Sell code for AI training: AI labs license real source code and git history. Sell a non-exclusive license, keep ownership, get paid each time.
Free to join. No public listing. A real estimate in minutes once Git is connected.
Invoicing platform, shelved 2024
48,210 lines and five years of real commits from a B2B product that never turned a profit. Tested and company owned.
AI labs pay to license real source code and git history to train and evaluate coding models. Source pays $100 per million tokens and history $40. You sell a license and keep ownership, and the same repo can be licensed again.
Coding models learn from code. To get better at writing, fixing and reviewing software, they need to see a lot of real software, and the public supply is largely used up. Open source code has been collected many times over. What labs lack is private code: the products, tools and systems that companies and developers built and never published.
That code is valuable for the same reason it was hard to write. It shows how working developers solve real problems under real constraints. And its git history shows something no finished snapshot can: how the code changed, which bugs were found, how features grew and what got thrown away. Labs license code to use as training data and to build tests that check whether a model can do real engineering work.
Real production code is the one input labs cannot generate synthetically. They pay for the real thing.
Labs are licensing real-world codebases to train and evaluate coding models, and the appetite is growing.
Commits, fixes and the way a codebase matured carry signal. You are paid for git history as well as the final source.
What matters is that the code is real and that you hold the rights.
Shelved products, dead startups, internal tools, R&D and side projects all qualify.Labs grade code on what it can teach a model. That turns the usual idea of what makes code valuable upside down. The things that make a product worth buying barely matter. The things owners worry about, like a messy history or an unfinished feature, often help.
Tests deserve a special mention. A test suite shows what the code is supposed to do and how its authors checked it, and labs use that to measure whether a model can make a change without breaking anything. A repo with tests and a long history of changes to them is among the most useful kinds of code a lab can license.
Value per license is source tokens at $100 per million plus history tokens at $40 per million. A token is a short chunk of text, roughly three or four characters of code. History usually holds far more tokens than source, so it carries most of the value: across the program, 79% of licensed tokens are history.
The specimen lot above is a shelved invoicing platform with five years of commits. Its per-license value works out at about the same as one typical repo, which earns about $1,000 to $1,500 per license. Rates can change on 30 days’ notice and never retroactively. For the depth table and an estimator, read how much is my code worth.
A lab trains a coding model by showing it large amounts of code and asking it to predict what comes next, then tests it on tasks it has not seen. Licensed code is used in both stages. In training, the model learns patterns from the source and from the sequence of commits. In evaluation, real repos become tasks: fix this bug, add this feature, pass these tests. Your code is not copied into anyone’s product, published or attributed. It becomes one small part of what a model learns from.
$100/M
per million tokens
$40/M
per million tokens
Value per license = source tokens × $100/M + history tokens × $40/M
Rates can change on 30 days' notice, never retroactively · Paid within 15 days of the buyer paying
Illustrative, conservative · Last verified 2026-10-10
Free to join. No public listing. A real estimate in minutes once Git is connected.
Yes. A license gives a lab the right to use a copy of your code for training and evaluation, and ownership stays with you. You keep ownership and full internal and commercial use, and because the license is non-exclusive, the same repo can be licensed to more than one buyer. Each license that closes pays you. For a plain guide to how code licenses work, see code licensing explained.
You do not need to clean anything up. The scrubbing happens on a review copy, so your repository is never touched. Only the repos you select are copied, and only the copy is graded and shown to buyers.
Secrets are the main worry for most owners, and with good reason: almost every real repo has a key or a password committed somewhere in its history. That is expected, and it is handled for you on the copy.
The repos you select are copied with their history through read-only access. Your repository stays exactly as it is.
Keys, tokens, credentials and personal data are removed from the copy, in the code and in every commit of the history.
The scrubbed copy is verified before anyone sees it, graded on quality, and taken to buyers in batches. There is no public listing.
And it is yours, which is exactly what makes it licensable. You decide what is shared: only the repos you choose, only code you hold the rights to. Secrets, customer data and third-party code stay out. Within the repos you submit, secrets and personal data are stripped from the review copy before anything reaches a buyer.
Owners in the licensing program earned $6M in the last 60 days, across 11,069 licensed repos.
One company licensed 326 repos through the program and earned $300,000 in the same 60 days. Most of its value came from history.
Published program figures, shown as illustrative, conservative. What your repos earn depends on analysis of the code and on buyer demand.
Yes. You grant a non-exclusive license and keep ownership. You can keep using, changing and shipping it.
No. The same repo can be licensed more than once, and you are paid each time. How often depends on demand.
The token count of your source and of your git history, how deep and real that history is, tests, and buyer demand. History carries most of the value.
That is fine. Third-party code is excluded from what is licensed. Your own code and history are what count.
Keys, tokens, credentials and personal data are removed from the review copy of both code and history and verified before a buyer sees anything. Your repository is untouched.
Request access. Connect Git. See a real per-repo estimate in minutes, then decide.
Free to join. Nothing upfront. You can withdraw a codebase at any time.