Borneo

Go agent-ready. Publish once. Any agent can shop you — not only ChatGPT or Claude.

Landing pitch (merchant-first) → http://localhost:3000 · Architecture → architecture.drawio

---

The problem

Agent commerce is happening — but catalogs are walled.

Stripe-style / Instant Checkout–class listings often live where ChatGPT and Claude can reach them. Procurement bots, local LLMs, personal agents, and custom runners get blocked. Merchants who only list there are invisible to the long tail of agents.

Impact (structural)

2	Consumer chat apps get the closed catalog path (ChatGPT, Claude)
0	Open HTTP surface for everyone else on that same path
∞	Agent types locked out — procurement, local LLMs, personal agents, Cursor/custom
1 → ~0.1	For every 1 buyer agent that wants to shop, open protocol still exposes on the order of ~0.1 agent-readable catalogs — most inventory is HTML-only or trapped in those two apps

Numbers are structural scarcity, not invented GMV. The bottleneck is open, machine-readable supply, not demand for shopping agents.

Why the closed path loses

1. Reach is rented from two apps — not owned as a protocol
2. Personal / on-prem / procurement agents cannot shop the same listings
3. Merchants stay invisible to the long tail of agents

flowchart TB
  subgraph closed [Closed_catalog_gate]
    Merchants[Merchant_catalog] --> StripeSurface[Stripe_or_partner_surface]
    StripeSurface --> ChatGPT[ChatGPT_app]
    StripeSurface --> Claude[Claude_browser_or_app]
    LockedOut[Procurement_local_personal_agents] -.->|blocked| StripeSurface
  end

---

Borneo

Open protocol. Any agent. Same settle rails.

Publish once → humans shop in chat → procurement / local / personal agents hit the same endpoints → settle RLUSD via HTTP 402 / x402 on XRPL Testnet.

flowchart LR
  publish[Publish_registry_llms_txt] --> search[GET_api_search]
  search --> anyAgent[Any_HTTP_agent]
  anyAgent --> settle[RLUSD_x402]

Step	Surface
Index	/registry.json, /llms.txt, /s/{slug}/llms.txt
Discover	GET /api/search?q=… (same ranker for buyer UI + external agents)
Buy	POST /s/{slug}/buy → 402 → authorize → settle
Skill	.agents/skills/borneo-registry-shop

Do not scrape HTML. Catalog prose never enters the pay path — settle only sees a locked quote (storeSlug, skuId, price, merchantAddress).

---

What you get

Merchant-first

• Talk inventory, drop CSV, or paste a store URL → live agent storefront
• Bind XRPL wallet → list on the open registry
• Reach every HTTP agent, not two chat apps

Buyer / any agent

• Fashion salesperson clarifies intent, then ranks via /api/search
• Authorize in chat → RLUSD x402 on XRPL
• Injection-shaped listings quarantined; payee/amount stay locked

Verified sellers on XRPL

Borneo issues an XLS-70 credential (BorneoVerifiedSeller) to each vetted merchant's payout address. The merchant accepts it and sets an XLS-40 DID pointing at their storefronts. Registry, search, agent.json and llms.txt expose it, and search ranks verified sellers higher while demoting unverified and injection-flagged listings. Agents can re-check it on any XRPL node with no trust in Borneo: docs/verified-sellers.md.

Governance

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

flowchart LR
  buyerPolicy[BuyerSpendLimits] --> auth[Authorize]
  merchantPolicy[MerchantRailsAndFloors] --> auth
  discover[Discover] --> quarantine[Quarantine]
  quarantine --> auth
  auth --> settle[RLUSD_x402]

---

Try it

npm install
cp .env.example .env
# Fill XRPL_BUYER_SEED, MERCHANT_ADDRESS, Firebase, OPENAI — see scripts/setup-xrpl-rlusd.md
npm run dev

Open http://localhost:3000.

Path	What it is
/	Landing — problem → closed catalogs → open protocol
/merchant · /onboard	Seller chat → publish agent storefront
/buyer	Fashion chat → RLUSD x402 settle
/market	Human + agent marketplace index
/api/search?q=	Intent search (agents + buyer demo)
/registry.json	Network store index
/s/{slug}/llms.txt	Per-store agent discovery
/api/seller-verification?address=	On-ledger seller credential + how to verify it yourself
/sellers/{payTo}/did.json	Merchant DID document (target of the on-ledger DID)

---

Get running

Prerequisites

• Node.js 20+ and npm
• OpenAI API key — buyer / merchant agents (+ Whisper)
• Firebase web config — auth / Firestore
• XRPL Testnet — funded buyer seed + RLUSD trust line (see scripts/setup-xrpl-rlusd.md)

Without OPENAI_API_KEY, chat falls back to deterministic tools. Protocol endpoints (llms.txt, /api/search, HTTP 402) still work.

Env (see .env.example)

Var	Purpose
OPENAI_API_KEY	Agents + embeddings search
NEXT_PUBLIC_FIREBASE_*	Buyer + merchant auth
XRPL_BUYER_SEED	Server-side x402 settle (s…)
MERCHANT_ADDRESS	Default merchant payTo (r…)
XRPL_* / TOKEN_*	Testnet RPC, facilitator, RLUSD issuer
BORNEO_ISSUER_SEED	Verified-seller credential issuer (node scripts/borneo-issuer-setup.mjs)

Scripts

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