MiroFish is a simple and universal swarm intelligence engine that turns real-world seeds such as news, policy drafts, analytical reports, or fiction into a parallel world populated by agents with memory, motives, and social behavior. The product is designed for graph-native predictive simulation, narrative stress testing, report generation, and deep interaction across a synthetic world. According to the site, MiroFish transforms source material into a scenario graph, runs a multi-agent simulation, and produces a reviewable AI forecast report, letting users inject variables, run social evolution, and inspect the likely trajectory before the real world catches up. The workflow follows five steps. Step 01 is Ontology Generation, which turns raw reports, notes, or fiction into structured entities, motives, and factual anchors. Step 02 is Graph Construction, which assembles a living relationship graph that exposes the actors, tensions, and memory structure behind the scenario. Step 03 is Parallel Simulation, where platform-native agents interact across Twitter and Reddit style channels over multiple rounds. Step 04 is Report Generation, which condenses the trajectory into a readable prediction report with key turning points, risks, and confidence signals. Step 05 is Deep Interaction, which lets users interrogate the generated world through ReportAgent or by interviewing individual characters directly. The engine console accepts PDF, MD, Markdown, or TXT files up to 50MB total, either by uploading a file or importing a link. Users then provide a simulation prompt describing how the uploaded material should evolve once agents begin reacting across public platforms, memory layers, and narrative pressure surfaces. The site advertises support for 1M+ parallel agents per run, Twitter and Reddit style simulation surfaces, and a prediction report as the primary output. The current product shell accepts PDF, Markdown, and plain text, and the workflow assumes the file contains enough narrative, analytical, or factual seed material to build a graph and simulation context. Built-in scenarios cover public opinion forecasting, launch reaction, policy impact, brand crisis, finance cases, and literary continuation. Each scenario pairs a typical input with a template and guide. The public opinion scenario models how institutions, media, influencers, and observers reshape the first narrative around an incident. The launch reaction scenario stress-tests product messaging before competitors, users, and commentators interpret the launch. The policy impact scenario inspects how stakeholder groups interpret a draft policy once incentives and compliance pressure collide. The brand crisis scenario shows how a fragile launch or reputational event expands when the public question drifts away from internal intent. The finance case scenario runs a market-facing scenario where management, analysts, and retail narratives react to the same financial signal differently. The literary continuation scenario treats a fictional world as a live graph of motives and memory, then tests how one new event changes the story. The Forecast Explorer shows the output surface before anything is uploaded, with graph, simulation, and report views. The operator readout presents the likely trajectory as an executive summary, compressing the path into a concise readout of risks, key actors, and the evidence line behind the forecast. Reports include an executive summary, predicted developments, major risks, evidence lines, key actors, and a long-form narrative explanation of how the outcome unfolds. A prompt recipe library offers ready-made prompts for public opinion forecast, launch stress test, policy reaction, and narrative continuation. The site also publishes field notes and guides, including how to use MiroFish, what MiroFish is, how MiroFish simulates the future, what multi-agent simulation is, and how to review a MiroFish forecast. The trust and research section states what MiroFish simulates, including actor incentives and motive conflict, narrative spread across platform-style surfaces, and how one reaction changes the next round, while noting what stays human: choosing the scenario boundary, judging whether the graph is missing pressure, and making high-stakes operating calls. The research context references generative agents and social simulation research, scenario planning framed as reviewable evidence, and forecasts treated as inspectable hypotheses rather than certainty. MiroFish is described as an independent simulation product, and its forecasts are exploratory outputs that should be reviewed before operational, financial, or policy decisions.