Source system · Engineering
Sell your company's private GitHub repositories for AI training
Outrich Data is a direct buyer of private GitHub repository data from companies with 50 or more employees. Source code, commit history, pull request reviews and issues can often be licensed to AI labs once secrets, personal details and third-party code are removed. Your engineers produce the export. Outrich Data does not connect to your organization.
What GitHub holds that AI labs value
A private repository holds more than the current code. The commit history shows every change with its message. Pull requests hold the diff, the review comments, the requested changes and the revised commits that answered them. Issues describe the bug or feature that started the work.
AI labs have already trained on public code. Private repositories are code their models have not seen, which makes them useful for training and more useful still for evaluation, where an unseen code base is the point. Production code also differs from open-source code in its tests, its internal tooling and its handling of real business rules.
Review history is the scarcest part. A pull request that was challenged, revised and then merged shows engineering judgment being applied to a concrete diff. An issue linked to the pull request that closed it, with tests that pass, is a task with a verified solution.
What a usable export looks like
A mirror clone of each repository carries every branch, tag and commit. Issues, pull requests, review comments and discussions are not part of the git data. They come from the REST or GraphQL API, or from an organization migration archive, which GitHub documents as including repositories along with their issues, pull requests, comments and related metadata. Workflow run logs are kept for a limited, configurable period, so older runs may be gone. What you can export depends on your plan and role; check GitHub's current documentation.
Cleaning has to cover history, not only the latest commit. A key deleted last year is still in an old commit. The usual approach rewrites history on a copy to remove secrets, large binaries and vendored directories, and maps author names to pseudonyms. The original repository is never altered.
Outrich Data does not connect to your GitHub account, and it does not ask for credentials or an access token. Your company exports the agreed records itself, removes what is out of scope and approves the packaged dataset in writing before delivery.
Records and fields in a GitHub dataset
| Record | Typical fields | Why it matters to AI labs |
|---|---|---|
| Source tree | Files, paths, languages | Production code a model has not seen |
| Commit history | Diff, message, timestamp, author pseudonym | How the code changed, step by step |
| Pull requests | Title, description, diff, merge status | A proposed change with its rationale |
| Review comments | File, line, comment, resolution | Engineering judgment tied to exact lines |
| Issues | Title, body, labels, linked pull request | The task behind the change |
| Tests and CI configuration | Test files, workflow definitions | A way to verify that a solution works |
Removed before licensing
What comes out of a GitHub export first.
- Keys, tokens, passwords and certificates, in the current tree and in every past commit
- Environment files and infrastructure configuration with real hosts and accounts
- Author names and email addresses in commit metadata
- Customer data in fixtures, seed files, database dumps and logs
- Vendored dependencies and any third-party code you did not write
- Security-critical logic your security lead flags, and open vulnerability reports
- Licensed binaries, datasets and model weights stored in the repository
Data title search
Rights and contract checks.
Vendor terms for business software generally treat the content a customer puts into the service as the customer organization's data. Read your own agreement with the vendor to confirm. That settles ownership between you and the vendor. It does not settle it with your employees, customers and partners, which is what these checks cover.
- Employee and contractor assignment. The company must own the code. Check that invention assignment or work-for-hire terms cover everyone who committed.
- Open-source licenses. Code under an open-source license stays under that license. You can license only what your company owns, and copyleft components need particular care.
- Customer-funded work. Code written under a statement of work may be assigned to the customer.
- Source licenses and escrow. If customers hold a source code license or an escrow arrangement, read its confidentiality terms.
- Generated code. Ownership of code produced by AI coding tools is unsettled. Raise it with counsel if a large share of the repository was generated.
This page is general information, not legal advice. Whether your company can license a given dataset depends on your contracts, your privacy notices and the law where you and your customers are. Review it with your own counsel before signing anything.
Value
What moves the price of GitHub data.
A written offer follows the rights review. The data valuation calculator gives an illustrative range first.
| Factor | Effect on value |
|---|---|
| Never public | Code that was never open-sourced or leaked is what labs pay for. Public repositories add little. |
| Review culture | Repositories with substantive pull request review are worth more than those merged without comment. |
| Runnable tests | A test suite that passes lets a lab verify solutions, which raises value sharply. |
| Third-party share | Heavy vendoring or a large copyleft footprint shrinks what is clear to license. |
Related data types and industries
Questions
Selling GitHub data: what companies ask.
Would a competitor end up with our code?
Are our public open-source repositories worth licensing?
What about secrets we rotated long ago?
Does Outrich Data need a GitHub App or deploy key?
GitHub is a trademark of its owner. Outrich Data is not affiliated with GitHub.
Find out if your data qualifies.
Eleven questions, about five minutes, no call required. You get an answer on screen and a written follow-up.