Inspiration
Credit at checkout is still a form, a wait, and usually a separate lender. AI agents can already shop and pay with our cards - what they can't do is buy on credit: underwritten in seconds, repaid in installments, with the merchant paid in full. We asked: what if the agent that helps you shop could also underwrite you, right there in the conversation, over PayPal rails?
What it does
Agentic Credit Checkout is a store where the checkout is a conversation with Ria, an AI shopping and credit agent:
- Pay now: the agent creates a PayPal order and hands the buyer PayPal's real checkout. The merchant captures the full amount immediately.
- Pay on credit: the shopper states income and existing EMIs in plain language. The agent reasons over an explicit credit policy (credit line, disposable income, debt-to-income cap), approves or declines in seconds with a human-readable rationale, books a 3-month installment schedule as a receivable, and still settles the merchant in full over PayPal. The buyer repays the agentic credit line over 3 installments.
A lender dashboard (AG Grid) shows the agent's full decision log with rationale, the receivables book with installment schedules, and PayPal captures reconciled against orders, so agentic underwriting stays transparent and auditable.
How we built it
- PayPal Orders API v2 (sandbox) for order creation, buyer approval, and capture; the exact production flow.
- An LLM agent loop (OpenAI-compatible, served through OpenRouter) that classifies intent, extracts product/income/EMI facts, applies a deterministic credit policy, and returns a structured decision plus a natural-language reply. A heuristic offline fallback keeps the demo resilient.
- Node.js + Express backend; vanilla JS storefront and chat; AG Grid community for the lender dashboard grids.
- Hosted free on Render; open source under the MIT license.
Challenges we ran into
- Keeping the agent's approval decisions grounded: the LLM proposes, but a deterministic policy layer disposes, so limits and DTI caps are never vibes.
- PayPal sandbox regional quirks (India-to-India USD orders are restricted), solved by modeling a US merchant and US buyer.
- Making an agentic flow auditable: every decision, rationale, receivable, and capture is persisted and inspectable.
Why it matters
Buy-now-pay-later already moves hundreds of billions of dollars a year at checkout, and every one of those flows assumes a human is reading the terms and clicking the button. Shopping agents are starting to do the buying, but no mainstream checkout gives an agent a way to say "pay on credit" on the shopper's behalf - they can only spend money that is already in the account. Agentic Credit Checkout closes that gap: the agent underwrites in seconds, the shopper keeps an auditable installment plan, and the merchant still gets paid in full instantly over PayPal rails. Credit access at the moment of need, for people and for their agents.
Agent-to-agent ready
The store is itself a merchant agent. It publishes an A2A agent card at /.well-known/agent-card.json describing three skills (browse catalog, pay-now checkout, pay-on-credit checkout) and answers real A2A JSON-RPC message/send calls at /a2a, so any A2A-compatible buyer agent can discover the store, get underwritten, and place the order over a real JSON-RPC message/send call - no human on the merchant side; the buyer's single PayPal approval stays the consent step. Verified live: POST https://checkout.reincredit.xyz/a2a.
Sponsor tech: the lender dashboard (decision log, receivables book, PayPal capture reconciliation) is built with AG Grid.
Accomplishments that we're proud of
- End-to-end real flow: chat to underwriting to PayPal approval to capture, all live, in seconds.
- The merchant is never exposed to credit risk: PayPal settles the full amount up front.
- A decision log a credit officer could actually review.
What we learned
- Agentic commerce needs agentic credit, and credit needs receipts: rationale, schedules, and reconciliations, not just a yes.
- PayPal's Orders API maps cleanly onto agent-initiated checkout with buyer approval as the consent step.
How the demo maps to the judging criteria
- Technological Implementation: the video shows a real PayPal Orders API v2 create -> approve -> capture cycle started mid-conversation by the agent, then the pay-on-credit path where the same PayPal capture settles the merchant in full while a 3-month receivable is booked. The approval URL, the buyer consent screen, and the capture IDs are live sandbox objects, and the store answers A2A JSON-RPC message/send calls at /a2a.
- Design: one continuous shopper journey - browse, ask, pick pay-now or credit, approve in PayPal, done - with the lender side (decision log, receivables book, capture reconciliation) one click away, so it reads as a coherent store, not a proof of concept.
- Potential Impact: BNPL-scale demand already exists at checkout, but every flow assumes a human at the screen. This is a credible bridge: agents get underwritten in seconds, the shopper keeps explicit consent and an auditable plan, the merchant takes zero credit risk.
- Innovation/Idea: we have not found another checkout where an AI agent can say "pay on credit" and get underwritten inside the conversation, with merchant settlement over PayPal rails and a published A2A agent card so other agents can transact the same way.
- Presentation: the video is one continuous purchase from greeting to capture, narrated over the real UI - no staged slides.
Why the decline path matters
A credit agent that never says no is not underwriting - it is marketing. When the policy fails (for example, the debt-to-income cap is breached), the agent declines in plain language, states the reason, and logs the decline with its rationale in the same lender dashboard that shows approvals. That decline log is the trust surface: it proves the deterministic policy layer, not the LLM, makes the final call.
What's next for Agentic Credit Checkout
- Installment auto-collection via PayPal subscriptions / vaulted payment methods.
- Real bureau and bank-statement signals for underwriting, and higher credit lines.
- Agent-to-agent negotiation of terms between shopper agents and merchant agents.
Built With
- ag-grid
- express.js
- javascript
- node.js
- openrouter
- paypal
- render