AgentSandbox

AgentSandbox is a web + desktop app for building and running agent-based models to stress-test operational and public-policy decisions. Users define agents, environments, and rules through a constrained visual editor plus a lightweight scripting layer, then run scenario sweeps (e.g., staffing levels, pricing rules, traffic restrictions) and compare outcomes with uncertainty bands. The product focuses on “decision-ready” outputs: dashboards that translate simulation results into KPIs, risk ranges, and sensitivity rankings rather than academic plots. It includes a calibration workflow to fit model parameters to historical data, and a reproducibility pack that exports model versions, seeds, and assumptions for audit. It’s a combination traditional + AI app: AI assists with model scaffolding, parameter suggestions, and plain-language explanations of what changed between runs, but users remain in control of assumptions and constraints.