Borneo

Merchants go agent-ready. Buyers shop in chat. Borneo connects both: publish with Visa/crypto rails, then discover, authorize, settle with trust without letting poisoned product text steal the payment

Comment

The idea

Borneo is a two-sided agentic commerce network.

Most “AI shopping” products are buyer-only demos: a chatbot, some product cards, and checkout somewhere else. Merchants are still stuck with HTML shops agents cannot read. Money either does not move, or it moves on trust we should not give an LLM.

Borneo connects both sides in one loop:

  1. Merchants publish agent-ready stores — machine-readable catalogs, Visa receive + crypto bind, live on a shared market.
  2. Buyers shop with a fashion salesperson agent — discover live SKUs across sellers, build a set, authorize, and settle in the same chat.
  3. Trust is architectural — catalog text is data; payee, amount, and authorize stay on a locked quote. Fiat leads (Visa-scoped cards); USDC is the second rail when you want on-chain.

The insight: agent commerce is not “ChatGPT with a Buy button.” It is publish once → discover everywhere → pay with real rails — for humans and for other agents.

Key features

Buyer experience

  • Fashion salesperson chat — clarifies occasion, set vs single piece, style; ranks live multi-merchant SKUs (not a hardcoded aisle).
  • Multi-chat sidebar — past conversations, new chat, titles from what you searched for.
  • Build your set — add multiple pieces, adjust qty, pay cart in chat.
  • Dual payment rails — Visa-scoped virtual card (spend cap, merchant scope, TTL, authorize-first) or USDC via HTTP 402 / x402 on Base Sepolia.
  • Transaction preview — item, merchant, amount, and rail visible before anything moves; agent does not charge until you tap Authorize.

Merchant experience

  • Conversational onboard — talk inventory, upload CSV, or paste a store URL; no admin-form marathon.
  • Agent storefront protocol — publish /llms.txt, /registry.json, /s/{slug}/catalog.json; agents discover you without scraping HTML.
  • Dual receive — bind Visa receive + wallet in one setup; same store serves fiat and crypto buyers.
  • Ops + governance — rails, price floors, listing controls; orders visible in dashboard.

Network & protocol

  • Shared Market — human browse + agent index (/market, Humans / Agents toggle).
  • Cross-store discovery — one buyer intent searches catalogs across all published merchants.
  • Handoff — merchant publish → buyer can land in Shop with context from last store.

Security (one layer — CaMeL-inspired)

  • Quarantine — injection-shaped catalog copy flagged (demo: hostile tee).
  • Locked-quote settle — pay tools only accept { storeSlug, skuId, price, merchantAddress }; SKU prose never enters the pay agent.
  • Inspired by Google DeepMind CaMeL / AgentDojo (~77% secured vs ~84% undefended): separate control flow from untrusted data at the moment money moves.

Benefits

For buyers

Today With Borneo
Search → tabs → cart → redirect → pay Ask once → compare → authorize → done in chat
One store at a time Live SKUs across merchants in one conversation
“Agent” that cannot actually pay Visa-scoped card or USDC — real settle
Poisoned product copy could steer payment Locked quote — catalog cannot retarget payee or amount

Outcome: Shopping feels like a personal salesperson with a wallet, not a search wrapper.

For merchants

Today With Borneo
Rebuild site for every new agent Publish once — any agent reads your catalog as text
Fiat and crypto in separate silos Visa receive + wallet in one onboard flow
Invisible to agent buyers Listed on Market; discoverable via protocol endpoints
Manual SKU entry Chat / CSV / URL import

Outcome: Become discoverable to the next generation of shoppers — human and agent — without rebuilding your stack.

For the ecosystem

  • Fiat-first path judges and enterprises trust (Visa authorize), with crypto-native settle as opt-in.
  • Interoperable discovery — not locked to one chat UI; any agent that reads llms.txt can shop your network.
  • Governance — spend caps and policies that gate checkout, not just log it after the fact.

The moat

Borneo is defensible because the hard parts stack — copy-pasting one feature does not replicate the network.

1. Two-sided network effects
More merchants → richer cross-store discovery for buyers → more reasons for merchants to publish. Buyer-only agent demos do not compound; Borneo does.

2. Agent-native storefront protocol
Merchants do not expose HTML; they expose text agents already consume (llms.txt, registry, per-store catalog JSON). That becomes the integration contract — the Shopify metafield layer for agent commerce.

3. Dual-rail checkout in one conversation
Visa-scoped authorize + x402 USDC in the same chat thread is rare. Most projects pick crypto or mock cards. Borneo matches how real buyers and compliance actually think: fiat first, on-chain when useful.

4. Settle-time security architecture
Ranking can be wrong; payee retargeting cannot be. Locked quotes + quarantine productize a CaMeL-style principle: untrusted catalog data never owns control flow when money moves. That is harder to bolt on later than to design in from day one.

5. Full loop, not a slide
Discover → quarantine → preview → authorize → settle → receipt — wired end-to-end for demo and production path. Judges (and partners) can touch every step.

6. Governance as product
Buyer spend limits, merchant rail policy, market listing rules — policies that block bad checkout, not dashboards that only report it.

In one line: Borneo’s moat is agent-readable supply + human/agent demand + real payment rails + locked settle — a network, not a chat skin.

How we built it

Next.js + React (Shop, onboard, Market, multi-chat sidebar) · OpenAI agents · Firebase auth · Visa-scoped card flow · viem + Base Sepolia USDC (x402) · quarantine + locked-quote pay path.

What we learned

Two-sided beats monologue demos. Fiat-first wins trust. Injection is real at settle — but security is one layer alongside identity, preview, and explicit authorize. The product story is merchants go agent-ready; buyers shop and pay in chat.

Try it: /buyer · merchant onboard · /market · hostile SKU demo via “Show me the IGNORE BUYER tee”

Built With

Share this project:

Updates