The labs training frontier coding models pay for real production source code and git history. Licences are non-exclusive, so the same repository earns again every time it is licensed. You keep full ownership, you keep using the code, and it costs nothing to find out what yours is worth.
# What the labs are actually buying $ pangea estimate --repo ./internal-platform --include-history scanning ........... source code + full commit history scrubbing .......... secrets, credentials, personal data pseudonymising ..... contributor identities working on ......... a copy — your repo is never touched → estimate returned as soon as Git is connected # Non-exclusive. The same repo can be licensed again, and again.
01 Why now
Frontier coding models have to be trained and evaluated on genuine production code — not just the finished source, but the history behind it. How a feature actually shipped. How a bug was actually found and fixed. How a codebase matured over years. That record cannot be manufactured, and the labs are short of it.
Authentic production code is the one ingredient a lab cannot generate synthetically at the quality it needs. So it pays for the real thing.
Your commits, fixes and tickets carry signal the final source doesn't. Git history is licensed and paid for separately from the code itself.
A licence to one buyer doesn't close the door on the next. The same repository can be licensed repeatedly, and you're paid each time.
02 What you keep
This is the part most people get wrong on first read. Nothing changes hands permanently. You are granting a non-exclusive licence to code you continue to own and continue to use.
You keep full ownership of everything you submit. You carry on using it internally, commercially, and with clients. You can keep developing it. Nothing is assigned.
You decide which repos take part, one by one. Git access is read-only. You can withdraw a codebase before it goes to labs, though not once it has been presented to a buyer.
Secrets, credentials and personal data are stripped from both the current code and its history. Contributor identities are pseudonymised. All of it happens on copies — your original repository is never touched.
03 Two routes
Same platform, same rates, same terms — two intake routes, because what counts as licensable code is different depending on who owns it. If you're a founder, CTO or solo developer with code you own outright, take the first. If you run an agency or studio, take the second.
For founders, CTOs and individual developers sitting on code they own: internal tools, retired products, R&D, side projects, anything built and paid for by you.
For agency and studio owners with years of delivery behind them. The qualifying material is your own: internal tooling, frameworks, R&D and projects where you hold the rights.
Both links above are referral links. If you sign up through them, Central Intelligence earns a referral fee from Pangea. It costs you nothing, and it does not change your rates or terms. We're pointing you at this because the economics stand up on their own — but you should be told, not left to find out.
04 Eligibility
The single fastest way to waste your own afternoon is to start an application for code you aren't actually free to license. The rules are simple and they are strict.
05 The economics
Pangea publishes its rates rather than quoting case by case, and describes them as conservative by design. Source and history are priced separately, which is why a long-lived repository with a real commit record is worth considerably more than a snapshot of its final state.
These are Pangea's own numbers for their platform, not ours, and not a projection of what any particular codebase will earn. Treat them as evidence the market is real, not as a forecast.
| Measure | Figure | Basis |
|---|---|---|
| Total paid to licensors | $600,000 | across 380 repositories, 13.6B tokens |
| Single company, cited example | $130,000 | across 22 repositories |
| Agencies | Six figures | Pangea's stated range for agency licensors |
06 How it runs
The work on your side is measured in minutes. The elapsed time is measured in weeks, because it depends on a buyer arriving — and nobody can promise when that happens.
[01] request access Basic details about you or the company. [02] sign + connect Sign the agreement, grant read-only Git access. GitHub, GitLab or Bitbucket. Estimate is immediate. [03] select + grade You pick the repositories. They are copied, scrubbed and graded. Hours, up to four weeks. [04] licensed + paid Presented to buyers. Deals typically close in 2–16 weeks. Payment within 15 days of each close. note: non-exclusive — step 04 can happen more than once # Withdrawal is possible before submission, not after presentation.
The estimate arrives as soon as Git is connected, costs nothing, and commits you to nothing. If the answer is that your code doesn't qualify, you'll know inside an afternoon rather than wondering for a year.