Licensing software company records to AI labs
Labs want code with the history around it: the ticket, the pull request, the review and the incident that followed. We check who owns the code and what is in it, then take it to code buyers and full-record buyers.
Last verified Oct 16, 2026
Deal summary · Confidential- Industry
- Software company
- Headcount
- 25–49
- Years of records
- 5–9 years
- What sells
- Source code and git history
- Tools
- GitHub, GitLab, Jira, Linear, PagerDuty
- Published AI lab range
- $100K–$450K per license
- Next step
- 20-minute confidential call
Example with placeholder values. Range from AI labs' published price tiers; not an offer.
The whole process, in less than 2 minutes.
What AI labs pay for and how personal details come out.
The records buyers ask software companies for
Source code and git history
Every commit and the order the work happened in.
- GitHub
- GitLab
Pull requests and code review
Review comments, requested changes and how disagreements were settled.
- GitHub
- GitLab
Tickets and specs
The request behind each change, linked to the code that answered it.
- Jira
- Linear
Incidents and postmortems
Pages, timelines and the write-up after an outage.
- PagerDuty
- Confluence
- Notion
Engineering chat
Design debates and debugging threads.
- Slack
Who buys from this industry
Public terms| Buyer | Published terms | Source |
|---|---|---|
| Collinear AI | Wants repos and pull requests linked to tickets. Gives an initial range within a day; price rises with 15+ peak staff, multi-year history and several tools. | Who Sells Data, buying programs · Observed Oct 14, 2026 |
| pre.dev | $100 per million source tokens plus $40 per million git-history tokens, paid within 15 days of funds landing. Non-exclusive by default. | Who Sells Data, buying programs · Observed Oct 14, 2026 |
| Turing, Project Lazarus | Buys code together with communications, tickets and wikis; code alone does not qualify. One payment of up to $1M, valued after sampling, usually within a week. | Who Sells Data, buying programs · Observed Oct 14, 2026 |
| HUD, DataVendor | Self-serve marketplace in beta for codebases and records. You receive sale proceeds less a 20% fee on withdrawal. | Who Sells Data, buying programs · Observed Oct 14, 2026 |
Marin Data is not an agent of any buyer listed. See every buyer program or the price index.
Check who owns the code and what is in it
Buyers look for clean title: no contractor claims and no open-source licensing issues that would block the sale (HUD, Jun 24, 2026). That means IP assignments from every founder, employee and contractor, the licenses of any copied or vendored code, and a sweep for customer data and credentials committed to the repo. Customer contracts and NDAs can limit what you share, and Slack's API terms bar third-party apps from using API data to train a large language model (Frankfurt Kurnit, Oct 1, 2026). We run these checks before a buyer sees a repo.
HUD, best platforms for selling a codebase, Jun 24, 2026 · Frankfurt Kurnit, Selling company data for AI training, Oct 1, 2026
We never license, in this industry
- Customer data, logs or database dumps committed to repos
- Secrets, keys and credentials
- Code you licensed from vendors or built for clients who own it
- HR records and performance reviews
A worked example
An illustrative case: a 30-person SaaS company with 7 years of code, reviews and incidents in GitHub, Jira, Slack and Notion. AI labs' published price tiers put a firm like this at an Estimated License Value of $100K–$450K per license. The AI lab pays that to the firm directly; Marin Data costs it nothing.
Illustrative. Range from AI labs' published price tiers; not an offer.
Questions software companies ask us
Can we license our code and keep running the product?
Yes. pre.dev licenses non-exclusively by default, and Turing says operating companies keep running while Lazarus works from a copy. Exclusivity is a term to negotiate; Labelbox Prism, for example, is exclusive by default (Who Sells Data, observed Oct 14, 2026).
What is code alone worth?
HUD's Jun 24, 2026 comparison quotes Turing at $10K–$100K for legacy code and up to $15K per modern repository, and pre.dev pays by the token. Lazarus pays up to $1M for code together with its tickets, reviews and chat. We price both routes before you choose.
What does it cost us?
Nothing up front, and nothing at all if no deal closes. Marin Data is paid by the AI lab, never out of your payment.
Find out what your records could be worth to AI companies.
Find out what your records could be worth to AI companies.
It costs you nothing.