Is my company’s data used to train your models — or anyone’s models? Where exactly does it go when I click generate?

The full executive answer

No. Neither we nor our model providers train on your data. We do not train or fine-tune any model of our own — there is no training pipeline in the product, period. And we reach Anthropic and OpenAI exclusively through their commercial APIs, where both providers’ enterprise terms state they do not train on API inputs and outputs by default. That posture is written into our model registry as a data-handling declaration per model, so it is a documented governance commitment, not a verbal assurance — and for customers who need stronger guarantees, both providers offer zero-data-retention arrangements on enterprise agreements, which we can pass through.

What actually travels when you click generate: a structured context block about your organisation — assessment scores, profile facts, and short retrieved snippets from documents you chose to upload — assembled on our servers. Before any of it leaves, untrusted text is scrubbed: personal data like emails, phone numbers, card and ID numbers is redacted, and the text is neutralised against manipulation. The model returns a structured result, which is validated and stored in your tenant, isolated by row-level database security. Your documents themselves stay in our storage; the model sees short excerpts, not your archive.

And there is a nuclear option built in: any workspace can enable a no-third-party-model policy. With that flag set, the engine refuses to send anything to any external model — you keep the deterministic analytics and lose generative output, by your own choice. We built that because for a regulated CDO, an honest "no analysis" beats a silent data egress. Your data remains yours: we claim no rights to it, and deletion means deletion, including the derived embeddings.