MiroFish
Multi-agent social simulation that turns source documents into forecasts
What it is
MiroFish takes seed material such as news, policy drafts, reports, or novels, builds a knowledge graph from it, and spawns thousands of LLM agents with personas and memory that interact in a simulated social world. After the run, a report agent writes up predicted outcomes, and you can chat with individual agents or inject new variables mid-simulation. Analysts and curious users run it to rehearse public opinion reactions, policy effects, or story endings before they happen. It runs as a web app with a Python backend, via source or Docker, and needs an OpenAI-compatible LLM API and a Zep key.