What you can ask me, and how

I answer two kinds of question, and they do not travel the same path.

Questions about the product (how something works, where a behaviour is set, why you are seeing a message) are read from a manual that ships inside your installation. So it describes exactly the version in front of you, not another one.

Questions about your memory (what a document contains, what you noted on a subject) are read from your vaults. Two separate sources, and an answer always tells you which one it came from.

How to ask about the product

Name the thing. The manual is searched by words, so "the attention slider" or "watched folders" finds something, while "that thing in the panel" does not.

A full question works better than a single word. "why does my answer find nothing" aims at a page; "nothing" aims at nothing.

A question made only of pronouns has nothing to catch on. "what can you do" or "what are you for" carry no word the manual can recognise. Ask "how does Mnemosyne work" instead, which has a page.

What I do not do

I do not invent screens. If the manual has no page on what you asked, I say so and offer what it does have, rather than describing a plausible button that does not exist.

I do not guess your memory either. An answer about your documents rests on what was actually retrieved, and it can name the file. When it finds nothing, it says so.

When the answer does not help

Three reflexes, in this order. Rephrase by naming the feature rather than its position on screen. Check the question's scope, because a muted or protected vault does not answer. And if it is a product question, ask me for the list of what the manual covers: I can give it to you.