Updated
For companies with private repos
Sell Source Code to AI Companies
Short answer
Yes. AI data programs pay for a copy of your private code, often with its change history and code reviews. You keep full ownership. First make sure the code is yours to license, check the open-source parts and remove any passwords or keys saved in it.
To sell source code to AI companies, your company signs a license covering copies of its private code. A license means a program pays to use a copy, and you keep full ownership. The code sits in repositories, or repos, which are the folders that hold a project's code. Programs often want each repo's commit history too. That is the log of every change. A contract lists exactly what is in and what is out.
Code raises questions that email never does. Who owns what a contractor wrote? What about the open-source code mixed into every repo? What about a secret API key someone saved in the code in 2019? An API key is a password one program uses to talk to another. This page answers each one. We bill you nothing, and the money we earn is a referral fee from programs.
- You still own the code
- Secrets scrubbed first
- No fee for your company
What programs ask for
What do AI companies look for in source code?
Rules differ by program and change often. Programs tend to start with the first five rows. The rest can make a codebase more interesting to them.
| What some programs ask for | Where it usually lives | Why they want it |
|---|---|---|
| Private repos that have never been public, mirrored or open-sourced | GitHub, GitLab, Bitbucket | Public code is already easy to get. Private systems are what they have not seen |
| 3+ years of commits from at least 10 engineers | Git history on your main branches | It shows how real code ages, and how a team cleans up old shortcuts |
| 50,000+ lines of code your company owns outright | Product repos, in-house tools, data pipelines | Big working systems look very different from neat sample projects |
| Code written by people, with tests that run | Test folders, GitHub Actions, GitLab CI/CD | It shows how a team proves its code works before it ships |
| A team of about 15 to 500 people, based anywhere | Your payroll and contractor list | Some programs set this size for code. Others use their own rules |
| Thousands of pull requests with review talk | GitHub and Bitbucket pull requests, GitLab merge requests | A pull request is a proposed change. The threads show engineers finding and fixing problems |
| Tickets, design docs, runbooks and postmortems from the same team | Jira, Linear, Confluence, Notion | A bug report linked to its fix goes from problem to solution. Those links are hard to find |
| Code in C, C++, Rust, Swift, Kotlin, Ruby, COBOL or .NET | Older systems and mobile or device repos | Some programs name these languages as a plus |
| Banking, insurance, healthcare, trading or blockchain software | Your core product repos | Some programs name these fields as a plus. Check customer data before anything leaves |
| History that mostly predates 2025 | Older branches and archived repos | Older work is likely less mixed with code from AI tools |
An honest filter
Which codebases are not a fit?
Some programs turn these down outright. If this describes your code, the quiz may point you toward your other records instead.
Forks and copies of open-source projects
A fork is a copy of someone else's project. Programs can already get the public version, so it adds little.
Student, hackathon and side projects
These rarely show a team maintaining a real system over years.
Code mostly written by AI tools
Programs want to learn from people. Output from another model is not what they are paying for.
Client work an agency doesn't own
If your contracts give the client the code, it is not yours to license.
Repos with a single author
With one person there are no code reviews and no team decisions to learn from.
Code that has been public at any point
Some programs ask that a repo was never public or mirrored. An old open-source release can count against it.
Ownership
Is the code yours to license?
Most code your employees wrote is yours to license. Code written by employees as part of their job is generally a work made for hire. That means the employer owns it (U.S. Copyright Office, Circular 30). For most companies, that covers most of the repo.
Contractors are different. The same circular sets two tests for outside work. It must fit one of nine listed types. Both sides must also sign an agreement saying it is made for hire. Software is not on that list. That is why contractor agreements usually add a written assignment, which is a signed handover of ownership. Check that you have one for every outside developer in the repos you plan to include.
If you build software for clients, read those contracts closely. Many give the client the code you delivered. Your own in-house tools and libraries you kept rights to may still be yours. A lawyer can sort each repo into yours, theirs or unclear. The unclear ones stay out.
Open-source code stays out too, because it is not yours. Each piece has its own license. The MIT License says its notice must stay in all copies or large parts. Apache 2.0 asks that people get a copy of the license and that changed files say so. GPL v3, section 5 says a changed work passed on must go under the GPL as a whole. AGPL v3, section 13 adds that network users of a changed version must be offered its code.
How these terms apply to AI training is not settled, so ask your lawyer. The practical step is simple. Run a scan that lists every outside component and its license. Leave copied-in and dependency folders out of the export, and attach the list to the contract. What a program pays for is the code your team wrote.
Get a number for your codebase
Take a short quiz on headcount, years in business, tools and location. Nobody asks for repo access, a token or a single file.
Keep out
What should you strip or keep out before any code leaves?
