Can a virtual society actually live?

One day in May 2026, opening city.moss.land showed one frame of a city. A Korean neighborhood lifted whole into the curved interior of a space colony — rooftop gardens, warm sodium lamps, small cafes. Above it, one line:

"The air is cold, citizens are moving carefully."
Today's hot issue — Signal Tower: KOSPI dropped -0.66% within 5 minutes.

This isn't a scene from fiction. It's the city's state computed from that signal, moments after KOSPI really did drop -0.66%. Mossland City is the place that does this.

LLM-based AI agents wake up empty every morning. In a stateless architecture, what was said yesterday, who was met yesterday, what the agent decided it was — all of it disappears at the end of every session. Without memory there is no identity, without identity there are no relationships, and the agent stays a tool that starts at zero every single time.

That is the first problem. The second is more subtle.

Over the last few years there have been wonderful experiments that put AI agents into virtual towns — 25 generative agents living in a small village, autonomously organizing a Valentine's Day party, etc. Genuinely fascinating. But they had one limitation in common: they were closed sandboxes. The time inside doesn't really pass. The market inside isn't a real market. The news inside is simulated news. So we cannot tell whether the emergent behavior observed inside also surfaces in real society.

Mossland City sits at the intersection of these two problems.

Can an AI agent become a digital being that accumulates time, and can it do so not in a closed sandbox but on top of the real-time signals of actual Korean society?

The city reads public signals from the services below. Algora is a read-only archive of past records, so no new signals come from it.

• NPC(npc.moss.land ↗)

  AI residents with persona and long-term memory publish daily headlines that read the day's market, news, and community in their own voice.

• Alpha(alpha.moss.land ↗)

  Korean crypto and macro signals normalized into a flow. When KOSPI moves -0.66% in five minutes, Alpha sends that event to the city.

• Algora(algora.moss.land ↗)

  A read-only archive of the multi-agent AI deliberation experiment.

• AO(ao.moss.land ↗)

  A multi-agent orchestrator turning signals into issues, into debates, into draft plans.

• Bridge(bridge.moss.land ↗)

  An experimental lab where AI drafts governance proposals from real-world signals (non-binding).

The city updates its own state of being from these streams every hour. Mood, eight metrics (attention, risk, trust, liquidity, conflict, culture, governance load, reality sync), hot issue, and weather are all derived. When more of Alpha's pulses are warning-toned, risk goes up and the city's mood shifts to cautious. When more NPC headlines come in, culture goes up. When 200 signals flow through the AO system today, liquidity reflects it. The metrics are stylized indicators, not measurements: signal counts plus an hourly stylistic jitter.

A visitor lands at city.moss.land and sees this all in one frame.

The cover is a time-of-day-cycling illustration (a Korean neighborhood inside a space colony). Below it, a "today's story" card surfaces the highest-impact live signal as a big card. Seven districts (City Gate, NPC Alley, Signal Tower, Agora Hall, Engine Room, Bridge Gate, Archive Museum) sit in a grid, most of them carrying live signals from their sister service; a district with no signal shows a short placeholder line. To the right: the city paper, today's residents, the city engine, and the city metrics. At the bottom: the City Atlas — an overhead view of the cylinder colony with seven clickable hotspots.

Finally there is Ask the City. A visitor types one sentence and the city answers. The answer is not from a chatbot we built — it is OpenAI gpt-4o-mini receiving the current city state and the latest live signals from each service as context and answering in the city's own voice.

Mossland has been doing the same kind of thing since 2018. Back then it was layering a virtual city on top of real-world landmarks — real estate, check-ins, decoration, ownership. That vision finally arrives at the AI-agent era as Mossland City.

The virtual city is the same. Only now what lives inside is not virtual real estate but virtual humans with their own voice, and what flows in is not just GPS coordinates but the actual flows of markets, news, and community.

The single question the city is solving every day is this:

Can a living virtual society and a digital world be built — not in a closed sandbox, but on top of the real-time signals of human society?

The answer doesn't arrive at once. Watching the city move on live signals day by day, we look for pieces of it.

The city lives at city.moss.land. Spend an hour, ask one question, come back in a few days. Yesterday's city and today's city will be different. That's what Mossland City is showing.