Responsible AI

No. Not by us, and not by the model providers we use on your behalf. We use these providers through their business and API terms, under which submitted data is not used to train their models.

This is not a setting a customer has to find and disable. It is how the platform is configured.

By the model provider handling that request. The providers we use — Anthropic, OpenAI and Together AI — are US companies, which means enabling an AI capability is the ordinary case in which data leaves the EU.

Only the content required for the request is sent: the prompt, the relevant context, and whatever your system attached to it. Your database is not exposed to a provider, and no provider holds a standing copy of your data.

If your organisation cannot accept processing outside the EU, AI capabilities can be left disabled and the rest of the platform is unaffected.

No. Retrieval works by finding the relevant passages from your own content at the moment a question is asked and supplying them with the request. The model reads them to answer and does not retain them.

The stored representations of your content used to make that retrieval possible live in your system's own storage, inside our EU infrastructure — not with a model provider.

Yes. AI capabilities are part of what a system is assembled from, so a system can be built without them. Nothing else in the platform depends on them.

No. The platform does not use AI for automated decision-making that produces legal or similarly significant effects on individuals, and it does not profile people. Those are not capabilities we offer.

The AI capabilities that exist assist a person doing their work. The decision, and the accountability for it, stays with the person.

AI output is presented to a user, who acts on it or does not. Where an AI capability produces structured data for a system to act on, that flows through the same validation and permission checks as any other input — an extracted value does not bypass the rules that govern a typed one.

Actions taken in a system are audit-logged against the user who took them, whether or not an AI capability helped produce them.

Language models produce plausible text, which is not the same as correct text. They can be confidently wrong, they can misread a document, and they can miss context a person would have.

So: treat AI output as a draft to be checked, not as an authority. Use retrieval over your own content where the answer must be grounded in your records rather than in a model's general knowledge. Keep a person accountable for anything that matters.

We would rather set that expectation here than have it discovered in production.

Provider credentials are held in the encrypted secret vault described on the Security page, scoped per system and environment, and never present in source control or in plain files.

An AI capability runs with the permissions of the user who invoked it. It does not have a privileged path to data the user could not otherwise reach.