MiroFish - An open-source AI prediction engine that automatically builds high-fidelity parallel digital worlds. - AiBoss

What is MiroFish?

MiroFish is an AI prediction engine based on multi-agent technology, developed by BaiFu, a 20-year-old student at the University of Science and Technology of China. MiroFish can automatically construct a high-fidelity parallel digital world from real-world seed information (news, novels, policies, etc.), generating thousands of intelligent agents with independent personalities, long-term memories, and behavioral logic to freely interact and socially evolve. Users can dynamically inject variables to accurately predict future trends, making it suitable for scenarios such as predicting major events, analyzing strategic trends, and writing continuations of literary endings.

MiroFish's main functions

Intelligent Predictive InferenceUsers upload seed materials (such as news, novels, and policy documents) and describe their prediction needs in natural language. MiroFish can automatically build a high-fidelity parallel digital world, deduce the future trend of events, and generate detailed prediction reports.
Multi-Agent Social SimulationThe system generates thousands of intelligent agents with independent personalities, long-term memories, and behavioral logic based on seed information, allowing them to interact freely and evolve socially in virtual space, simulating the dynamics of real-world groups.
Dynamic variable injectionUsers can inject new variables or adjust parameters in real time during the simulation from a "God's-eye view" to accurately observe the impact of different conditions on the development path of events and achieve zero-risk decision-making simulation.
Deep Interactive ExplorationAfter the simulation is completed, users can have a one-on-one conversation with any intelligent agent in the digital world, or consult with a professional ReportAgent to gain a deeper understanding of the simulation details and logical basis.
Visual report generationThe system automatically integrates the results of multiple simulations and outputs a structured analysis report that includes the event context, key nodes, trend judgments, and risk warnings to help users make informed decisions.

MiroFish's technical principles

Map ConstructionThe system extracts key entities and relational networks from the input seed text, constructs a dynamic knowledge graph using temporal GraphRAG technology, and injects unique background memories and initial states into each agent to form an evolvable digital social infrastructure.
Environment setupMiroFish extracts entity relationships from knowledge graphs to generate intelligent agent personas, including dimensions such as panoramic experience, behavioral patterns, social relationships, and even MBTI personality. It also configures dual-platform simulation parameters and initial activation events to prepare for simulation operation.
Simulation engineDriven by the OASIS engine open-sourced by the CAMEL-AI team, it supports the parallel operation of thousands of agents and maintains long-term memory through Zep Cloud. Agents can generate real social interactions such as citations and comments, realizing the emergent effect of swarm intelligence.
Report generationMiroFish is equipped with a dedicated ReportAgent that has a rich toolset, enabling deep interaction with the simulated environment, automatically analyzing user forecasting needs and integrating multi-round simulation data to ultimately output structured forecasting analysis reports.

How to use MiroFish

Online experienceYou can directly access the official MiroFish demo at https://666ghj.github.io/mirofish-demo/ to experience the pre-built public opinion analysis and Dream of the Red Chamber continuation case studies without installation.
Local deployment (source code method):
- Environmental preparationInstall Node.js 18+, Python 3.11-3.12, and UV package manager.
- Configuration keyCopy .env.example to .env, and fill in the LLM API Key and Zep API Key.
- Install dependencies:runnpm run setup:allInstall front-end and back-end dependencies with one click.
- Start service:implementnpm run devThe front-end accesses localhost:3000, and the back-end API is located at localhost:5001.
Docker deploymentAfter configuring the .env file, execute...docker compose up -dYou can pull the image and start the service.
Usage process
- Upload seed materialsUpload original materials such as data analysis reports, novel texts, or news events.
- Describe the predicted demandUse natural language to describe the problem or scenario you want to deduce.
- Waiting for system constructionMiroFish can automatically complete knowledge graph construction, agent generation, and environment configuration.
- Run simulationSet the number of simulation rounds and observe the interaction and social evolution of intelligent agents.
- View reports and interactRead the generated prediction reports or engage in in-depth dialogues with specific agents to explore the details.

MiroFish project address

GitHub repository: https://github.com/666ghj/MiroFish
Experience the demo online: https://666ghj.github.io/mirofish-demo/

Application scenarios of MiroFish

Public opinion crisis rehearsalMiroFish can input trending events and social media data to simulate the path of public opinion fermentation, predict the outbreak point of crisis and the turning point of public sentiment, and provide a zero-risk trial and error environment for corporate public relations and government decision-making.
Financial Investment DeductionBased on market signals, draft policies, and corporate financial reports, a digital sandbox is built to simulate the market chain reaction after mergers and acquisitions, financing, or macroeconomic policy adjustments, assisting investors in formulating forward-looking strategies.
Literary creation continuationUpload the first part of a novel and let an AI agent play the role of a character in the book to autonomously deduce the subsequent plot. For example, based on the first 80 chapters of "Dream of the Red Chamber", predict the lost ending and explore different possibilities of the characters' fates.
Corporate Strategy SimulationSimulate variables such as competitors' new product launches, price wars, or changes in industry regulations to deduce the impact on one's own market share and supply chain, and verify the feasibility of strategic adjustment plans.
Public Policy AssessmentTo construct a virtual society that includes different interest groups, test the long-term impact of tax policies, environmental regulations, or urban planning, and observe the obstacles and unintended consequences in policy implementation.