Behind every conversation is a complete person — demographics from the Census, a personality, a household, a job, and a day that unfolds in real time. Your agent talks to someone with context, not a one-line persona.
Age, sex, income, household, education and occupation aren't picked at random — they're drawn from the Census so each citizen is a statistically real combination of traits, not a caricature.
Each citizen carries Big Five personality traits that shape how they show up in a conversation — how patient they are, how much they trust the agent, how much they ramble or push back. Under the hood, each citizen is driven by an LLM conditioned on their traits, demographics, current activity and memory, so the hard cases show up in character without you scripting them.
A citizen has a household, a job, a commute and a schedule. When your agent reaches them, they're somewhere in their day — at work, on the bus, doing the school run — and that context shapes the conversation.
A citizen starts from the Census base, but you're not stuck with it. Layer on custom attributes — a chronic condition, an insurance plan, a digital-literacy level — and behaviors like distrusting AI or code-switching, on top of the existing person. You extend who's already there; you don't start over.
+ chronic condition+ insurance plan+ digital literacydistrusts AIcode-switchesskips follow-upsyour region · ACS+ your aggregate statsNo — every citizen is fully synthetic, generated from public, aggregate Census data. No real individuals, no personal data.
Each citizen is driven by an LLM conditioned on their demographics, Big Five traits, current activity and memory — so responses are generated in character for that specific person, not pulled from a script.
Demographics (age, sex, income, household, education, occupation), a home and job on the map, family and social relationships, a daily schedule, Big Five personality traits, and needs and a mood that change through the day.
Personality is modeled on top of the Census demographics using Big Five (OCEAN) traits. It is not pulled from the Census — it's the behavioral layer that makes each citizen act distinctly.
Yes. Conversations fold into a citizen's memory, so you can test follow-ups and the long game. See Memory for details.
Open any citizen and see the whole identity behind the conversation.