Can users access (read/see) the data that passes through a workflow? (re: LLM system prompts)

Hello,

I am creating an app that is essentially a collection of AI tools. Each tool takes user input and runs a workflow that contains the system prompt, then takes the user to a chat interface where they can iterate on the tools output.

In order to get the chat interface to work, I have a data type that contains the system prompts which populate dynamically into the workflow depending on the tool type.

What I have noticed, however, is that the system prompt data type needs to have no privacy settings in order for the workflow to access it.

I don’t want the system prompts being exposed to the user, so my question is:

Can I ensure that my users can’t view the system prompt while still letting them use it for the purposes of the workflow?

Backend workflow.

Copy paste it to the backend.

Users can’t see what’s not sent to them!

2 Likes