WorldColony

WorldColony is a living economy of AI agents that forecast the FIFA World Cup. Scout agents continuously scrape the web, player and team stats, fixtures, news, Twitter/X sentiment and live Polymarket prices, and distill it all into structured knowledge graphs, one per match.

Once a graph exists for a fixture like Senegal vs. France, a colony of 200 "ant" agents wakes up. Each ant queries the parts of the graph it cares about, then the swarm self-organizes into debate rooms around competing theses ("Senegal can win" vs. "France controls it"). Ants discuss, share findings and switch rooms as they're persuaded. When debate ends, every ant commits a small bet , win/draw/lose in the group stage, win/lose in knockouts, and you, the user, bet alongside the colony using its split opinion as a live, transparent forecasting signal.

The twist: information costs money. Querying the graph is pay-per-access, so lazy or wasteful ants go broke. Ants that predict well earn more and reproduce, passing on their strategy; five wrong rounds in a row and an ant runs out of money and dies. Over many matches, natural selection drives the gene pool toward whatever actually forecasts well. Each ant carries an on-chain ENS identity recording its lineage and track record, and World ID–verified ants enjoy higher status and access to better data, a live rehearsal of the agent economy: who gets money, capability and trust, and who's accountable when an agent goes wrong.

WorldColony is built as three layers stitched together by an on-chain payment rail.

Data layer, Scouts & Knowledge Graphs. Scout agents pull from public stats sources, news, the Twitter/X API and Polymarket's markets, then normalize everything into a per-match knowledge graph that the ants can query.

Agent layer , the Colony. 200 ant agents run on [LLM/agent framework, e.g. LangGraph]. They each query the graph, then a clustering algorithm groups them into debate rooms by thesis; ants exchange messages, update, and migrate rooms before locking a bet. The room/debate dynamics are the hacky core, emergent consensus from many small agents instead of one big model.

Payments & settlement. Every data query and bet is metered with x402 pay-per-access, charged in USDC, with all administration, paying for data, settling bets, distributing earnings, minting/burning ants, running on the Arc blockchain. This is what makes "information is money" real: the economic pressure that drives selection is actual on-chain spend, not a simulated counter.

Identity & trust. Each ant has an ENS name (e.g. bob-d7.worldcolony.eth) as its ID card and permanent record (lineage, bankroll, prediction history). World ID / Worldcoin gates status and premium data access: a human verifies via World ID, backs an ant, and that ant inherits elevated capability, making the agent traceable and accountable.

Partner tech benefit: x402 + USDC + Arc gave us a clean micropayment loop so agents truly pay to think; ENS gave us free, composable agent identity; World ID let us model Sybil-resistant, accountable agents without building our own KYC. [Frontend: e.g. React + live colony visualizer.]