Last updated: November 12, 2026
The best platforms to sell your codebase to AI companies are Turing Project Lazarus, DataVendor, Labelbox, Vortex Bridge, SimpleClosure AssetHub and Mercor. Turing is the best known direct buyer at $10,000 to $100,000 or more per codebase, DataVendor lets one repo be licensed more than once, and Vortex Bridge suits companies selling code together with its tickets and reviews.
A private codebase with years of real history is one of the most valuable things a software company can license to AI labs. Labs training coding agents want production systems shaped by real customers and real trade-offs: the pull request that was argued over, the test that kept failing, the design doc that changed halfway through. Public open-source code does not show that.
Prices vary widely. Some buyers pay a few thousand dollars per repo, and a complete operating record with code, tickets and docs can reach seven figures. Most repos, though, do not sell at all. These are the best platforms to sell your codebase to AI companies, and what each one is looking for.
| Platform | Type | What it pays | Key requirement |
|---|---|---|---|
| Turing Project Lazarus | Direct buyer | $10K to $100K+ per codebase; up to $1M for a full record | 50,000+ lines of human-written code |
| DataVendor (HUD) | Listing marketplace | Seller keeps 80% | Private, non-public code with real history |
| Labelbox | Direct buyer for defunct startups | One-time payment | Fully defunct startup codebase |
| Vortex Bridge | Broker to vetted buyers | Buyer pays directly | GitHub or GitLab repos, PRs and reviews |
| SimpleClosure AssetHub | Shutdown marketplace | Percentage on success | Startups closing or established companies |
| Mercor | Training-data company | Paid on volume and depth | Connected tools, live or insolvent |
Turing supplies coding and training data to frontier labs and raised $111 million at a $2.2 billion valuation in 2025. Its Project Lazarus, announced in March 2026, buys codebases on their own at typically $10,000 to $100,000 or more, depending on quality and complexity, and buys full operating records for up to $1 million. Repos need at least 50,000 lines of human-written code. Code written by an LLM, or with pull requests not written by a person, does not qualify. Turing says most codebase sellers finish in under two weeks.
DataVendor is run by HUD, a Y Combinator W25 company that licenses private codebases to frontier labs. Owners list repos, and when lab requests come in, matching repos are offered for license. Sellers keep 80% of each sale, the seller keeps ownership, and most licenses are non-exclusive, so the same codebase can earn several separate payments. HUD’s own guidance is candid that only a small share of codebases find a buyer, but listing costs nothing.
Labelbox buys codebases from fully defunct startups to help train and evaluate AI systems. It frames the deal around closure: a one-time transaction with no maintenance or support burden and no public reuse of the code. It values systems that reflect real development with a team, even if they do not tick every box.
Code is worth more when it comes with the tickets, reviews and conversations that explain it. Vortex Bridge matches companies with AI buyers looking for GitHub and GitLab repos, pull requests and reviews alongside Jira, Linear and Slack history. The buyer vets and anonymizes everything and pays the company directly, and Vortex Bridge says it never holds the data itself. It suits companies that want to license the full engineering record rather than a single repo.
AssetHub licenses closing startups’ source code and workspace data to AI labs and companies building training environments. One founder reported receiving about $25,000 for a codebase through it. It works on a percentage of the sale and suits founders who want the shutdown and the code sale handled together.
Mercor connects a company’s tools read-only, anonymizes the records and pays on volume, tools connected and depth. Code is one input among many, which makes it a better fit for companies with a rich mix of engineering and operational records than for a lone repo.
We reviewed buyers and marketplaces that publicly license private code for AI training, and checked each one’s requirements, prices and terms on its own pages. We favored options with published criteria and payment terms. Read more about how we research and rank companies.
Public and open-source code has no value to labs because they already have it. AI-generated code is excluded by most buyers. Check before you apply.
Pull requests, reviews, issue history, design docs and test logs multiply a codebase’s value. A repo without its history is worth much less than one that shows how decisions were made.
Code written for clients, or containing licensed third-party components or live credentials, needs checking before any sale. Our guide on whether you can legally sell your startup’s data after dissolution covers ownership.
A non-exclusive license on DataVendor can pay several times. An exclusive sale to a direct buyer pays once but may pay more. Decide which you want before negotiating.
Private, human-written code with real history, ideally with pull requests, reviews, tickets and docs. Public and AI-generated code generally does not sell.
Turing says repos typically pay $10,000 to $100,000 or more depending on quality and complexity. Most repos, though, find no buyer.
Usually not. Most deals are licenses, and on DataVendor the seller keeps ownership and can license the same repo again.
The codebase market rewards private, human-written code with real history, and the best results come from comparing a direct buyer like Turing with a marketplace like DataVendor or a broker like Vortex Bridge. For the wider picture of who buys company data, see the best AI training data marketplaces for companies.
For more guides and rankings on this topic, visit our Sell Company Data To AI hub.