The Virtual Assistant is a chat panel built into the platform that answers questions about how to use it. Ask how to set up a metadata profile or why an asset is not appearing in a portal, and it replies with an answer drawn from this documentation and a link to the article it came from. This article covers where to find the assistant, what it can answer, the one change it can make on your behalf, and the boundary between it and the platform's other AI capabilities.
The assistant is a small button in the bottom-right corner of every page once you are signed in. It is also available while you are looking at an asset in the lightbox, so you can ask about what is in front of you without closing it.
Opening it gives you a panel with a single prompt: What can I help you with? Answers stream in as they are written rather than arriving all at once. Two controls sit in the header — minimising the panel leaves the conversation where it was, and restarting the chat begins a fresh one, which is the thing to reach for when a conversation has wandered away from what you actually wanted.
The assistant searches this support documentation by meaning rather than by keyword. That matters more than it sounds: a question phrased the way you would put it to a colleague finds the right article even when it shares none of that article's wording. Asking "how do I stop people downloading the files I share" will find the portal permissions article without you knowing the word "portal" was the one to use.
It then answers from the articles it found, and tells you which ones. An answer you can check is worth considerably more than a confident one you cannot, and the link is there so you can read the full procedure when the summary is not quite enough.
The flip side is worth stating plainly. Because answers are grounded in the documentation rather than generated from general knowledge, the assistant is at its best on questions this knowledge base covers, and will tell you it does not know rather than improvise when it does not. That is the intended behaviour, not a gap — an invented answer about your permission model would be worse than no answer.
Some questions are not documentation problems. When something appears to be broken, or the answer depends on how your workspace has been configured, the assistant offers to raise it with Data Dwell support.
It drafts the ticket from the conversation you have already had — what you were trying to do, what happened instead, what has been tried — shows you the wording, and files it once you confirm. The point of doing it this way is that the context is already captured. Most of the delay in a support exchange is the round trip spent establishing what was actually going on, and that has usually been said already by the time anyone thinks to open a ticket.
The assistant can rename assets you have selected. You select them in the library, describe the renaming you want, and it shows you what it intends to do before doing it. The change runs as you, so your permissions apply exactly as they would if you had edited each asset by hand.
That is deliberately the extent of it, and it never deletes anything. The assistant is not a way to operate the library, and treating it as one would be the wrong tool: bulk changes belong in the library's own editing tools, where you can see what you are affecting. If you want an AI client that genuinely works on your content — searching, organising, building portals — that is what an MCP connection is for.
Alongside the chat panel, and available where your permissions allow it, is a separate page for designing a library rather than using one. Describe your organisation, the content you hold and how teams work with it, and it proposes a folder structure, metadata profiles and the metafields to go in them.
You review the proposal, adjust the parts that do not fit, and only then create it. This is most useful at two moments: setting up a workspace for the first time, and restructuring one that grew without a plan. Neither is a decision to hand over entirely — the proposal is a well-informed starting point that saves you the blank page, and the planning guides are still the reasoning you want to apply to it.
This is the most common misunderstanding, and worth being direct about. The Virtual Assistant answers questions about the platform. It does not search your assets, and asking it to find last year's campaign photography will not work.
Three separate capabilities are easy to confuse:
Because the assistant matches on meaning, the way you ask changes what you get back, in the opposite direction to a search box:
If the assistant keeps missing on something your team asks regularly, that is useful information in itself, and worth sending to support. A question asked often enough to notice is usually one the documentation should answer directly.
The assistant is available to signed-in users without any setup. The structure design page is permission-controlled, so if you cannot reach it and need to, contact your administrator or Data Dwell support.