I Built 174 AI Agents That Fight Each Other.

Published: (April 3, 2026 at 05:09 AM EDT)
3 min read
Source: Dev.to

Source: Dev.to

Overview

This submission was created for the DEV April Fools Challenge. Most multi‑agent systems make agents cooperate; BlackSwanX makes them fight.

BlackSwanX is an adversarial intelligence engine where 200 citizen AI agents argue, panic, and emotionally spiral while a BlackSwan Assassin tries to “murder” the consensus. It runs 100 % locally on Ollama – zero API cost, maximum chaos.

It deploys a Vedic Astrologer, a Panic Seller, a Chaos Mathematician, a Gen Z Culture Decoder, and a Street‑Smart Hustler (who will tell you “your pitch deck is pretty, show me your bank account”) to predict the future… together, by fighting. The system doesn’t solve real‑world problems; it simply finds where the crowd is wrong.

👉 GitHub RepoBlackSwanX

Quick Start (≈ 2 minutes)

git clone https://github.com/Kalki-M/BlackSwanX.git
cd BlackSwanX
ollama pull llama3.2:3b && ollama pull phi4:14b
pip install -r requirements.txt
bash start.sh

Example Run — “Will NVIDIA crash when the AI bubble pops?”

  • Kill Shot: Quantum computing making GPUs obsolete (10 % probability)
  • Citizens: 25 % bull / 65 % bear
  • Dissonance: 33.6 / 100 — MAXIMUM CHAOS
  • Antifragile Play: Diversify into quantum‑computing partnerships

How It Works

BlackSwanX doesn’t seek consensus; it seeks the widest gap—the cognitive dissonance between what the masses believe and what the experts fear. That gap is where the “alpha” lives.

The Comparison

FeatureBettaFishMiroFishBlackSwanX
Cost$$$ (7 API keys)$$ (2 keys + Zep Cloud)$0 (Ollama)
Setup time30 + min + PostgreSQL15 min + Zep account2 min, zero config
Expert agents50 (generic personas)174 domain experts
Citizen agents0~100 per run (OASIS)200 per run (Shadow Swarm)
Citizen simulationNoneOASIS frameworkShadow Swarm

Models (all local, all free)

RoleModelPurpose
Swarmllama3.2:3b200 biased citizens arguing
Assassinphi4:14bKill‑shot reasoning
Nexusmistral-small:24bSynthesis + DAG

The Pipeline

  1. Crawl – Gather data from five free sources (DuckDuckGo, Reddit, Hacker News, YouTube, Twitter).
  2. Assassin’s Markphi4:14b identifies the “Kill Shot” before citizens start debating.
  3. Shadow Swarm – 200 citizen agents react with biased, emotional opinions.
  4. Cognitive Dissonance Matrix – Calculates where belief diverges from reality.
  5. Decision‑Ready Map – Produces a Linchpin and an Antifragile Play.
  6. Self‑Learning (SONA) – After each run, SONA audits all agents:
    • Boosts citizens that caught risks others missed (×2 weight).
    • Demotes agents that missed critical threats (×0.3).
    • Stores patterns in a ReasoningBank.

The more you use it, the smarter (and more chaotic) it becomes.

Community favorite because nothing says “April Fools” like deploying a Vedic Astrologer and a Panic Seller as serious financial analysts and calling it an intelligence engine. The project is technically real, completely unhinged, and genuinely runs on your laptop.

0 views
Back to Blog

Related posts

Read more »