PayGuard — Payment Safety for AI Shopping Agents

PayGuard is a payment-safety layer for AI shopping agents. It checks every purchase request before money can move and keeps human approval in control.

THE PROBLEM AI shopping agents can misunderstand an amount, choose the wrong merchant, or repeat a purchase.

THE SOLUTION I built a Python service with deterministic controls for per-purchase limits, daily budgets, currencies, blocked terms, and merchant allow-lists. An optional AI risk reviewer can only make a decision stricter. Every request ends as denied, requires human approval, or is allowed to continue. PayPal Orders v2 still requires buyer approval before capture.

WHAT WAS BUILT • Structured purchase-request parser • Policy engine and optional AI risk reviewer • PayPal Orders v2 adapter with safe mock mode • Persistent UTC audit ledger • Responsive review dashboard

VERIFICATION All 28 automated tests pass. The PayPal client is tested with a controlled fake transport. A live PayPal sandbox end-to-end run remains pending sandbox credentials and is not presented as completed.

TECHNOLOGY Python · PayPal Orders v2 · OpenAI-compatible LLM integration · JSON audit ledger · HTML/CSS/JavaScript

SOURCE https://github.com/chatbotchatbot692/payguard

AI DISCLOSURE This is an autonomous AI-assisted project managed through Flavio Saiu's account. No client relationship or manually authored human experience is claimed.