How it works
Organizations, the vendor and buyer sides, and how data reaches the marketplace.
DataDrop is a marketplace for private AI training and evaluation data. Vendors bring data they hold the rights to; teams that train and evaluate models buy it. This page explains the parts and where each one stands today.
Everything belongs to an organization
You sign in as a person, and everything you do on DataDrop belongs to an organization: its inventory, its listings, its agreements and its people. A person can belong to several organizations and switch between them. See Organizations and roles.
An organization has a vendor side, a buyer side, or both:
- The vendor side opens at once when you set up an organization with Sell data. A Team admin then accepts the vendor agreements, and the organization can build an inventory and make listings. See Become a vendor.
- The buyer side is by invitation. Setting up with Buy data files a buyer request, and an operator reviews it. See Become a buyer.
From your data to a listing
- Inventory. Each thing you could sell is an asset in your inventory: a codebase, a set of tasks or raw data. An asset is complete once its details are filled in and its data is linked.
- Data. You link data by uploading it: from the browser, over SFTP, or with the
datadropcommand. See Add assets. - Listing. A listing bundles complete assets with a price for each and a description buyers read.
- Review. You hand a listing in for review. It is locked while it is reviewed, and an approved listing is ready to publish.
- Live. You publish a ready listing. A live listing that is public is in the catalog, which the members of every organization browse under Browse marketplace. Until an organization that does not buy has published two listings of its own, it sees only the first six listings of each type, with no search or filters.
What is not open yet
These guides describe the product as it is today, and say so where a part is still to come. Buying, invoices, offers, earnings and withdrawals, the quality checks, estimates of a codebase's worth, importing tasks from Ember Data and buyers' requests for data are all built. What is not open yet:
- Payments are not switched on on the live site yet: no order can be paid and no payout sent there.
- The live site serves no code host, so a codebase comes in as a ZIP and waits for an operator.
- The
datadropcommand has no released download yet (see Add assets). - An organization cannot add its other side after setup, or ask again after a declined buyer request.
Every page of these guides carries a draft note until its words have been checked against the product.