What documents can be used with an AI agent for SCM? And how to achieve the best answers?

Summary:

I'm testing the AI Agent for RSSP.

Is there documentation on what type of documents you can upload and what it can read on those, does it need to be all text, or can it read text within a screenshot or slides, pictures? Can it also read links in the document and provide it to users? e.g. asking location of policy document

And are there any controls over what the AI agent can say or a way to train its answers? I've noticed it can give incorrect/inconsistent answers, do we solve this by feeding more documents or is there an optimal amount or solution to this?