Home / Data types / Code

Private codebases with real git history for AI training

Datadhara licenses private, production codebases from the companies that own them, including startups that have shut down. Repositories come with their real git history, pull requests, reviews, issues and docs. Secrets and personal data are removed first, and every codebase is licensed for AI training under a signed agreement with the owner.

What we can source

Record typeFormatTypical labels
RepositoriesGit mirror with full historyLanguage, size, commit count, contributors, years active
Pull requests and reviewsJSON exportDiffs, review comments, approvals, linked issues
Issues and ticketsJSON / CSV exportStatus history, labels, linked commits
Docs and design notesMarkdown, PDFsArchitecture, runbooks, READMEs
Legacy systemsCOBOL, PL/I, JCL, mainframe JavaPrograms, copybooks, change history, tests

We collect to your spec, so volumes, languages and labels are agreed per project. Tell us what you need.

Example record

{
  "repo": "repo_027",
  "languages": ["TypeScript", "Go"],
  "years_active": "2021-2025",
  "commits": 14820,
  "contributors": 11,
  "merged_prs": 2310,
  "secrets_removed": true,
  "consent_ref": "src-0441 / v1"
}

Values are illustrative. Real deliveries follow the schema agreed in your spec.

How personal information is removed

Credentials, API keys and tokens are scanned for and removed across the full history, not just the latest commit. Customer data in fixtures, logs and seed files is removed, and contributor names and emails can be pseudonymised on request. See the full methodology.

What teams use it for

How licensing works

Every source signs a written agreement that covers AI training. You start with a small pilot and a dataset card, and if the data works for you we agree volume, timeline and license terms (exclusive or non-exclusive) in writing. Read the full licensing walkthrough.

Samples

These records are private by nature, so samples are shared under NDA once an owner has agreed to license. Tell us your spec and we'll confirm what we can source.

Have data like this?

Own a private codebase, or did your startup shut down? See how licensing to AI labs works. You keep ownership, and you don't pay us.

Further reading

Questions

Where do the codebases come from?

From the companies that own them: operating software companies and startups that have shut down, where the founders or an administrator control the IP. We never use public open-source code or scraped repositories.

What makes a codebase useful for AI training?

Real history matters most: years of commits, many contributors, merged pull requests with reviews, issues and tests. A typical bar is 2+ years, 10,000+ lines, 1,000+ commits and 5+ contributors, never published publicly.

Is the code sold outright?

No. The owner licenses a cleaned copy for AI training and keeps ownership of the IP. Terms, including exclusivity, are agreed per licence.

Tell us what your models are missing

A few lines is enough: the domain, the format, and roughly how much.

Request data →