(Data licensing, B2B)
Validated training and evaluation data with known provenance, documented rights and defined license terms. The plan starts with direct agreements and a proposed software engineering pilot. No dataset is available to buy on this site.
For AI companies
Discuss a pilot(BYO Data Marketplace)
Direct agreements first
Self-serve catalog later
Licensed data with known provenance
Direct B2B contracts come first, to validate demand and pricing. A self-serve catalog follows once quality, rights and licenses are standardized. Nothing here is a live catalog.
(01) Data products
What you can license
- 01
Training datasets
Licensed, curated examples for training.
- 02
Task, solution, correction
Validated traces of how a task was solved, corrected and checked.
- 03
Preference and evaluation
Human feedback and model assessment datasets.
- 04
Expert datasets
Specialized contributions, with expertise and quality assessed separately.
- 05
Commissioned tasks
Data collected to your requirements through demand-led tasks.

(02) License terms
A purchase is not unrestricted
- 01
Allowed purposes
What the data may be used for - 02
Term
How long the license runs - 03
Training permissions
Whether and how models may train on it - 04
Redistribution
Whether it may be passed on - 05
Security duties
How it must be stored and protected - 06
Restrictions
Anything else that is excluded
(03) BYO Data Engine
Checked before it is licensed
Removing names and identifiers does not guarantee anonymity, so sensitive material gets additional restrictions and controls. Volume alone is not training value.
-
(A)
Contributions
Authorized by category or item -
(B)
Rights check
Permitted uses, third-party data -
(C)
Privacy filtering
Identifiers removed, sensitive data restricted -
(D)
Quality evaluation
Independent checks, abuse prevention -
(E)
Structured examples
Task, solution, correction, outcome -
(F)
Provenance record
Permission scope, licensing history
(04) The pilot
Software engineering data
A proposed first category, not a permanent limit and not an available commercial dataset. Software outcomes can often be checked by reproducing failures, running tests and verifying fixes.
- 01
Validate demand
Ask AI teams which data, rights and prices they need.
- 02
Controlled pilot
Build a small, permissioned collection of validated software examples.
- 03
Commercial proof
Seek buyer interest, ideally a paid pilot, and test payout economics.
- 04
Scale
Add fields and automation only once the unit economics hold.