Agent Spending Policy Playground

Set the rules, simulate an agent payment, watch the decision. This is the model behind SpendShield — a deterministic policy layer that sits between an AI agent and its payment rail, returning ALLOW / APPROVAL (human) / DENY before money moves.

1 · Policy rules

Daily budget ($) — cumulative, enforced outside the agent loop

Per-transaction cap ($)

Approval line ($) — over this = human required

Allowed merchants (comma separated) Blocked merchants (comma separated)

2 · Simulate a payment

Merchant / recipient Amount ($)

$25 · mcdonalds.com $75 · mcdonalds.com $500 · scam-vip.com

Recent attempts

Agent payment	Decision

What this is: an interactive preview of SpendShield's policy model for the common rules (budget, per-transaction cap, merchant allow/block lists, approval line). It is a simulation for comprehension — the real engine additionally enforces verifiable agent identity, signed one-time grants that execution must consume, replay protection and a tamper-evident audit chain. For the full engine: pip install spendshield → connect it in 2 minutes. MIT licensed, open source on GitHub.

← Home · FAQ · Connect (2 min) · Failure map · GitHub