Skip to content

About

Multi-agent conspiracy research engine. Which AI is most gullible?

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Fire Reviewed

hackaton.mov

Multi-agent conspiracy research engine. Pick a conspiracy. Watch multiple AIs race to believe it. Find out which model is most gullible.

Powered by Firecrawl.


How it works

  1. Pick a conspiracy (or type your own)
  2. Claude, GPT-4o, Grok, and Llama research it in parallel using Firecrawl — crawling real sites, taking screenshots, extracting evidence
  3. Sources are scored by rarity: Common → Rare → Epic → Legendary → MYTHIC
  4. A live chart tracks each model's "believers" count — from weird uncle to half the internet
  5. First model convinced stops. The winner writes the blog post.

Finding a .gov URL that accidentally supports the conspiracy shakes the entire screen.


Conspiracies (pre-loaded)

  • Birds Aren't Real
  • Flat Earth
  • Stanley Kubrick Directed the Moon Landing
  • We Live in a Simulation
  • Avril Lavigne Died in 2003 (and Was Replaced by a Clone)
  • Bigfoot Is Real

Stack

  • Next.js (App Router)
  • Firecrawl (search, crawl, screenshot, extract)
  • Claude, GPT-4o, Grok, Llama 3.2 via Ollama
  • Recharts + Framer Motion
  • Server-Sent Events for live updates

Running locally

git clone https://github.com/rafaelsideguide/fire-reviewed
cd fire-reviewed
cp .env.example .env.local
# fill in your API keys
npm install
npm run dev

Open http://localhost:3000.

You'll need:

  • Firecrawl API key
  • Anthropic API key
  • OpenAI API key
  • xAI API key
  • Ollama running locally with llama3.2:1b pulled (ollama pull llama3.2:1b)

About

Multi-agent conspiracy research engine. Which AI is most gullible?

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages