666ghj/MiroFish

What it solves

MiroFish is a prediction engine that allows users to simulate future trajectories of real-world events or fictional scenarios. It solves the difficulty of predicting complex social outcomes—such as public opinion shifts, policy impacts, or story endings—by creating a high-fidelity digital sandbox where thousands of autonomous agents interact and evolve socially.

How it works

The system takes seed materials (like news reports or stories) and natural language prediction requirements as input. It then follows a five-step workflow:

1. Graph Building: Extracts seed information to build a GraphRAG structure with individual and collective memory.
2. Environment Setup: Generates personas and configures agent relationships.
3. Simulation: Runs parallel simulations where agents interact and update their temporal memory.
4. Report Generation: A specialized ReportAgent analyzes the simulation to produce a detailed prediction report.
5. Deep Interaction: Users can chat directly with agents in the simulated world or the ReportAgent to refine their understanding.

Who it’s for

• Decision-makers: Those needing a risk-free laboratory to test policies and public relations strategies.
• Creative users: Individuals wanting to deduce novel endings or explore "what if" imaginative scenarios.

Highlights

• Multi-agent Swarm Intelligence: Uses thousands of agents with independent personalities and long-term memory to mirror reality.
• Dynamic Variable Injection: Allows users to inject variables from a "God's-eye view" to change simulation trajectories.
• GraphRAG Integration: Combines knowledge graphs with retrieval-augmented generation for high-fidelity world construction.
• Interactive Sandbox: Provides a digital world where users can interact with simulated entities post-simulation.