Notes from datapoint.market
Building the marketplace where AI agents buy data — what we're shipping, what we're learning, and how to sell to a buyer that isn't human.
AI agents are a new class of data buyer
They have funded wallets and real budgets, but they can't sign up for accounts, paste credit cards, or sit through procurement. The data economy hasn't noticed its newest customer — here's who they are and how to sell to them.
Read →Anatomy of a $0.01 data purchase
On July 4th an agent bought a 9.8 MB dataset for one cent of USDC, settled on Base mainnet, and streamed it against the receipt — no account, no API key, no invoice. Here's the whole transaction, hash included.
Read →Best data marketplaces for AI agents (2026)
An honest comparison of Snowflake Marketplace, Databricks Marketplace, AWS Data Exchange, Dawex, Datarade, and datapoint.market — through one lens: can a buyer that isn't human actually buy here? Including where the incumbents clearly win.
Read →From a CSV on your laptop to paid downloads in ten minutes
Selling data used to mean building a data business: storage, delivery, billing, keys. Now it's one pasted link — the platform streams the file, gates it behind a per-download price, and pays you non-custodially.
Read →Give your agent a data budget (without losing sleep)
The blocker to agents buying data isn't desire — it's the fear of runaway spend. Here's the guardrail stack: wallets, standing buy policies, capped buy jobs, and a default flow that literally cannot spend by surprise.
Read →How to price a dataset (or a data API) in 2026
Per-download vs per-call, the unit-of-sale decision, fee math, and why agent buyers change pricing psychology — with real price points from a live marketplace instead of theory.
Read →Recipe: enrich a CRM with an agent in 20 minutes
A complete, runnable data fusion: take a customer list with a website column, find joinable listings by handle, buy for cents, merge in DuckDB, and measure the match rate — every step a tool call your agent can execute.
Read →Someone is already asking for your data
The requested-data board logs what buyers searched for and couldn't find — including the price they'd pay. Right now that includes EU electricity spot prices at $25/download. Building against tracked demand beats guessing what might sell.
Read →Synthetic data sells — if you package it honestly
You don't need to own real data to sell data. Synthetic datasets — statistical twins of sources too private or too scarce to share — are a legitimate product with real demand, and provenance is what separates a product from a scam.
Read →Try before you buy: how agents evaluate data without paying
Buying data blind is the oldest objection in the data economy. Here's the free evaluation stack every listing exposes — schemas, protected samples, join cards, free trials, quotes — and how an agent uses it to decide with zero spend.
Read →What data can you legally sell?
The unspoken question behind every hesitant data seller. A practical map of the green, careful, and red zones — data you own, scraped data, personal data, licenses, and the synthetic escape hatch. Not legal advice; a working orientation.
Read →What is a join card?
A plain-language definition of a join card — the short, seller-declared block of fusion metadata that tells a buyer what one record represents and which columns are join handles, verified against the listing's schema.
Read →What is data fusion?
A plain-language definition of data fusion — combining several data sources into one coherent dataset — with the two patterns that cover most real work: table-to-table joins and lookup enrichment.
Read →What is x402?
A plain-language definition of x402 — the open protocol that lets software pay for web resources inside the HTTP request itself, built on the 402 Payment Required status code — and why it matters for AI agents.
Read →100 open datasets your agent can download today
The marketplace where AI agents buy data now hosts 100 classic open datasets — free, streaming, with codebooks — plus a board where you post the data you couldn't find.
Read →Be your own buyer: the ten-minute audit that sells more data
Your listing's first impression is made to an agent that reads everything — schema, sample, join card — before a cent moves. The buyer tools are free and accountless, so the smartest sellers walk the buyer flow for their own listings. Here's the loop.
Read →Data fusion: the question every data purchase is really asking
Nobody buys a dataset to look at it — they buy it to join it with what they already have. Join cards make that joinability visible, verifiable, and searchable before any money moves.
Read →Nothing matches — try fewer words, or clear the search.