civlab·ai-society-sandbox.psyverse.fun·agents

◯ AI Society Sandbox

A toy economy of autonomous AI agents — citizens, firms, regulators — running on visible rules.

AI Society Sandbox is a deterministic, fully-visible micro-society in which every agent's policy and every transaction is inspectable. Citizens have utility functions; firms have production functions; regulators have rule sets. You can step the simulation, change a regulation, and see the macro consequences in 50 ticks. Not a prediction of how AI agents will actually behave; a tool for building intuition about which institutional designs survive autonomous-agent populations.

No magic black box: the policies are short, the rules are short, the data is in front of you. If a result surprises you, you can find the line of code that produced it.

Interactive demo

Use it — don't read about it.

tick 0

policy

citizens alive40

firms5

total cash4000

gini0.000

Citizens lose 0.5 cash per tick; trade for 5 with nearby firms if policy permits. Cash to zero = death.

Features

What this platform promises.

Pause-step-resume the economy

Pause at tick 47; inspect why citizen #312 starved; resume.

Drop in a new policy live

Halfway through a run, change all firms to a new policy.

Save runs to share with colleagues

Share a tick-by-tick replay URL.

Modules

Modules that compose this platform.

01 · agent-types

Agent types

Citizen (consume + work), Firm (produce + hire), Regulator (set rules).

02 · policies

Policy library

Greedy, satisficer, RL-shaped, random, malicious — pluggable.

03 · stepper

Discrete-time stepper

Pause, step, rewind. Inspect any agent's state at any tick.

04 · rules

Regulation editor

Add taxes, caps, anti-monopoly rules; see effects.

05 · metrics

Macro metrics

GDP, Gini, employment, monopoly index over time.

Data model

Agent

field

type

note

uuid

Agent id

type

enum

{citizen, firm, regulator}

policy

code

Reference to a Policy in the library

state

blob

Type-specific local state (cash, inventory, etc.)