Your company’s AI. Your company’s information.

Foundry trains the core model, configures Sabina for your company, and operates her AI and supporting systems. We manage the infrastructure, updates, and ongoing support.

Sabina runs on a core model trained and operated by Foundry. Our proprietary memory and teaching systems shape that foundation around your company, while your permissions define what she may do.

When she is wrong, we answer for it. We remain responsible for the service we provide, and the record of what she did is held in our own system for your company.

Your company’s data is held for your company alone, and it is yours. If you leave, it is turned over to you. Foundry never holds your data hostage.

She does not unlearn, and we will not describe a mechanism that doesn’t exist. A packaged export is not built yet, so the form and timetable of the handover are set out in writing when you start the service. It is not a promise of model weights: your company’s data and Foundry’s model and software are different things.

Your team can teach Sabina how work should be handled. The authority system determines which actions she may take and which need approval. What she learns about your business sits in a separate system from the one that governs what she is authorized to do in it. Insight and authority are kept apart on purpose, and the purpose is this: what she learns never turns itself into an action inside your business. Nothing she reads, notices or infers writes her a grant.

Authority is a written object: a sentence you sent her, naming a capability, in force until you withdraw it. Learning is recorded elsewhere, and the gate described on the authority page reads the sentence, not the learning. She may know, from six months of your record, exactly which customer would accept a discount. Knowing it grants her nothing.