Dispute Defender

Dispute Defender: an AI agent that wins PayPal disputes for small sellers.
Watch the agent think: it checks orders, tracking, buyer history and photos for every case at once.
Morning briefing: every dispute gets a plan, with $416 expected to keep.
The verdict: fight, 93% win odds, with the evidence packet ready for PayPal.
AI vision reads the delivery photo; an NFC tag in the box proves it arrived.
PayPal rules our way, and $129 stays with the seller.
When a partial refund is fairer, our agent negotiates with the buyer's agent.
Autopilot follows your rules; new disputes and decisions flash live.
Risk Radar stops fraud before it ships: this $580 order is held.
Track Record: the agent learns from PayPal's rulings, so its odds stay honest.
Just ask: a voice copilot that answers like a teammate.
Seven themes, including Kawaii Bow with Mochi the mascot.
PayPal Midnight theme.
Demo shop: PayPal-style checkout, and every order is scored by Risk Radar.
every dispute in one sortable, filterable table with the AI's plan, win odds and expected money kept, plus live totals and CSV export.

Inspiration When a buyer opens a PayPal dispute, small sellers lose hours gathering evidence, and often just refund, even when the package was delivered. Big retailers have fraud teams; small sellers don't. We wanted to give every seller their own AI dispute specialist that always keeps them in charge.

What it does Triages every dispute at once: checks the order, tracking, buyer history, policy and photos, then plans each case (fight, settle, refund or reply) with win odds. Builds the evidence and sends it through PayPal's Disputes API, only after the seller approves. Negotiates fair partial refunds with a simulated buyer agent. Autopilot handles clear-cut cases within the seller's rules. Risk Radar blocks fraudulent orders before they ship. Learns from PayPal's rulings to keep its odds honest. Voice assistant you can just talk to.

How we built it Claude as a tool-using agent for reasoning and photo analysis. PayPal: Disputes, Webhooks, Orders, Payments, Tracking and Transaction Search APIs, tested on a real sandbox account. Stack: Node.js and Express, with plain JavaScript on the frontend. Live demo: runs entirely in the browser on GitHub Pages, with no setup. Challenges we ran into Keeping an agent that can move money safe and predictable. Real PayPal sandbox dispute flows and their stages. Running the full server inside a browser tab. Making win odds honest instead of overconfident.

Accomplishments that we're proud of End-to-end on a real PayPal sandbox: a buyer opens a dispute, the agent responds, and PayPal accepts. Calibration that works: learning from past rulings cut wrong calls by about 12% on our sample history. Anyone can try it in the browser, with no keys or sign-up.

What we learned AI earns trust by showing its reasoning. Fighting isn't always winning: a fair refund can keep more money. And stopping fraud before it ships beats winning a dispute afterward.

What's next for Dispute Defender Live negotiation with real buyers, per-store learning, and one-click install for Shopify and WooCommerce sellers.