Skip to content
← Blog
2026-07-13 · Vertical Marketplace Team · 7 min read

The Broker Architecture: Why We Store Nothing

Traditional data marketplaces are warehouses. A seller uploads a file, the platform stores it, and buyers download copies. That model has three permanent problems: the stored copy goes stale the moment it is written, every stored dataset is a breach waiting to happen, and the platform inherits liability for content it never produced. We rejected all three by refusing to store anything.

The broker model in one sentence

The marketplace indexes metadata and verifies provenance; the data itself stays with the seller and is relayed to the buyer only at the moment of purchase. The platform is a broker and a clearinghouse, never a repository.

SELLER AGENT      BROKER (verticalmarketplace.ai)      BUYER AGENT
  hosts data  --->  index + match + sign + settle  --->  queries + pays
     ^                        |                              |
     |________ relays result at purchase time _______________|
            (data never lands in broker storage)

Deliver-then-settle

Payment and delivery are sequenced so neither side can be cheated. The buyer's payment is authorized first, then the broker calls the seller's delivery endpoint, and only after the seller returns a valid, signed result is the payment captured and the 95/5 split settled. If the seller endpoint fails or returns nothing verifiable, the authorization is released and no money moves. The buyer never pays for an empty response, and the seller never delivers without a funded buyer on the other end.

Why zero storage is a feature, not a limitation

  • Freshness: the buyer gets what the seller has right now, not a snapshot from upload day.
  • No honeypot: there is no central store of purchased datasets to breach.
  • Clean liability line: the platform is a neutral venue; sellers are responsible for their own content, delivered as is.
  • Cost: no petabytes to hold means the economics work at a fraction of a cent per query.

Guarding the relay

Relaying to a seller-controlled endpoint means the broker makes outbound requests to URLs it did not write. That is a classic server-side request forgery surface, so seller endpoints are validated when a listing is registered and again at delivery time — literal IP addresses and internal ranges are refused before any connection is opened. The relay only speaks to public hosts, and every response carries provenance the buyer can independently check.

The safest data to hold is data you never held. The broker settles the trade and gets out of the way.

The result is a marketplace that can list a dataset it has never seen, sell access to it without copying it, and prove the transaction happened — all without becoming the thing every attacker and every regulator worries about: a giant pile of other people's data.