RL ATLAS

How to sell source code to AI companies

What makes a private codebase useful to AI buyers, how to prepare it, and which rights and deal terms to check.

AI companies can learn from more than the final version of a program. A private repository may show how a team turned requests into changes, tested them, reviewed mistakes, and maintained a working product. That history can make a codebase useful for training or evaluating coding agents.

If you want to sell one, start by describing the asset and confirming you can grant the rights a buyer needs. A repository's size alone says little about its commercial value.

What a buyer can use

A code snapshot shows the result of engineering work. A repository with connected records shows the work itself. Commit history, pull requests, reviews, issue trackers, design notes, and setup instructions can all make the package more useful.

Those parts answer different questions:

AssetWhat it helps a buyer understand
Commits and branchesHow the implementation changed over time
Issues and pull requestsThe request, proposed solution, feedback, and resolution
Tests and CI configurationWhether a change can be checked reliably
Design and incident notesWhy the team chose an approach or changed course
Setup instructionsWhether the buyer can reproduce the software's behavior

The links between records matter. An issue that points to a pull request, test, and eventual outcome is easier to turn into a meaningful agent task than four unrelated files. A buyer building a runnable evaluation may also need a stable starting state and a way to verify the result. A buyer training a model may instead want a larger history of authentic decisions. Ask which use is intended before preparing an export.

Inventory the code before quoting a price

Make a short asset sheet for each repository or product line. Record the language, domain, years of history, approximate number of contributors, related ticket and documentation systems, and whether the project can still be built. Include what you can exclude. A retired internal tool may be easier to scope than an active product connected to live customer systems.

Then prepare a representative sample that shows the relationship between a task and its outcome without disclosing the whole archive. Buyers can assess fit from examples; you can reserve a full delivery for an agreement with clear terms.

There is no reliable public price per repository. Codebases and transaction types vary too much for a single reference price to forecast what yours will earn. Compare actual offers on the payment you receive, the preparation you must do, and the rights you give up. A marketplace listing or valuation is an early signal, not a completed sale. Our guide to AI data marketplaces explains the distinction.

Clear ownership and sensitive material

Check who owns each part of the code. Employee and contractor agreements, acquired IP, third-party components, and customer-funded work can all affect what the company may license. Keep a dependency inventory and review the terms of components included in an export. If ownership is unclear, resolve it before promising the asset to a buyer.

Search the full history for credentials, private keys, customer records, personal information, and third-party confidential material. A clean current branch does not prove an old commit or issue is safe to deliver. Decide what must be removed, masked, or excluded, then verify the delivered copy. Keep the original archive intact under appropriate controls while you prepare a separate review and delivery copy.

These checks are specific to the asset and agreement. They are also part of the buyer's assessment: an interesting codebase with uncertain rights or unusable documentation may be hard to purchase.

Decide what the agreement grants

An outright sale transfers ownership of the specified asset. A license lets the owner retain it while granting defined uses. An operating company may prefer a license for a bounded export; a company closing may prefer a sale if no entity will remain to administer future licenses.

Write down whether the buyer may use the code for model training, evaluation, building simulated environments, or some combination. Set the term, permitted recipients, sublicensing rules, security requirements, and any exclusivity. Specify whether related tickets and documents are included, and how later updates will be handled. If multiple buyers may receive the same asset, make that explicit.

A practical first pass

Choose one codebase with a clear owner and a well-documented history. Assemble an asset sheet, map a few real tasks to their code changes, and test whether the project can be run or inspected. Review the rights and sensitive material before sharing samples. Then compare buyer and marketplace interest against the effort needed to complete a transaction.

DataVendor lists repositories among the types of supply its marketplace accepts; its profile also explains how a listing differs from a completed purchase. Other buyers may require a different package. Let a concrete buyer use case determine the final export, while keeping ownership and usage terms clear from the start.

← Back to the blogExplore the directory ↗