An agent for data security #40

User Story: "As a user within an organization, I want to utilize a data privacy agent in order to safely use LLM without the risk of indirectly sending internal processes, ideas, and plans to the cloud."

We create a new agent: First, it checks whether the data security of the chat thread is already restricted. This is the case, e.g., when a data source in the RAG process requires such restriction. In these cases, the agent takes no action.

When the data security is not restricted yet, the entire thread is checked. Here, the agent is supposed to determine whether the current interaction indirectly reveals information—such as when a user (presumably anonymously) asks questions related to internal processes, ideas, and plans. If this occurs, the agent stops sending this thread to the cloud and instead restricts the data security setting to self-hosted models only.

Consequently, this agent should use a self-hosted model.