An agent should never answer a user's question with data that the user would not ordinarily have access to. FoundationaLLM can be used to create an expert agent who can answer questions using a variety of enterprise data sources. In this example, a user first asks a question about HR policy which is defined in a collection of documents that is available to all signed-in users. Then the user asks about her specific compensation package, which the agent determines it needs to query the compensation database on behalf of the user. FoundationaLLM integrates with the row-level security capabilities of the database to make sure that the user can only get the compensation data she is supposed to see. In the last case, the user asks about the CEO’s compensation and when the agent tries to query for that data, no records are returned because the user does not have permission to see that data. The agent recognizes that permission was denied and politely declines to share the data.