What Is MiroFish? The New AI Engine That Uses Thousands of Agents to Model Human Behaviour

A new open source artificial intelligence project called MiroFish is gaining attention in the global technology community for its ability to simulate human behaviour and market dynamics using thousands of digital agents.

The platform has rapidly gained popularity on GitHub, where developers have given it thousands of stars.

According to reports, the project was created by a young Chinese programmer and has received support from Chen Tianqiao, the founder of Shanda Group.

Developers describe MiroFish as a swarm intelligence engine designed for large scale simulations.

How the MiroFish System Works

Unlike traditional predictive models, MiroFish builds a digital environment populated by thousands of independent AI agents.

These agents interact with one another simultaneously, creating a complex network of behaviours and responses.

The system first gathers real world data from sources such as news reports, financial documents, policy papers and social media discussions.

This information is converted into a structured knowledge graph that maps relationships between individuals, organisations and events. The graph then forms the foundation for the simulated environment in which the AI agents operate.

Modelling Collective Social Behaviour

Inside the simulation, each AI agent is assigned its own behavioural profile, memory and decision making process.

Agents communicate with each other, react to new information and influence decisions within the system. Over time, these interactions produce patterns that resemble collective social behaviour.

Supporters of the project say the approach focuses on modelling how groups of people respond to events rather than relying solely on statistical prediction models.

Technical Structure of the Platform

MiroFish runs on a multi agent architecture. The backend is built with Python, while the visual interface is developed using Vue.js to allow users to monitor interactions between agents.

The platform also uses GraphRAG, a method that organises information into connected entities and relationships instead of treating documents as separate text blocks.

This approach allows agents to analyse complex networks such as economic relationships, influence patterns and social group dynamics.

Flexible Deployment and AI Integration

The system integrates with the Zep platform to maintain long term memory for agents. This enables agents to store and retrieve experiences across multiple simulation rounds, allowing behaviours to evolve over time.

According to project documentation, MiroFish can run locally on individual systems or through container environments such as Docker.

The platform also supports integration with multiple large language models compatible with the OpenAI API framework.

Potential Applications of the Technology

MiroFish can generate interactive analytical reports based on how agents behave during simulations. Users can modify parameters and observe how the virtual environment responds to different conditions.

The technology may be used for analysing market sentiment, studying public opinion, testing policy responses and exploring narrative scenarios.

However, researchers behind the project emphasise that the system is intended for scenario exploration rather than precise forecasting.