Asking for a dashboard

Ask for it in one sentence

A dashboard in NSGPT is a board — cards, charts and tables laid out on a grid — that lives in a project as an item of its own. You do not assemble one. You ask for it in chat, in one sentence, and say what it should be called.

Say where it goes as well as what it is. Naming the file in the request is what ties the sentence you typed to the row that appears afterwards.

What happens while you wait

One request starts one run. The composer swaps its send button for a stop button and keeps it for the whole run, so the composer is the status light: when the send button comes back, the run is over.

The board is written during that run. There is no background job to come back and check on later.

Where it lands

What arrives is one item in the project — a file whose name ends .dashboard.py, at a project:// address of its own.

It appears in the project's item listing. Filter that listing to Dashboard to see only boards, or open Dashboards in the sidebar. Opening the row opens the board.

That last part is the part worth checking, because it is the difference between a dashboard and a document about one: a real dashboard opens as a grid of cells. If what opens is a page of text, see If it does not work below.

What a dashboard is made of

You do not have to read one to use it, but it is not a black box either.

A dashboard is a Python file. Each block of code in it is a cell, and each cell is placed on a 12-column grid by its own declaration — @cell(x=0, y=0, w=12, h=4) — so the layout is written down in the file rather than dragged into place. Only what a cell outputs is drawn. Prose written between cells does not appear on the board.

Because it is an item like any other, it travels like one: a dashboard that is part of a template arrives in your project when you start from that template.

Where the numbers come from

A dashboard reads. It is not meant to go and fetch anything itself — the guidance the assistant follows when it writes one says exactly that — so a board is built over data that is already in the project as an item something else produced and saved.

That is also the shape to ask for if you want a board that keeps up. Put the fetching in a separate item — a page or an automation — that saves its results into the project. The board reads those results, and the cells that read them redraw when they are rewritten.

What ends up on the board is decided per request. NSGPT keeps no standard project dashboard to fill in, so the cells, their number and their titles are chosen each time, and two people asking the same question will not get the same board. Read it before you rely on it, the same way you would read the query behind a number — whether a tile measures what you think it measures is a question about that board and nothing else.

Whether it is up to date

A board is a saved result, not a live view.

The first time it is opened it computes, and you watch the tiles fill in. Every opening after that shows you what it computed then — the stored output, replayed. It does not re-run because you came back.

To recompute it, use Run all cells — the circular arrow above the board. A single cell can also be re-run on its own. If a board carries controls, changing one re-runs the cells that depend on it.

Nothing recomputes a dashboard on a schedule. If a board has to be current without someone opening it and pressing the button, schedule the automation that produces its data instead; when that writes a new result, the cells reading it redraw.

Who can change it

A dashboard is editable by anyone who can open it. Cells can be edited in place, re-run, moved and resized, and what you change is written back to the same item.

There is no view-only access. NSGPT has no role that lets someone read a project's dashboards without also being able to change or delete them, and access extends to everyone in the organisation the project belongs to. "Can see this board" and "can edit this board" are the same permission.

What a dashboard is not

It is not something you can send someone. There is no public link and no view for anyone outside the application. A board reaches a person by their having access to the project it is in, and no other way.

You cannot export the board. There is no download of it as an image, a PDF or a slide.

It is not a route around access. You reach the application through an NSGPT engagement and no other way — request a demo.

If it does not work

The run can finish without leaving you a dashboard, and the ways it does are worth recognising because the fix is different for each.

Nothing was written. The assistant describes the dashboard it would build instead of building one, and no new item appears. Ask again, saying explicitly that the file should be written rather than described.

A page landed instead of a board. An item appears and opens, but as a document rather than a grid of cells. This is the near miss to look for: the item is real, so nothing looks broken, and only the way it opens tells you. Ask for it again as a dashboard.

The row is there and the board is empty. A board can be created with nothing in it. There is nothing to recover; ask again.

Two boards landed. A retry inside the same run can leave two items. Both may be fine — delete the one you do not want.

A tile shows an error. A cell whose code fails prints what went wrong into the tile, and it prints it the same way a cell prints a successful result — no red border, no warning icon. A tile of unexpected technical text is a failed cell, not a finding.