What is DataShelf?
DataShelf helps operating companies prepare and license business records to AI labs for model training and evaluation. It starts with an inventory of systems, operating history, and record types. This is a license, not a sale of the company. DataShelf handles coordination, packaging, and anonymization. Your team answers questions. It does not run a project.
How does DataShelf work?
You describe what you have in a data manifest: systems, volume, rights, and a few commercial details. The first form is answers, not files, and takes about 15 minutes. If it looks promising, a short fit call follows. DataShelf takes the inventory to several buyers and comes back with an indicative range. You decide whether to go further before anything is packaged. The agreement is signed, and you are paid per the paper.
Who buys the data?
AI labs license de-identified operating history to train and evaluate models. DataShelf takes the inventory to several buyers, so you are not limited to the first price offered. The same record can be worth different amounts depending on who needs it. A partner receives the cleaned record under contract, not as a customer list. DataShelf does not name who licensed which data.
Why now?
AI labs used to train on the open web. That well is running dry. They now pay for real operating history: the files a company already produced by doing the work. Most owners have never been asked, because the category is new.
What happens to the data?
The first form is answers, not files. Anything a buyer sees is de-identified first. A partner receives the cleaned record under contract to train or evaluate models, not as a customer list. You keep the underlying files. We do not name who licensed which data.
Do I need to send samples?
Not to start; the manifest comes first. If a buyer needs to see files, we agree with you on a call what to send and what to leave out, then send a private upload link. Nothing is uploaded through the public website. How to export each system.
How much do you get paid?
It depends on how long you have been operating, what is in the record, and which partner wants it. We do not quote a number until we have seen what you have. If it is not worth much, we say so.
Who owns the data?
You do. A license is not a sale; the buyer gets rights to use a de-identified copy under contract.
What happens after I share?
It is de-identified, held only for the agreed processing, and deleted after; you review the de-identified copy first.
Am I committing to anything?
No commitment until you sign. If you go ahead, you sign a short agreement so DataShelf represents you exclusively while we take your record to buyers and run the bids. Your own lawyer can review it first.
How is the payout set?
Volume, complexity, uniqueness, quality, and domain; each buyer weighs them differently, so we run several bids.
How much work is it for my team?
An inventory that takes about 15 minutes, one call, and pointing an engineer at an export. We do the rest.
Is my data going to train a competitor?
Buyers license de-identified history to train general models, not to build a product against you; the paper says so.
Do I need my own lawyer?
Yes, and we expect it; our counsel drafts the paper, yours reviews it.
Ready to discuss your records?
Check your data