Dashboards

A dashboard is a screen of live tiles on a grid: key figures, charts, tables, matrices, texts and images. It is an object of a folder, with its versions, like a flow or a report. A reader opens it to see the state of something at a glance, chooses its parameters at the top, clicks a bar to filter the other tiles, or leaves it on a wall screen refreshing by itself. It downloads as a PDF or an image; a flow group sends it by email; a script takes its file through the API.

Use a report when the answer has to arrive as a document with a fixed page, the same layout every month. Use a dashboard when the screen is the answer: it adapts to the window and answers clicks. The two share what they can: a dashboard's datasets and parameters are written as a report's, read with the same rights, in the same expression language. Every installation has dashboards, Docker Compose included.

Sales overview, the dashboard of the sample data: open it in the Sample folder, and everything on this page can be tried on it.

At a glance

Reading a dashboard

Analysis → Dashboards lists the dashboards you may open, the most read first: the people who opened one in the last 30 days, counted once a day. The search looks in the names, the descriptions and the folders. A click on a row opens the dashboard; so does a click in Explore, or a link someone sent you, after signing in.

Parameters

The bar at the top holds the dashboard's parameters. Choose values and press Apply: the tiles read the data again with them, and the address holds them. A list can follow another: in Sales overview, choose an Area and the Agent list holds only the agents of that area, before you press Apply.

With Nord-Ovest chosen the Agent list holds 13 names of the 90. An agent chosen before goes back to Any when the area changes.

A click filters the other tiles

A click on a bar, a slice, a point, a cell of a heatmap, a group's name in a table or a header of a matrix filters the other tiles on the same dataset. A strip under the parameters says what is active, with a cross on each filter and Clear all.

A click on Centro, then Clear all. The tile clicked keeps its bars and marks the one chosen.

• Ctrl (or Cmd) with the click adds a value, or takes it away. A plain click on the only value chosen clears it.
• In one tile, values add up: Centro or Nord-Est. Between tiles, and between the rows and the columns of a matrix, each narrows the others.
• A click stays on its dataset. In Sales overview a click on an area leaves the margin at the bottom as it is, because the margin reads another dataset, with its own Category parameter and its own click on the donut. To filter several datasets with one choice, use a parameter that each of them reads.
• A KPI, a text and an image do not filter. Neither do «Other», a waterfall's total, a gauge or a detail row of a table. The designer can turn a tile's filtering off (A click filters the other tiles).
• An aggregate that names its dataset, SUM([ricavo], "agents"), still reads all of it, so that a share of the total stays one; so does a calculated field.
• A parameter changed afterwards keeps the clicks whose values still exist, and drops the others.
• On a table that can filter, a group's name filters and the arrow beside it closes the group.

Clicks are in the address too (?area=Nord-Est&click.by_area=Nord-Est), so Back undoes one, and a link gives the same state to someone else with the rights. A link naming a tile that is gone, or one that no longer filters, opens without that click.

From the keyboard, a group's name, a matrix's header, the cross on a filter and Clear all are buttons (Ctrl with Enter adds). A chart's drawing cannot be reached: when it takes the focus a list of its categories shows, and choosing there adds a value. The parts of a split chart, the cells of a heatmap and the points of a scatter cannot be chosen from the keyboard, only their category where they have one.

Refresh and the cache

Beside the title, Data from says when the oldest numbers on the screen were read from their source.

• Everyone shares the numbers. The first person to open a dashboard reads the source; whoever opens it in the next five minutes (DASHBOARDS__CACHE_SECONDS, or the dashboard's own interval when that is longer) gets the same numbers without a new query. On a database that charges by the query, this is what keeps the bill down. A click works from the rows already read and does not go back to the source. Your rights are checked at every request all the same: sharing never shows you what you may not read.
• Refresh reads the source again, for everyone, at most once every 10 seconds per dashboard. Pressed sooner, it shows what there is, without an error.
• By itself. A dashboard can refresh every so many seconds, from 60 to 86,400: Refresh → By itself in the designer, and the viewer says so («refreshes every 5 minutes»). It runs only while the tab is in sight. When a refresh fails it keeps the last numbers, with the time they are from, and says so; it tries again at the next interval.
• A datasource refreshed or a flow run is new data: its new snapshot is read at the next request, with no wait for the five minutes.

