MiroFish

What it is

An open-source multi-agent simulation engine. Give it seed materials and a question: its documented workflow builds a knowledge graph, creates agent personas, runs a simulation and generates a report. You can then interact with simulated agents and the report agent.

Setup

1. Start with the official demo to explore the prepared example before setting up your own instance.

2. For local setup, use Node.js 18+, Python 3.11 or 3.12, Git and uv. Clone the repository and enter its folder.

   git clone https://github.com/666ghj/MiroFish.gitCopy
   cd MiroFishCopy

3. Copy .env.example to .env. Set LLM_API_KEY, LLM_BASE_URL, LLM_MODEL_NAME and ZEP_API_KEY for your providers. Keep your keys private.

4. From the project root, install the frontend and backend dependencies, then start both services.

   npm run setup:allCopy
   npm run devCopy

5. Open localhost:3000 in your browser; the backend API runs on port 5001. Follow the app workflow with your seed materials and scenario question.

Cheat sheet

• Try the official demo: the project's prepared simulation example.
• Repository and current setup instructions.
• Backend requirements: Python 3.11-3.12 and the simulation dependencies.
• Commands and Node.js requirements: run setup and development commands from the repository root.
• LLM API and Zep Cloud credentials are required by the documented local setup. Check their usage charges; an open-source license does not make hosted services free.

Axo's take

AXO

Try it if you want to explore how a scenario could unfold with interacting agents. Treat the report as a simulation shaped by its inputs and model assumptions, then compare it with real evidence. Axo has not independently tested its predictive accuracy.