An autonomous AI agent that takes any two unrelated topics and builds an increasingly unhinged conspiracy board connecting them. It searches the internet, finds tenuous connections, analyzes "evidence" images, and constructs a live graph that deepens with each round — getting progressively more conspiratorial.

Built at the Autonomous Agents Hackathon (Feb 27, 2026) at AWS Builder Loft, San Francisco.

1. Search — Tavily searches the web for both topics and connections between them
2. Extract — GPT extracts entities and relationships from the search results
3. Graph — Neo4j stores the conspiracy graph of entities and connections
4. Vision — Reka Vision analyzes found images for "hidden clues"
5. Remember — Senso.ai stores findings so each round builds on the last
6. Narrate — The agent provides unhinged commentary as it connects the dots

Each round feeds into the next, making the conspiracy deeper and wilder.

pip install -r requirements.txt
cp .env.example .env  # fill in your API keys

Required API keys (see .env.example):

• TAVILY_API_KEY
• NEO4J_URI, NEO4J_USERNAME, NEO4J_PASSWORD
• OPENAI_API_KEY
• REKA_API_KEY
• SENSO_API_KEY

Web UI (conspiracy board + graph visualization):

python -m agent.main --web

Open http://localhost:8000, enter two topics, and watch the board build itself.

CLI mode:

python -m agent.main "dolphins" "the pyramids" --rounds 3

agent/
  main.py       — CLI entry point (--web or --cli)
  agent.py      — Main agent loop (search → extract → graph → narrate)
  search.py     — Tavily web search
  extractor.py  — LLM entity/relationship extraction
  graph.py      — Neo4j graph operations
  vision.py     — Reka Vision image analysis
  senso.py      — Senso.ai knowledge base
  narrator.py   — Conspiratorial narration generator
  server.py     — FastAPI + WebSocket server
  static/       — Web UI (vis.js graph + cork board view)

A render.yaml blueprint is included. Set the env vars and deploy.