Yes, if your user terms allow licensing de-identified usage data or you can get consent. Published buyer ranges run $20K to $480K for companies with 5 to 200 employees that hold complete sessions from real users: prompts, responses and every tool call. Coding agents, AI chat and search apps, model routers and vertical assistants with 6+ months in production qualify. Check your model provider's terms too: some restrict using outputs to develop competing models.
- Complete traces are the product: truncated or partial sessions do not qualify.
- Your own terms of service decide whether usage data can be licensed at all.
- Model providers' usage policies may restrict how their outputs are used to train other models.
What usage data buyers want
Published buyer requirements name AI coding agents and IDE plugins, LLM routers and gateways, AI chat and search apps, and AI assistants for health, fitness and work.
| Record | Why buyers want it |
|---|---|
| Full sessions: prompts, responses and every tool call | Show how real users and agents work through tasks. |
| Coding sessions with the original repository or environment | Let a buyer replay and grade the work. |
| Planning, to-do and clarifying-question steps | Capture how agents decompose tasks. |
| User ratings or preference votes | Label which answers worked. |
Where the traces usually live
Buyers name Claude Code, Codex, OpenHands, Cline, OpenRouter, Langfuse and LangSmith. Observability tools often keep fuller traces than application logs, so check retention settings before scoping.
Terms to check before licensing
The rights question is usually harder than the technical one.
- Your terms of service and privacy policy: do they allow sharing de-identified usage data with third parties for model training?
- Enterprise customer contracts: many forbid any training on customer content.
- Model provider terms: some prohibit using outputs to develop competing models. Read the current version of each provider's terms.
- Personal data in prompts: GDPR and similar laws apply to EU and UK users regardless of where you are based.
What rules a product out
Buyers list synthetic or scripted traffic, logs that cannot be stripped of personal information, and truncated or partial traces as not a fit.
Can companies outside the US license these records?
Buyers of these records accept companies anywhere. What matters more is where your users are, because their privacy law travels with the data.
The privacy law that applies to any personal data left in the records:
| Country | Main rule to check | Country guide |
|---|---|---|
| United States | Your terms; state privacy laws | US guide |
| United Kingdom | UK GDPR | UK guide |
| Ireland | GDPR | Ireland guide |
| Germany | GDPR | Germany guide |
| Canada | PIPEDA | Canada guide |
| India | DPDP Act 2023 | India guide |
| Singapore | PDPA 2012 | Singapore guide |
Next step
Related: workflow data for AI agents and data for AI evaluation. Start with a company-level description: which systems hold the records, how many years they cover and who owns them. No files are needed. Request a free review.
Origin Data Partners is independent and is not affiliated with any AI data buyer. We help owners compare buyers and find the best fit for their records. See what your data is worth.
Common questions
Can an AI startup sell its user logs?
Only where its terms of service allow licensing de-identified usage data, or where users consent. Logs must also be stripped of personal information; Buyers exclude logs that cannot be.
How much do AI labs pay for agent trajectories?
Published buyer ranges run $20K to $480K, depending on session volume and completeness.
What makes an agent session valuable?
Complete multi-turn sessions with every tool call, ideally saved with the original repository or environment. Planning, to-do and clarifying-question steps, web search beyond documentation, and user ratings all add value.
Does it matter which model we used?
Buyers list sessions run on open-weight models as nice to have. Sessions produced with closed models raise a terms question: check whether your provider's terms restrict using outputs to develop competing models before licensing them for training.
Sources and further reading
Origin Data Partners is independent and not affiliated with any buyer named on this page.