mirofish-offline-simulation

What This Skill Does

MiroFish-Offline is a powerful, fully local multi-agent swarm intelligence simulation engine. It allows users to ingest complex documents—such as policy drafts, financial reports, or marketing press releases—and subject them to a stress test of public opinion. By leveraging Neo4j for graph-based knowledge retention and Ollama for local LLM inference, the skill generates a population of AI agents, each assigned unique, nuanced personalities and background beliefs. These agents interact within a simulated environment to produce time-indexed social media threads, arguments, and shifts in consensus. Because it runs entirely on your local hardware using Docker, your sensitive data remains air-gapped from cloud vendors, providing a secure sandbox for predictive social analytics.

Installation

Installation is streamlined via Docker Compose for a unified experience. After cloning the repository, ensure your environment variables are configured in the .env file to point to your Neo4j instance and local Ollama server. Use docker compose up -d to spin up the backend and frontend services. Once running, you must ensure the necessary models are pulled into your Ollama container: qwen2.5:32b for the primary reasoning engine and nomic-embed-text for handling vector search tasks. For manual installations, ensure your Neo4j instance is reachable via the bolt protocol at bolt://localhost:7687.

Use Cases

This skill is ideal for communication strategists, policy analysts, and market researchers. You can use it to anticipate backlash against a new corporate policy, test the phrasing of a public statement for maximum positive sentiment, or analyze how different ideological personas interpret economic data. It is a critical tool for "red teaming" public communication strategies before they go live, allowing you to observe simulated conversations in a safe, controlled environment.

Example Prompts

1. "MiroFish, ingest the Q3 fiscal report in /data/reports and simulate how a group of tech-skeptic investors would react to our growth projections over 24 hours."
2. "Analyze the attached policy draft. Generate a simulation of 100 agents with diverse political demographics and provide a summary of the top three polarizing topics that emerge in the discussion."
3. "Start a new simulation based on this press release. I want to see if the sentiment shifts favorably if we emphasize the environmental impact clause in the second paragraph."

Tips & Limitations

MiroFish-Offline is resource-intensive. Running a 32B parameter model requires at least 24GB of VRAM for optimal inference speed. If you have limited hardware, utilize the qwen2.5:14b model to reduce memory pressure. Ensure your Neo4j database has sufficient memory allocation if you plan to scale the simulation beyond several hundred agents. Remember that this is a predictive simulation; agents are stochastic and results should be interpreted as potential scenarios rather than absolute certainties.