MiroFish is an AI prediction engine for testing scenarios and predicting what happens next with multi-agent simulation. It is an independently operated hosted service that hosts the open-source MiroFish project, and it supports a workflow from seed material to a structured report. It can start with one free guest simulation and provides tutorials to begin. The service emphasizes simulating scenarios where outcomes depend on reactions, incentives, and narrative shifts.
MiroFish describes its workflow as turning source material into a prediction report, building a GraphRAG knowledge graph, creating agent personas, running interaction rounds, and producing a structured report. It presents a console-like experience that handles setup, simulation, and reports in one conversation, with a focus on inspectable prediction workflows.
The tool supports a hosted plan and aims to cover applications in public opinion, crisis management, policy testing, market narrative research, and creative writing. It mentions an emphasis on graph-grounded context and the use of independent AI agents within a simulated social environment.
It also references a self-hosted deployment path, Docker, and an official app stack with Vue/Vite frontend and Flask backend.