogate the simulated world
    Use deep interaction to chat with any agent in the world or with the ReportAgent, and adjust variables from the god's-eye view to test alternative trajectories before settling on a conclusion.

Pros and cons

Pros

  • Open source engine with a public GitHub repository you can inspect and self-hostAI
  • Five-step pipeline covers seed extraction, persona generation, simulation and reporting without manual modelingAI
  • Dual-platform parallel simulation with dynamic temporal memory updates during a runAI
  • Post-simulation interaction lets you question individual agents as well as the ReportAgentAI
  • Demo use cases span finance, policy, crisis PR, marketing, research and fiction deductionAI

Cons

  • The hosted online edition has not launched; the waitlist signals an unfinished core productAI
  • No pricing, plan tiers or usage limits are published anywhere on the homepageAI
  • Predictive validity depends entirely on the seed data and persona setup, with no accuracy benchmarks shownAI
  • Multi-agent simulation is computationally heavy, and no runtime or infrastructure requirements are statedAI
  • Getting a meaningful result requires domain framing and tuning, despite the simple five-step framingAI

How much does Mirofish cost?

Features

Five-stage prediction pipeline: GraphRAG based graph construction with memory injection, automated environment setup with persona generation, dual-platform parallel simulation with dynamic temporal memory, ReportAgent driven report generation, and deep interaction with any simulated agent. Seed input is natural language, and variables can be injected from a god's-eye view to remap future trajectories.

Frequently asked questions about Mirofish

  • How do you use Mirofish?
    The walkthrough on this page covers 6 steps: 1. Check the live demo and repository 2. Watch the demo videos 3. Prepare seed material and a prediction request 4. Build the graph and configure the environment 5. Launch the simulation and read the report 6. Interrogate the simulated world.
  • What platforms does Mirofish support?
    Mirofish is available on Web App.
  • What are the limitations of Mirofish?
    The hosted online edition has not launched; the waitlist signals an unfinished core product. No pricing, plan tiers or usage limits are published anywhere on the homepage. Predictive validity depends entirely on the seed data and persona setup, with no accuracy benchmarks shown.

Status

StatusActive
Views0
Outbound clicks0
Added10/21/2026

Platforms

Web App

Pricing

Free