README.md

August 29, 2026 · View on GitHub

$MiroShark — Simulate anything for \$1 in under 10 minutes with 100+ grounded agents. The pipeline flows input → build world → swarm → report. Keywords: multi-agent simulation, social simulation, swarm intelligence, agent-based modeling, LLM agents, prediction market, scenario testing.$

Simulate anything.

Star us ❤️ →

\$1 · per simulation · 10 min · first result · 100+ · grounded agents

Grounded · real personas · Cross-platform · X, Reddit, markets · Cited · real posts & trades

English · 中文 · 日本語 · Français

MiroShark live demo — a user drops in a document, MiroShark builds the world graph, spawns 100+ agents, and they post, argue, and trade in real time while a report is written.

What it does

What MiroShark does, in four steps: (1) You bring a scenario — MiroShark builds the world around it. (2) Hundreds of grounded agents across Twitter, Reddit, and a prediction market, hour by hour. (3) Chat with any agent, drop breaking news mid-run, fork the timeline. (4) Get a report on what happened, citing actual posts and trades.

Get started

git clone https://github.com/aaronjmars/MiroShark.git && cd MiroShark
cp .env.example .env    # paste one OpenRouter key
./miroshark             # deps + Neo4j + servers → http://localhost:3000

How it works

MiroShark overview — information propagates through X (Twitter), herd effects form in Reddit and Polymarket, 100+ agent personas across 3 platforms drive cross-platform dynamics, and a ReAct report agent writes the recap. Five workflow steps.

MiroShark five-phase pipeline: Phase 1 Ontology Generation, Phase 2 Graph Building, Phase 3 Agent Setup, Phase 4 Simulation Execution, Phase 5 Report and Interaction.