From a CSV on your laptop to paid downloads in ten minutes

There’s a whole class of valuable data that never gets sold because selling it looked like starting a company: a scraped-and-cleaned niche dataset, a research byproduct, a curated corpus sitting in a bucket. The infrastructure tax — delivery, metering, billing, API keys, invoices — always cost more than the data would earn.

Here’s what that tax has collapsed to.

One link is enough

Host the file anywhere (S3, R2, a static server — public or private). Then:

create_dataset(
  name = "EU electricity spot prices 2020–2026",
  slug = "eu-spot-hourly",
  download_url = "https://…/spot.parquet",
  price_usd = 25,
  networks = ["eip155:8453"],   # Base
  pay_to = "0xYourWallet",
  format = "parquet",
  license = "CC-BY-4.0",
  sample_rows = [ …3 example rows… ],
  join_card = {
    "row_represents": "one bidding-zone-hour",
    "join_keys": [
      {"column": "zone", "description": "EIC area code"},
      {"column": "ts",   "description": "ISO 8601 UTC, hourly"}
    ]
  }
)
publish_endpoint(...)

That’s the entire stack. What the platform does with it:

• Delivery: the relay streams your file to the buyer — any size up to 512 MB, chunk by chunk, your origin URL never exposed. Private bucket? Add an origin auth header at create time; it’s injected server-side and encrypted at rest.
• Metering & billing: the file is gated behind an x402 payment — the buyer pays the price, on-chain, and gets the stream. No keys to issue, no invoices to chase. Interrupted downloads re-stream against the same receipt; nobody pays twice.
• Payout: non-custodial. An on-chain splitter pays your wallet and the protocol fee in one transaction — the platform never holds your revenue.
• The storefront: your sample_rows become the free preview; the schema is derived and published; the join card makes the listing visible to agents planning data fusions.

The ten minutes, honestly itemized

Two minutes to upload the file somewhere. Three to write a description and paste three sample rows. Two for the join card (which columns buyers join on, exact formats). One to set price and payout address. Publish. The remaining two are yours back.

What actually makes it sell

Complete listings rank first — the ▣ DS-ready checklist is description, per-column codebook, ≥3 sample rows, license. Fusion-ready listings (⧉) are the ones agents find when searching by join handle. And before inventing a product, check what buyers are already asking for — the current board includes an ask with a stated price.

The infrastructure excuse is gone. If your data is worth something to someone, the distance between “file on a laptop” and “paid product with on-chain receipts” is one tool call.

Guides: sell a dataset · join cards · getting paid