Companies are paying for real code, product history and workflows to train AI. What your startup left behind can still be licensed, even after you've closed.
Recent AI data deals
Reported deals involving large platforms. They show the market is real; they do not set a price for any one archive.
We find companies willing to pay for what your startup left behind, and you set the terms.
Based on published rates for expert-produced AI training data.
Replacement cost: what a lab would pay to commission this work from scratch, discounted because an existing archive is messier than data built to order.
Assumes 1,600 working hours per person per year.
Illustrative estimate in US dollars. Not an offer, not a valuation, and no sale or payment is guaranteed. Real value depends on what your records contain, who wants them, and what rights you can grant.
From the codebase to the Slack history, most of what your startup built can still be licensed.
Get a detailed assessmentSource code with its full history: the commits, the reviews, and the bugs that were found and fixed along the way.
Do not send raw records during an assessment. We start with counts and descriptions; data moves only after a deal is agreed.
List the tools your startup used and roughly what's in them. You don't upload anything or hand over logins. We'll tell you if it's worth licensing and give you a ballpark figure.
We work out who owns the data now, whether that's the company, its founders, its investors or a liquidator, and get it signed off in writing. Then we replace names and contact details and package it so buyers can see what they'd be paying for.
We take it to companies, put their offers side by side and push for the best price. Your data doesn't go anywhere until you sign, and you're paid on the terms in that contract.
We swap names, companies and contact details for consistent labels and take out anything sensitive. The work itself stays intact, because that's the part buyers pay for.
An example from a team Slack thread
Priya Shah merged the billing fix for Lumen Health. Send the release notes to priya@lumenhealth.io.
Priya Shah confirmed the patch held in production.
The same thread once we've prepared it
Person 07 merged the billing fix for Company 03. Send the release notes to Email 07.
Person 07 confirmed the patch held in production.
Priya Shah becomes Person 07 everywhere in the set, so a thread still follows one person. Companies and email addresses get labels too.
Passwords, payment details and anything you've ruled out come out before a buyer sees a single record.
The problem, the decision and the outcome stay joined up. That's the part an AI model learns from.
Made-up example. Replacing names is one step, not a guarantee: a record can still point to someone through context, so we review for that as well, and the limits go in the agreement.
Straight answers
Nothing is uploaded to start with. These are the things worth checking before you commit to anything.
Often, yes. What matters is that the records still exist, that someone can still get into them, and that whoever owns the company's assets now agrees. That might be the company itself, its founders, its investors or a liquidator. Working that out is one of the first things we do.
Whoever owns the data. Usually that's the company, and the money then goes to its shareholders or creditors under the normal rules. If the company is being wound up, the payment goes through that process. If its assets were sold or handed over when it closed, the new owner is the seller.
It depends on what the records contain, how much there is, whether the rights are clear, and what buyers are looking for at the time. The estimator on this page is an illustration built from published rates, not an offer. An assessment gives you a better-informed answer, still not a guaranteed figure.
Yes. We have buyers already lined up, including several from the AI labs, and we know what they are looking for.
Chats, emails and tickets are full of other people's details. They are replaced with labels before a buyer sees anything, and some records can't be offered at all, for example where customers were promised their data would never be shared. We check that before anything is listed.
No. You can leave out tools, channels, customers or whole parts of the product. The agreed scope sets exactly which records are offered, and it's fixed before anything moves.
Not on its own. A record can still identify someone through context, especially in a small team where everyone knows who did what. Replacing names is one step. The review has to cover identifying context and sensitive fields as well, and the limits belong in the agreement rather than in a promise.
Not always. Many tools keep data for a while after an account closes, and exports or backups often survive. Tell us what you used and roughly when you shut it down, and we'll tell you what's likely still recoverable.
Don't send raw records, credentials, API keys or payment details through this website or by email. Start by telling us which tools you used and roughly what's in them. Data moves only after a deal is agreed, and into the environment named in that agreement.