Full screen leaves only the title, the strip of clicks, the tiles and the time of the last update, for a wall screen. Esc comes back.

Any window

The grid adapts to the width of the window. From 1,100 pixels it is as designed, on 12 columns; from 700 to 1,099 it has 6, and the tiles take what is free; below 700 pixels one column, top to bottom and left to right. A tile hidden by its condition takes no place.

PDF and image

PDF and Image download what is on the screen: the same parameters, the same clicks, the same sort and closed groups, and the same numbers, because they come from the same reading. It is a document of the numbers, not a picture of the screen: it has a report's look, on white, whatever the theme you use.

On top, what it is showing: the name, the parameters that have a value, the clicks, when the rows are from. At the foot, the page's number.

• A row of tiles stays together; a table takes its row alone and the whole width, goes on over the pages, and repeats its title and head on each. The columns that do not fit go on in another table with the first column again.
• The page is A4 across with margins of 10 mm, unless the designer sets another in Page of the PDF and the image.
• The image is the same document on one long page, PNG at 192 ppi. It is one page in memory: past REPORTS__RENDER_MEMORY_MB (at the default, a table of 1,000 rows fits and one of 2,000 does not) it stops, says so, and points you to the PDF, which holds more.
• A tile that cannot say its numbers says why in its place. Of a dataset you may not read it says only that you cannot.

Example: Sales overview

The sample data ends with this dashboard in the Sample folder, made by sampledb-init (the sample data). It reads two tables the sample flows write, in Sample / Private tables:

It has three parameters: Area and Agent filter agents, Category filters months. With the sample data at the default scale:

Build one yourself

A smaller version, with the mouse, in a folder where you have EDIT, here My dashboards; the reader needs VIEW on it and on Sample / Private tables.

1. The dashboard. In Explore, in your folder, New → Dashboard. In the designer, write My sales over «New dashboard». The empty page says to start with the data.
2. The data. Add a dataset (or Data → + Dataset): name agents, Datasource, choose «Agent Leaderboard», Add. Its fields show under the dataset.
3. The first parameter. + Parameter: name area, label Area, type Text, Choices → From a dataset, value [area]. Filter this dataset by what is chosen is already ticked. Add: its list shows above the grid.
4. The second. + Parameter again: name agent, label Agent, From a dataset, value [agente], Add. Because it comes after area and filters the same dataset, the agents follow the area. Nothing more is needed.
5. A key figure. Insert → KPI. It opens with a number on it, from the fields of the dataset. In the panel on the right, Value SUM([ricavo]), label Revenue, format € #,##0.
6. The chart that filters. Insert → Chart, kind Column: along the axis [area], the series' value SUM([ricavo]), title Revenue by area. A click filters the other tiles is already ticked.
7. A table. Insert → Table; it opens with columns from the fields. Give it the title Agents. Leave a column or two, or click fields in Data to add them.
8. The places. Drag the tiles where you want them, and their handles to size them.
9. Done saves and opens it as readers see it. Revenue says € 18,961,683. Click the bar Centro: € 10,042,680.

The designer

The designer opens from Explore, from the tab, and from Edit in the viewer; with VIEW only you are taken to the viewer. It needs a window at least 900 pixels wide. It has the three parts of the report designer: on the left Insert and Data, the grid in the middle with the real numbers on it, on the right the properties of what is selected.

The tile selected shows its handles; the panel holds its place, its kind and its data. With nothing selected, it holds the dashboard: language, time zone, refresh, page of the documents.

