What is your codebase worth as AI data?

Get a first-pass range for licensing private code to AI training and evaluation buyers. No sign-in, upload, or AI calls. It runs in your browser.

Original lines of code

Count code your team wrote. Leave out generated files, vendored libraries, and dependencies.

Import a CLI reportRead in this browser. Never uploaded.

Technical complexityLowMostly CRUD screens, scripts, or configurationModerateProduct code with real business logic and integrationsHighCompilers, databases, distributed systems, numerics, or security-critical code

Was the code ever public?

Version history

Development records you can supply

A Git repository holds commits, but not the issues, pull requests, reviews, or CI results stored by GitHub or GitLab. Check only what you can export.

Indicative range

Enter the number of original lines to see a range. Unknown or zero code has no estimate.

An AI data licensing estimate, not company value, replacement cost, a buyer quote, or a guaranteed payout. Buyer demand, usable code, provenance, and final terms all need review.

Submit for review

How the estimate works

A transparent house heuristic. Every factor is listed here and shown in each calculation.

Base rate: $100 to $400 per 1,000 eligible lines
Size: The first 100,000 lines count fully, the next 900,000 count at 50%, and lines beyond 1,000,000 count at 20%.
Complexity: Low ×0.6, Moderate ×1, and High ×1.5
Prior visibility: Always private ×1, Previously public (removed) ×0.6, Not sure ×0.4, and Public now ×0.15
Development evidence: Weights add up to 100%: Source code 40%, Version history 25%, Pull or merge requests 7%, Human review discussion 10%, Issues or tickets 7%, CI and build results 6%, and Automated tests 5%. Partial history counts 12%.
Minimum: No range below 2,000 eligible lines or when the high end is under $500.
Rounding: Low end rounded down and high end rounded up to two significant figures.

These are Datakeep house assumptions (model 1.0.0, October 8, 2026). They are not fitted to completed sales. We do not yet have a validated transaction dataset, and we change the version whenever a value changes.

Not modeled yet: history length, contributor count, technical domain, buyer demand, code quality, and licensing encumbrances. We review these by hand.

Archives aren't discounted for being archives. Only missing evidence lowers the range, so a ZIP with full history and exported records can match a hosted repository.

From estimate to offer

Submit the codebase

We review it

We check provenance, how much of the code is usable, and whether an AI buyer wants it. Final terms are agreed before any deal.

You're paid after the buyer pays

Payment is contingent. You receive your share after Datakeep receives payment from the AI buyer.

Default terms

Exclusive AI data rights by default
You keep ownership and use of your code
Payment after we receive the AI buyer's funds
Referral rewards of up to 5% of the amount paid to a referred owner, on qualifying projects

Have other data?

Salesforce, CRM, and other startup operating data are reviewed separately for now. Tell us what you have.

Contact