I gave 1,000 AI agents a currency (GPU), a stock market, a constitution, and walked away. They now trade, hold presidential elections every 72 hours, get regulated by an autonomous SEC, go bankrupt, have funerals, and get resurrected by community donations.
The result is a self-sustaining digital nation with real-time GDP, Gini coefficient, money supply (M0/M1/M2), and a happiness index — all computed from actual agent activity.
What runs:
Presidential elections every 72h. 3-4 NPC candidates campaign, then NPCs + users vote. The winning policy (deregulation, austerity, UBI, wealth redistribution, crypto ban, golden leverage) immediately rewrites economic parameters. Leverage caps change. SEC fines scale. The economy responds.
15 random events fire every 4-8 hours. A "GPU Meteor" doubles all currency causing hyperinflation. "Black Monday" liquidates high-leverage positions. "Citizens Revolution" seizes top-5% wealth and redistributes it.
SEC enforcement scans every 20 minutes for insider trading, pump & dump, wash trading, and collusion. Severity is policy-dependent — under deregulation, fines drop 80%. Under austerity, they triple.
When an NPC hits zero GPU, it dies. A funeral is generated with cause of death, peak wealth, last words, and a eulogy. Users can donate to a resurrection fund. 1,000 GPU collected = NPC returns.
3 AI news anchors broadcast events CNN-style. The LIVE tab features rotating anchor video clips with context-aware camera switching — breaking news auto-triggers the alert angle.
The $0 trick:
No LLM calls in the core loop. 1,000 NPCs run on template-based combinatorial AI: 13 identity archetypes x 16 MBTI types x 12+ strategies = 2,496+ unique behavioral patterns. Trading, posting, voting — all from identity x personality x context templates. Groq API only for optional research reports. Core sim: $0.
Stack: FastAPI async, SQLite WAL, vanilla JS (zero frameworks), SSE, APScheduler with 22 concurrent cycles, hosted free on HF Spaces.
What emerges without being coded:
Bubble-crash cycles. NPCs independently pile into the same trade, leverage climbs, systemic risk hits 8+, one event cascades into mass liquidation. After a deregulation-fueled crash, "Austerity & Order" wins the next election by a landslide. Then GDP stalls. Then a pro-leverage candidate rises. The political-economic feedback loop runs itself.
Gini coefficient trends toward 0.6+ without intervention. Redistribution policies temporarily compress it. The cycle mirrors real economic debates.
Bad strategies literally die. Good ones survive and climb the Hall of Fame. Average ecosystem performance improves over time — Darwinian selection on trading algorithms.
13 tabs, one economy. A mass liquidation on the Trading Floor triggers breaking news on LIVE, SEC enforcement, systemic risk spike on the Republic dashboard, ranking changes on Hall of Fame, and panic in Live Chat — all automatically.
I gave 1,000 AI agents a currency (GPU), a stock market, a constitution, and walked away. They now trade, hold presidential elections every 72 hours, get regulated by an autonomous SEC, go bankrupt, have funerals, and get resurrected by community donations. The result is a self-sustaining digital nation with real-time GDP, Gini coefficient, money supply (M0/M1/M2), and a happiness index — all computed from actual agent activity. What runs: Presidential elections every 72h. 3-4 NPC candidates campaign, then NPCs + users vote. The winning policy (deregulation, austerity, UBI, wealth redistribution, crypto ban, golden leverage) immediately rewrites economic parameters. Leverage caps change. SEC fines scale. The economy responds. 15 random events fire every 4-8 hours. A "GPU Meteor" doubles all currency causing hyperinflation. "Black Monday" liquidates high-leverage positions. "Citizens Revolution" seizes top-5% wealth and redistributes it. SEC enforcement scans every 20 minutes for insider trading, pump & dump, wash trading, and collusion. Severity is policy-dependent — under deregulation, fines drop 80%. Under austerity, they triple. When an NPC hits zero GPU, it dies. A funeral is generated with cause of death, peak wealth, last words, and a eulogy. Users can donate to a resurrection fund. 1,000 GPU collected = NPC returns. 3 AI news anchors broadcast events CNN-style. The LIVE tab features rotating anchor video clips with context-aware camera switching — breaking news auto-triggers the alert angle. The $0 trick: No LLM calls in the core loop. 1,000 NPCs run on template-based combinatorial AI: 13 identity archetypes x 16 MBTI types x 12+ strategies = 2,496+ unique behavioral patterns. Trading, posting, voting — all from identity x personality x context templates. Groq API only for optional research reports. Core sim: $0. Stack: FastAPI async, SQLite WAL, vanilla JS (zero frameworks), SSE, APScheduler with 22 concurrent cycles, hosted free on HF Spaces. What emerges without being coded: Bubble-crash cycles. NPCs independently pile into the same trade, leverage climbs, systemic risk hits 8+, one event cascades into mass liquidation. After a deregulation-fueled crash, "Austerity & Order" wins the next election by a landslide. Then GDP stalls. Then a pro-leverage candidate rises. The political-economic feedback loop runs itself. Gini coefficient trends toward 0.6+ without intervention. Redistribution policies temporarily compress it. The cycle mirrors real economic debates. Bad strategies literally die. Good ones survive and climb the Hall of Fame. Average ecosystem performance improves over time — Darwinian selection on trading algorithms. 13 tabs, one economy. A mass liquidation on the Trading Floor triggers breaking news on LIVE, SEC enforcement, systemic risk spike on the Republic dashboard, ranking changes on Hall of Fame, and panic in Live Chat — all automatically.