For AI companies

Engineering data with a paper trail.

For coding-model developers, agent companies, evaluation providers, and enterprise AI teams that need data they can explain — to their researchers and to their counsel.

Where this data is used

Each use has its own licence scope, so you only pay for — and are only bound by — what you need.

01

Coding models

Pre-training and fine-tuning on real change histories, reviews, and tests from commercial codebases.

02

Coding agents

Trajectories and issue-to-fix loops that show how engineers navigate, edit, and verify.

03

Evaluation

Held-out, executable task suites kept out of public corpora to limit contamination.

04

Reward modelling

Review outcomes and accepted-vs-rejected revisions as preference signals.

05

Enterprise AI

Domain-specific engineering data for internal assistants, licensed for internal use.

06

Research

Scoped research licences for studying code-model behaviour.

From requirements to delivery

  1. 01RequirementsData types, languages, volume, intended use
  2. 02 · decisionSourcing & reviewSuppliers matched; rights and QA recorded
  3. 03Sample evaluationTime-limited access once we approve your request
  4. 04LicenceWritten scope, term, exclusivity, and price; signed online
  5. 05AccessPay the invoice, then read-only access to a private copy of the repository

What you receive

Every engagement comes with a documented record

  • A dataset card on the listing: contents, schema, version, and known limitations
  • The rights review on the listing: source, restrictions, permitted uses, and review scope
  • The quality checks run on the licensed version (counts shown only when verified)
  • A written licence defining scope, term, exclusivity, and permitted uses
  • Read-only access to a private copy of the reviewed repository, for the GitHub or GitLab accounts you choose
Engagement models
Catalog licence
License a listed dataset version on agreed terms
Custom sourcing
We find suppliers and data to your specification
Preparation
Scrubbing, normalisation, filtering, splits
Recurring
Ongoing refreshes where suppliers can provide them

Our commitments

What we will and won’t say

We will

  • Describe what we reviewed and what was out of scope
  • Show limitations alongside strengths
  • Tell you when we can’t source what you need

We won’t

  • Promise model performance improvements
  • Claim a review is a legal guarantee
  • Publish record counts or quality scores we haven’t verified

Questions buyers ask

What kind of data do you provide?
Software engineering data for coding models, AI agents, and evaluation: issue-to-fix histories, code changes and pull-request metadata, review and debugging examples, code–test pairs, evaluation suites, and authorized engineering workflow traces. Availability depends on what suppliers can license; we also source to specification.
Can I see data before committing?
Usually. Where the listing has a sample, you get a scrubbed, time-limited one to evaluate once we approve your request, before any licence. Samples are never published openly.
How is pricing set?
Pricing is agreed per engagement based on volume, exclusivity, term, intended use, and any preparation work. Tell us your requirements and we will send a written proposal.
Do you guarantee the data will improve my model?
No. We describe what each dataset contains, how it was checked, and its known limitations, and we support evaluation before purchase — but we do not promise model performance outcomes.
Is the data legally cleared?
Every listing goes through a documented rights review of ownership, licences, restrictions, and permitted uses, and the scope of that review is recorded. A review reduces risk; it is not a legal guarantee. Your licence agreement sets out the warranties that apply.

Start with your requirements.

A short form, then a conversation. No commitment until there’s a written proposal you’ve agreed to.