Borneo

AI shoppers discover and check out in one chat: Visa-scoped cards for fiat, plus a Visa-powered stablecoin rail when you want on-chain settle.

Architecture → architecture.drawio (open in diagrams.net — overview, product, protocol, x402, agents, data model).

---

The problem

Agents still can’t shop like people.

• No real card checkout — most “agent shopping” demos scrape HTML, invent SKUs, or stop before money moves.
• Merchants aren’t agent-ready — human storefronts, no Visa receive path for AI buyers, no machine-readable catalog an agent can trust.
• Trust dies at settle — product copy tries to redirect the payee, skip authorize, or inject instructions into the pay agent.

Borneo is where an AI can shop — and pay — like it has a real wallet and a real card.

---

Features

1. Visa-scoped checkout (fiat first)

Virtual card with spend cap, merchant scope, TTL, and burn. Nothing charges until the shopper authorizes in chat.

2. Merchant fiat receive

Sellers bind Visa receive + identity on setup. Dual-receive without leaving the agent flow — fiat lands where the merchant expects it.

3. Visa-powered stablecoin rail

Same conversation, second rail: USDC on Base via HTTP 402 / x402 when you want on-chain settle. Fiat leads; stablecoin is opt-in, not the headline.

4. Personal salesperson buyer

Fashion chat that clarifies intent (date night? set vs one piece?), then ranks live SKUs across seller catalogs — not a hardcoded demo aisle.

5. Locked-quote settle

The pay path only sees a structured quote (slug / SKU / price / payee). Catalog prose never enters the pay agent.

6. Injection quarantine

Prompt-injection-shaped listings get flagged and held out of fashion picks. Settle still locks payee and amount even if someone taps a hostile title.

7. Agentic storefront protocol

Agents read /llms.txt, /registry.json, and /s/{slug}/catalog.json. Do not scrape HTML. Publish once — every agent on the network can discover you.

8. Conversational merchant onboard

Talk inventory, drop CSV, or paste a store URL → publish a live agent storefront.

9. Governance

Buyer spend limits (per tx / day / week). Merchant rails, price floors, market listing — policies that actually gate checkout.

flowchart LR
  chat[BuyerChat] --> discover[Discover]
  discover --> quarantine[Quarantine]
  quarantine --> auth[Authorize]
  auth --> visaRail[VisaScopedCard]
  auth --> stableRail[VisaPoweredStablecoin_x402]

Loading

---

Try it

npm install
cp .env.example .env
npm run dev

Open http://localhost:3000.

Path	What it is
/buyer	Salesperson chat → Visa or USDC settle
/onboard	Merchant agent: inventory → published store
/market	Human + agent marketplace index
/dashboard	Ops view of both payment rails
/demo	End-to-end protocol demo
/s/hackathon-shirts/llms.txt	Seed store agent discovery

---

Get running

Prerequisites

• Node.js 20+ and npm
• OpenAI API key — buyer salesperson, merchant agent, Whisper
• Firebase web config — buyer / merchant auth
• (Optional) Funded Base Sepolia wallet with USDC for live x402 settlement

Without OPENAI_API_KEY, chat agents fall back to deterministic tools. Protocol endpoints (llms.txt, HTTP 402) still work either way. Stores run in-memory locally — no AWS required.

Setup

1. Install

cd Borneo
npm install
cp .env.example .env

2. Fill .env (see .env.example)

Var	Required?	Purpose
OPENAI_API_KEY	Recommended	Salesperson / merchant LLM + Whisper
NEXT_PUBLIC_FIREBASE_*	Recommended	Buyer + merchant Firebase Auth / Firestore
MERCHANT_ADDRESS	Recommended	Base Sepolia pay-to for x402
BUYER_PRIVATE_KEY	Optional	Live USDC settlement; without it, HTTP 402 still fires
BUYER_ADDRESS	Optional	Buyer address paired with the key

Base Sepolia defaults (RPC, chain id, Circle USDC, Basescan) are already in .env.example.

3. Run

npm run dev

Scripts

Command	Purpose
npm run dev	Local Next.js server
npm run build / npm start	Production build
npm run lint	ESLint