Security and data

What the agent sees

• The user’s request, typed or dictated then transcribed.
• The session language and the current screen, as declared in the model.
• The state of the current flow: step reached, step completed.

What the agent never sees

• The content of the user’s screen: fields, values, documents.
• Your application’s business data, except what you explicitly declare to the model (for example: signed in or not).

The language model: agnostic, minor, replaceable

NAVAGE is LLM agnostic. You pick the provider, impose your own, or use none. Without a model the assistant still works: commands, local recognition of requests, answers extracted from your content with their source.

The model has only two roles: understanding a request that local recognition did not resolve, and phrasing an answer from the passages of your site it is given, and nothing else. It never receives the screen or your application’s data, and it never decides a flow: guidance is computed from the navigation model.

An answer that rests on no provided passage is rejected. The model therefore cannot invent what it was not given.

What is kept

• Only what serves guidance and measurement: flow state, language, the requests made and how they were resolved.
• Never a form entry, a password, a document or the content of a screen.
• Dedicated hosting or your own infrastructure depending on the plan. Precise terms are part of the contract.

Your rights and your users’ rights

Access, rectification, erasure, objection: contact@navage.ai. Details in the privacy policy.