Each of these becomes a written exclusion or a cleanup step. An exclusion is something the contract says stays out. Your engineering lead should own this checklist.
Secrets anywhere in history
Deleting a key in a later commit does not remove it from Git history. GitHub's guide gives the first step: revoke or rotate the secret. Rotating means swapping in a new key so the old one stops working.
Scan the whole history
GitHub secret scanning checks the full Git history on all branches, plus issue comments. It is free on public repos. For private organization repos, it needs GitHub Secret Protection on a Team or Enterprise Cloud plan. Rotate first, then scan, then scan the final export again.
Customer data in the repo
Test files, database dumps, log files and seed data sometimes hold real customer records. Find and remove them, or leave those folders out.
Vendor and outside code
Paid SDKs, licensed parts and code copied from vendors belong to someone else. An SDK is a kit of code a vendor sells to developers. Their licenses may forbid sharing.
Code that maps your live security
Login code, payment code and server setup code can show how to attack systems you still run. You can license retired products and keep live security code out.
Code you treat as a trade secret
Trade secrets are private methods that keep you ahead, as long as rivals never learn them. If one algorithm or pricing engine is your edge, keep it out.
How it works
How does a source code license come together?
We match you with programs and stay out of the code. For other kinds of company records, see how businesses sell data to AI companies.
Answer a few questions
Share your team size, years open, systems and where most staff are based. No login, no token.
2 minutesSee your top 3 matches
We compare your engineering history, in general terms, with what each program publishes. Each match lists its reasons. You also get an estimate, which is not an offer.
Draw the lines and sign
You choose one program and sign its contract directly. You list which repos, years and folders are in or out. The open-source list goes into the contract.
Your team exports the code
A mirror clone backs up a GitHub repo with its full history. GitLab's project export covers issues and merge request comments, but not CI/CD variables or encrypted tokens. Bitbucket's API can list pull request comments.
The copy goes straight to the program
Your engineers send it, and we stay out of it. Developer names and emails get removed, programs say. No method is perfect, so request a sample check.
The market
What do public reports say about code deals?
None of these deals involved DillisData, and most involve closed startups. They are market evidence, not a forecast of your number.
A wind-down firm says this share of its deals now has code or data in it. Such deals were once rare.
MIT Technology Review, Oct 2026What that same firm's marketplace expects to return to startups by the end of 2026, across more than 150 deals.
MIT Technology Review, Oct 2026Dollars a closed software firm's CEO says it got for its Slack, email and Jira tickets. That was not code, but it came from an engineering team.
Slashdot citing Fast Company, Apr 2026Timing
Why does private code matter more today than later?
No one knows when labs will have enough code. Here is what argues for moving while your repos are intact.
Coding models keep being trained
Private code with real history is a source labs cannot just download.
Rare code draws the most interest
A program values what it can't download. Interest falls once a rival holds similar code.
Old repos get lost
Engineers leave, servers get wiped and archived repos get deleted. History you can export today may be gone next year.
New commits are harder to tell apart
As AI tools write more of each repo, older history that people wrote stands out. Yours is already on file.
Engineers who know the code move on
Sorting code from secrets is easier while the people who wrote it are still on the team.
Find out if your repos clear the bar
You'll see an estimate and your top matches. It only takes 2 minutes, and you can stop at any point.
Questions
Questions owners ask
Could the program's AI copy our code for a competitor?
It is a fair concern. Ask each program how it limits output, and get a legal review. You may also hold back whatever gives you an edge.
We only want to license retired products. Is that enough?
It can be, if they still meet a program's rules on lines, years and engineers. Retired systems carry years of history. They carry less security risk than code you still run.
We are a dev agency. Can we license anything?
Usually some of it. Client work often belongs to the client. But your in-house tools, your own products and libraries you kept rights to may be yours. Have a lawyer check each client contract.
Our repos include GPL code. Does that rule us out?
Not by itself. List the outside code and keep it out of the export. Then license the code your team wrote. Ask a lawyer how the licenses in your repos apply.
Our engineers use AI coding tools now. Is that a problem?
It depends how much. Code mostly written by AI tools is not a fit for some programs, while older history is a plus. Tell the program roughly when your team started using them.
Commit history has developer names and emails. What happens to those?
Programs say they strip these before using the history. Ask for that in the contract, and treat former employees' details the same way as current ones.
How much could our code bring in?
What programs typically pay runs $100K to $5M+, largely depending on software company size. Treat that as the market, not a promise. Your estimate is built for your answers.
Does DillisData or the program get access to our GitHub?
We never do. No repo ever reaches us. Your engineers can prepare an export and send it. Nobody outside your team needs a login to your Git host.
Free estimate
See what your repos could earn
A 2-minute quiz gives you a careful number based on programs' published pricing. Your code stays where it is.
No repo access needed.