MiroFish is a Python-based swarm intelligence engine designed to simulate and predict outcomes across diverse domains including financial forecasting, public opinion analysis, and social dynamics. The project operates as a multi-agent simulation platform where thousands of intelligent agents with independent personalities, long-term memory, and behavioral logic interact within a digital environment to model real-world scenarios and generate predictions.
The core functionality involves a five-stage workflow beginning with graph building, where seed information extracted from real-world sources undergoes entity relationship extraction and knowledge graph construction. The system then moves to environment setup, generating agent personas and configuring individual behavioral parameters. The simulation stage executes dual-platform parallel simulations that parse prediction requirements and dynamically update temporal memory. Following simulation, a ReportAgent with specialized tools generates detailed prediction reports, and users can engage in deep interaction by chatting with individual agents within the simulated world.
The repository is written in Python and maintained primarily by contributor 666ghj, who accounts for 252 tracked events according to GitGenius activity data. Secondary contributors Jimmy-L99 and yuanxiaodong79 each have 11 tracked events. The project shows strong community engagement with 323 total tracked issues and pull requests, maintaining a median response latency of 2.4 hours and a mean latency of 50.9 hours. The most active issue labels are question (54 instances), enhancement (34 instances), and LLM API (19 instances), indicating ongoing feature development and user inquiries around language model integration.
MiroFish's stated applications span serious prediction use cases and playful simulations. For decision-makers, it functions as a rehearsal laboratory enabling zero-risk testing of policies and public relations strategies. For individual users, it serves as a creative sandbox for exploring hypothetical scenarios, from deducing novel endings to analyzing trending public opinion events. The project includes a live demo environment showcasing prediction simulations on public opinion topics and has produced demonstration videos including a Wuhan University public opinion simulation and a "Dream of the Red Chamber" lost ending prediction based on analysis of the novel's first 80 chapters.
The project is built with Node.js 18+ for the frontend and Python 3.11-3.12 for the backend, using the uv package manager. Deployment options include source code installation or Docker containerization. The frontend runs on localhost:3000 while the backend API operates on localhost:5001. MiroFish's simulation engine is powered by OASIS (Open Agent Social Interaction Simulations) from the CAMEL-AI team, and the project has received strategic support and incubation from Shanda Group.
The repository is classified by GitGenius across multiple security-related categories including phishing, campaigns, page generation, social engineering, red teaming, penetration testing, and attack simulation, though these classifications appear to reflect the platform's capability to model adversarial scenarios rather than its primary intended use. The project maintains connections to major open-source repositories including microsoft/vscode, microsoft/typescript, and rust-lang/rust through overlapping contributors. The team is actively recruiting full-time and internship positions for developers interested in multi-agent simulation and LLM applications.
