Sell source code to AI companies: rates and rules

You sell source code to AI companies by licensing a copy, so you keep the code and can keep shipping it. Pangea licenses private repositories and their git history to AI labs at published rates of $100 per million tokens of source code and $40 per million tokens of git history. Labs pay because, in Pangea’s words, “authentic production code is the one input labs can’t generate synthetically”.

Pangea pays per token, for source and history separately

The rates can change on 30 days’ notice, never retroactively. Pangea’s own illustration is roughly $1,000 to $1,500 per repository at typical depth, which it labels “not a quote”. The worth calculator applies these rates to your repositories.

Pangea’s license is non-exclusive, so the same codebase can be licensed to several labs. An exclusive license or a full acquisition needs a separate written agreement on pricing.

Git history has its own rate, so don’t squash it

A repository with five years of commits, branches and merges carries more licensable tokens than the same code squashed into one commit. Rewriting, squashing or pruning history before you submit removes the part with its own price. The worth calculator explains how source and history are each counted.

Size and history set the price, and generated code doesn’t count

Because the rate is per token, more hand-written code and a longer history mean a higher figure. Pangea excludes generated files, vendored code, binaries, data dumps and third-party code from the valuation, so a repository that is mostly a framework scaffold and a vendor folder has little left. You don’t need to remove those files first: Pangea asks for repositories as they are and strips them on its side.

Pangea won’t take code that was public, forked or client-owned

Pangea’s page lists what not to submit:

• Previously public code. A repository that was public and later made private is out.
• Forks. The upstream code belongs to someone else.
• Client work. Agencies and consultancies can’t license code they wrote for clients.
• Confidential data or conflicting third-party rights. For example, a former co-founder with no IP assignment, a contractor who kept copyright, or an acquirer who bought the IP.

Secrets, customer data and personal data are stripped before a buyer sees the code. That handles a stray credential in the history. It doesn’t fix ownership: Pangea’s agreement makes you responsible for owning what you submit, with no cap, so if ownership is unclear, ask a lawyer before you submit.

How a Pangea sale runs, from estimate to payment

1. Request access. You give your details and connect Git. Pangea returns an estimate.
2. Sign and connect. You review the agreement, sign, and grant read-only access to GitHub, GitLab or Bitbucket. You choose the repositories.
3. Grading. Pangea scrubs and grades a separate copy, in a few hours to four weeks.
4. Licensing. Accepted repositories go to labs in batches. Licensing typically takes 2 to 16 weeks and is not guaranteed.
5. Payment. Within 15 days of the buyer’s payment, once your tax and payout details are complete.

You can withdraw a repository before it goes to labs. After that it stays in the active process, and you can remove it from future opportunities with 30 days’ notice.

Licensing the code is separate from selling the company

When a company is sold, the code usually goes with it. If you license the code first, tell the acquirer and have a lawyer check how the sale contract treats the license. Once the code belongs to someone else, you can’t license it. For a SaaS you plan to list for sale, see selling the code separately from the business.

A company that is closing has a second asset besides the code: Slack, tickets, docs and email, which buyers of company records license separately.

Keep the unsquashed history, proof of privacy and the IP paperwork

• The repository with its full git history, unsquashed and unpruned.
• Proof it was never public: your GitHub, GitLab or Bitbucket settings history, or your own records.
• IP assignments from co-founders and contractors, and the contract if an acquirer or client was involved.
• Read-only access ready to grant, from an account the owner controls.