                          .
                        A       ;
              |   ,--,-/ \---,-/|  ,
             _|\,'. /|      /|   `/|-.
         \`.'    /|      ,            `;.
        ,'\   A     A         A   A _ /| `.;
      ,/  _              A       _  / _   /|  ;
       /\  / \   ,  ,           A  /    /     `/|
      /_| | _ \         ,     ,             ,/  \
     // | |/ `.\  ,-      ,       ,   ,/ ,/      \/
     / @| |@  / /'   \  \      ,              >  /|    ,--.
    |\_/   \_/ /      |  |           ,  ,/        \  ./' __:..
    |  __ __  |       |  | .--.  ,         >  >   |-'   /     `
  ,/| /  '  \ |       |  |     \      ,           |    /
 /  |<--.__,->|       |  | .    `.        >  >    /   (
/_,' \\  ^  /  \     /  /   `.    >--            /^\   |
      \\___/    \   /  /      \__'     \   \   \/   \  |
       `.   |/          ,  ,                  /`\    \  )
         \  '  |/    ,       V    \          /        `-\
          `|/  '  V      V           \    \.'            \_
           '`-.       V       V        \./'\                \
               `|/-.      \ /   \ /,---`\
                /   `._____V_____V'
                           '     '

███╗   ███╗███████╗ ██████╗  █████╗ ███████╗██╗███████╗██╗  ██╗
████╗ ████║██╔════╝██╔════╝ ██╔══██╗██╔════╝██║██╔════╝██║  ██║
██╔████╔██║█████╗  ██║  ███╗███████║█████╗  ██║███████╗███████║
██║╚██╔╝██║██╔══╝  ██║   ██║██╔══██║██╔══╝  ██║╚════██║██╔══██║
██║ ╚═╝ ██║███████╗╚██████╔╝██║  ██║██║     ██║███████║██║  ██║
╚═╝     ╚═╝╚══════╝ ╚═════╝ ╚═╝  ╚═╝╚═╝     ╚═╝╚══════╝╚═╝  ╚═╝

Simulate Reality. Locally.

Swarm intelligence engine that models public opinion, market sentiment, and social dynamics — zero cloud, zero API keys.

$ curl -fsSL https://megafish.sh/install.sh | bash

View on GitHub

// capabilities

Everything runs on your machine.

No subscriptions. No rate limits. No data leaving your hardware.

🖥

100% Local

🧠

Multi-Agent Swarm

Hundreds of AI agents with unique personas, biases, and communication styles simulating real social dynamics.

🕸

Knowledge Graph

Automatic entity and relationship extraction from your documents, stored in a live Neo4j graph database.

📊

Full Reports

A dedicated ReportAgent synthesizes simulation output into structured analysis with charts and sentiment breakdowns.

💬

Chat with Agents

After a simulation run, interview any simulated persona directly. Probe their reasoning, test reactions to new info.

⚡

One Command

Type megafish and the entire stack spins up — graph, agents, simulation engine, and report pipeline.

// origin

MiroFish vs MegaFish

MiroFish — the original

Open-source swarm simulation.

Feed it a news event, financial signal, or any seed data — it spins up thousands of AI agents with unique personalities and memory, lets them interact in a digital sandbox, and outputs a prediction report on how things play out.

Requires an LLM API key (e.g. Alibaba Qwen-plus) to power agents
Requires Zep Cloud for persistent agent memory
Ongoing API costs and cloud dependencies
Not fully local out of the box

→ upgraded to

MegaFish — the upgraded version

A window into what humans do.

MiroFish requires cloud APIs and an internet connection to run. MegaFish runs everything locally — no API keys, no subscriptions, your data never leaves your machine.

World simulation — simulate any scenario against 8.3 billion people
Focus group mode — chat with agents after the simulation to ask why they reacted the way they did
Auto-generated reports — structured analysis grounded in your document
One-line install — everything set up automatically, no manual configuration

// main purpose

Given any event — what does the world actually think, and how does that opinion spread?

MegaFish is a public reaction simulator. Instead of running a survey or waiting for real social media data, you feed it a document and get a statistically-grounded simulation of global reaction — broken down by culture, age, income, and personality type. Entirely offline. No cloud. No API costs.

MiroFish was the seed. MegaFish is what grows from it.

// pipeline

From docs to insights in five steps.

MegaFish orchestrates a full simulation pipeline automatically.

STEP 01

Graph Build

Feed in documents, articles, or raw text. MegaFish extracts named entities, events, and relationships into a live Neo4j knowledge graph.

STEP 02

Environment Setup

The EnvAgent generates a population of distinct AI personas — each with demographics, opinions, and social positions — seeded from the graph.

STEP 03

Simulation

Agents interact on simulated social platforms: posting, replying, reacting. Opinion dynamics, information spread, and sentiment shifts emerge organically.

STEP 04

Report Generation

ReportAgent analyzes the simulation log and produces a structured report — sentiment timelines, key influencers, narrative clusters, and risk flags.

STEP 05

Interactive Mode

Enter a live chat session with any simulated agent. Ask follow-up questions, introduce new information, and observe updated reactions in real time.

// applications

What can you simulate?

MegaFish is purpose-built for high-stakes opinion and sentiment modeling.

🔥

PR Crisis Testing

Simulate how a population responds to a leaked document, product failure, or controversial statement — before it happens. Identify narrative flashpoints and optimal response windows.

📈

Trading Signal Generation

Model how market participants react to earnings releases, regulatory news, or macro events. Capture sentiment shifts that precede price movement.

🏛

Policy Impact Analysis

Forecast public reaction to proposed legislation, tax changes, or government communications across different demographic and ideological segments.

// hardware

What you need to run it.

MegaFish is designed for commodity workstations. No datacenter required.

Component	Minimum	Recommended
System RAM	16 GB	32–64 GB
GPU VRAM	8 GB	24 GB (RTX 3090 / 4090)
CPU	8 cores	16+ cores
Storage	50 GB free	200 GB SSD
OS	Linux / macOS	Ubuntu 22.04 LTS
Runtime	Docker + Ollama	Docker + Ollama + CUDA 12