Look inside the machine

Most software asks you to trust it. Dex would rather show you.

Type /xray after any answer and Dex stops working and starts explaining: what it read, which of your files it opened, what it was told to do before you typed a word, and why it took the route it took. Not a generic help article — a walkthrough of your last conversation.

It's the most useful word in Dex, and the least used. Two reasons to make it a habit: it's how you learn whether an answer deserved your trust, and it's the fastest education in AI you'll get, because the example is always something you actually cared about.

The journey of one question

Before the example, the shape of the thing. Every answer Dex gives you takes the same five steps — and /xray is simply Dex narrating the ones you didn't see.

Steps two and three are the ones people are surprised by. Step four is the one worth arguing with.

What it looks like

You ask something ordinary. Then you ask how it happened.

That last section is the one worth reading twice. Dex separates what it looked up from what it decided — and those are different kinds of wrong. One you check; the other you argue with. Knowing which you're dealing with is most of the skill of working with AI.

Notice the shape of it: the left column is checkable in seconds, and the right column is where a wrong answer would actually come from. That's where to spend your attention — and it's exactly the column most software never shows you.

Five ways in

/xray on its own explains the conversation you just had. Add a word and it opens a different door.

/xray boot is the quiet favourite. Most people are surprised how much has already happened by the time the cursor blinks.

Ask it to draw you the picture

Here's the part people don't think to try. Dex can explain itself in words — but it can also draw itself, and the drawing lands in your own notes where you can keep it.

Ask for a diagram and you get one in the standard format that Obsidian and most note apps render as an actual picture. So "how does this thing work?" becomes a diagram you own, made from your setup rather than a generic marketing graphic.

Draw me a diagram of what actually happens between me pressing enter and
you answering — in my setup specifically, not in general. Show what's
already loaded before I type, what you go and fetch, where judgment
enters, and where the answer gets written back. Use a diagram I can save
into my vault, then walk me through it in plain language underneath.

The same trick works on anything you're trying to hold in your head:

• "Draw how my meeting notes turn into tasks and person pages."
• "Draw the path a promise takes from something I said in a meeting to something I actually did."
• "Draw my week — where my time goes, by pillar."

Make it your own: ask for the diagram once, save it in 06-Resources/, then ask Dex to redraw it whenever you've changed your setup. A picture of your own system, kept current by the system itself.

Using it to check the work

The honest reason to look inside isn't curiosity. AI is mostly right and occasionally confidently wrong, and the difference is invisible from the outside — a wrong answer is written in exactly the same tone as a right one. What isn't invisible is the trail.

Take the answer you just gave me. Show your working: for every factual
claim, which file did it come from and can I open it? Then mark each
claim as one of — read it in my notes, worked it out, or assumed it.
Anything in that third pile, say so plainly and tell me how you'd
confirm it.

Three habits worth forming:

• X-ray answers that surprised you. Either you learn something about your own business, or you catch a wrong turn early.
• X-ray before you act on something that matters. Board numbers, a promise to a customer, anything with a person's name on it.
• X-ray when Dex feels off. Nine times out of ten the trail shows the problem instantly — a stale file, a connection that quietly stopped working, an old instruction still steering it.

Then teach it back

The last thing /xray is good for is checking that you understood — which is the difference between reading about how AI works and being able to use it.

Explain what you just did to me as if I'm smart but have never used AI —
no jargon, and stop and explain any word I'd have to look up. Then flip
it: ask me three questions to check I actually followed it, correct me
where I'm wrong, and finish with the one thing from this that's most
worth me knowing cold, given how I use you.

Once the machine stops being a black box, the next question changes. It goes from "can it do this?" to "what would I have to change for it to do this?" — and that question has a page of its own: make Dex your own.

First, fetch and read the page I'm referring to: https://heydex.ai/help/xray.md

I just read the Dex Guide's page on /xray. Do it live on our last
exchange: what loaded before I typed, which of my files you opened, what
you looked up versus what you judged, and which instruction of mine was
steering you. Keep it in plain words. Then draw me a diagram of that
whole path in my setup and save it into my vault. Finish by telling me
the one thing about how you work that would most change how I use you —
and what I should change to get it.