What founders ask us first

Models are now trained to do real work inside companies: handle a support ticket, follow a runbook, move a deal through a pipeline. Public web data doesn’t show that. Your tickets, wikis and processes do, which is why labs and the data companies that supply them license it.

You license it. You keep your data and keep using it. The buyer gets the right to use a defined, prepared copy for training and evaluation, on the terms written in the license.

AI labs and the data companies that supply them. Which buyer fits depends on what your data shows, how much of it there is, and where your company is registered. We tell you who fits after the first call.

It depends on volume, years of history, how much of it is in English, how distinctive your workflows are, and how clean the rights are. After the 15-minute check we give you an honest range, or tell you it isn’t worth doing, along with what would raise it and how to push for a better offer.

Often, yes. It depends on what your client contracts say about ownership and data use. We go through them with you source by source, and buyers de-identify what they receive.

Licensing your own business records is legal when you hold the rights to them and handle personal data properly. We check rights source by source, and the license spells out which personal data must be removed or masked before delivery, and who does it.

Your own counsel should sign off on the license. We go through the commercial terms with you and them, so that review is quick.

Six terms decide most of the value: the liability cap and whether indemnities sit inside it; the consents you are asked to warrant; exclusivity, both how long and over what; acceptance criteria and the review window; what happens if part of the data is rejected; and when you get paid.

Confidentiality of your identity is a term to agree before you sign. If it matters to you, we make sure it is in the license.

Sometimes. Many licenses ask for exclusivity for a period. We help you keep it narrow: limited in time and limited to the data actually delivered.

Nothing. When your deal closes, the buyer pays us a finder’s fee. We tell you exactly how much, in writing, before you sign anything.

A few weeks from the first call to a signed license. The buyer then reviews the delivered data before paying, often within 30 to 60 days.

That’s common. Buyers check language source by source, so your English sources can often go ahead on their own. We never translate data and present it as original: buyers test for that, and it would put the whole license at risk.

Often, yes. Some buyers contract only with US entities; others also accept UK, EU or Canadian ones. A company with a US entity has the most options. We tell you which route fits on the first call.

Gemdata was started by Nazar Gulyk, a San Francisco tech founder with over ten years of building software companies. We have taken a company’s data from the first offer to a signed license ourselves, and we are independent of every AI lab and data company.

NDA first. A mutual NDA is signed before you share any detail about your data.

We never touch your data. We advise on calls. We don’t ask for access to your systems or for copies of your data; the 15-minute check and any follow-up calls work from what you tell us.

Your sign-off, source by source. Nothing goes to a buyer until you have approved exactly what goes, in writing.

Access on your terms. After you sign, the buyer’s team gets only the access the license grants, for as long as it grants it.

Your notes stay yours. What you tell us on calls stays under the NDA, and we delete our notes at the end of the engagement if you ask.

No contact without you. We never contact your clients, staff or a buyer about your company without your approval.