MiroFish
A multi-agent simulation engine that turns source materials into interactive digital worlds, forecasts, and detailed reports.
MiroFish is a multi-agent AI prediction engine that turns seed material from the real world—such as breaking news, policy drafts, financial signals, reports, or stories—into a parallel digital world. It populates that world with agents that have independent personalities, long-term memory, and behavioral logic, then lets them interact and evolve to explore possible outcomes. A user uploads source material and describes the prediction requirement in natural language. The system can build a knowledge graph, extract entities and relationships, generate personas, configure agents, run a parallel simulation, and produce a detailed prediction report. It also supports deep interaction with the simulated world: users can chat with individual agents and interact with the report agent after the simulation. The project presents two sides of the same idea. For decision-makers, it is a sandbox for rehearsing policies and public-relations scenarios; for individual users, it can explore creative or imagined situations, such as alternative story endings. The documented workflow covers graph building, environment setup, simulation, report generation, and interaction. To try it locally, the README documents Node.js 18 or newer, Python 3.11–3.12, and uv, followed by the project setup commands and npm run dev. It exposes a frontend at http://localhost:3000 and a backend at http://localhost:5001; Docker deployment is also documented. The live demo is linked from the repository. MiroFish is open source under AGPL-3.0. The repository is split between a Python backend and a Vue frontend. Its simulations depend on configured LLM and Zep Cloud credentials, so local deployment requires those environment variables; the README does not promise a free or keyless self-hosted setup.