Nothing you give us trains anyone's model

There are two ways your data could teach an AI something: by training the model itself, or by Vaulty carrying what it learned from you into another customer's agent. We do neither.

The model

The AI model behind Vaulty runs on Microsoft Azure OpenAI. Under Microsoft's data commitments for that service, prompts and completions are not used to train the models, are not shared with OpenAI, and are not available to other Azure customers. Microsoft runs automated abuse monitoring on the service; that is a safety check, not training, and it is listed among the providers we name to you.

What Vaulty learns

As your people use Vaulty, the agent learns things about your company: how you define a new customer, which account holds raw materials, what "the end of the quarter" means in your business. Those facts are stored under your company and nowhere else. Every customer's agent is isolated at the database level; one company's rules, corrections and history cannot be read by another's. We do not use them to improve the product for other customers.

What this does not mean

It does not mean the agent remembers nothing. It remembers what your own people teach it, for your own people, and you can see and delete those memories from inside Vaulty. The promise is about who benefits: only you.