Sizes are in cells: 12 columns that share the width, rows of 40 pixels, 10 pixels between tiles. Drag a tile to move it, a handle to size it; it snaps to the grid, and a tile it lands on goes down rather than being covered. Click a tile in Insert to add it where it first fits, or drag it; click a field in Data to add it to the tile selected or make a new one. A new tile opens with data on it, from the fields of its dataset. One half made says what it needs; one with a mistake says so inside the tile and at its field, and the others go on.

The keys: Ctrl+S saves, Ctrl+Z and Ctrl+Shift+Z undo and redo (a whole drag is one step, with what it pushed), the arrows move the tile selected by one cell (5 with Shift), Ctrl+D duplicates it, Del removes it. A click on the grid selects a tile and filters nothing. Shown when, a condition, hides a tile from readers: on the grid it stays and says so.

Datasets, parameters and expressions are a report's, with the same dialogs: see Datasets and parameters, A parameter that filters the data, Expressions and Report functions, with SUM([x], "group"), PREVIOUS, LOOKUP, SUMIF and COUNTIF. PAGE() and PAGES(), which speak of a page, write nothing on a screen. Every save is a new version; the server checks the whole definition each time, and a save it refuses names the tile and the field. If someone else saved since you opened it, Save says so and asks before yours becomes the newest.

AI comments

An AI comment tile holds a short text written by the configured AI model on the numbers of the tiles you pick, such as what moved and what stands out. The numbers never depend on it, and never wait for it: a provider that is down, a budget that is spent or a mode that forbids it costs the comment only, and the tile says so.

To add one, click AI comment in Insert, and in its panel write what you ask of it (at most 1,000 characters; empty, it says what stands out) and tick what it reads, from 1 to 10 KPIs, charts or tables. Try it shows what it writes on the tile: the designer asks the model only then, never at a keystroke, and it needs EDIT on the folder.

Every comment says Written by AI · the model · numbers of the moment they were read, and that it may be wrong, on screen and on paper; no setting takes that away. It is plain text, at most 1,500 characters, with no link.

When it cannot write, it says why in its place, and a PDF, an image or an email goes out with one line there: that the comment is not available, that the day's spending limit has been reached, that this installation sends no data to AI models (AI__METADATA_ONLY).

Any element it reads that you may not read stops the comment, with what a tile with no access says and nothing about which one, on screen and in the PDF, with a comment kept for others and with none: a text written from fewer numbers would mislead, one from more would leak. An element hidden by its own condition shows nothing, so it is left out.

What its model reads goes where the administrator said: AI in Tabularia.

Sending it on a schedule

A dashboard is sent by a flow group: a Pushmail cell attaches it as PDF or PNG, up to 5 of them, with the values of its parameters. The group's schedule is the dashboard's: the cell runs after the steps that refresh the data. Nobody looks at it before it leaves, so it is stricter than the screen: a dataset the group's author may not read, or a tile that cannot say its numbers, stops the cell with its name, and nothing is sent. It reads its source and not what the screen keeps: what the steps before it refreshed is what it says. An AI comment never stops it: with the provider down the email goes out, with one line in its place.

With whose rights

A dashboard never lends its author's rights. What a tile reads is checked dataset by dataset, at every request, with the rights of whoever it is read for, before anything shared is looked at:

A tile whose dataset you may not read says so inside the tile, and the others work: «You do not have access to the data of this tile», with no number, no column and no name of the dataset, whether or not someone with the rights has just read it. A click on such a tile counts for nothing. Row-level security does not exist in Tabularia, here as in reports: rights are on the dataset.

Whoever reads Sales overview needs VIEW on Sample, which reaches Sample / Private tables where its data is. The Sample folder is created with no permissions: the administrator shares it.

Administrators see every dashboard. Dashboards show in Explore, with their own icon, in the search and in the tab. A view, a SQL script, a datasource or a connection that a dashboard reads cannot be deleted while it reads it: the refusal (409) names it. A dashboard a flow group sends cannot be deleted until it is taken out of the group's cells.

The audit log has every change (dashboard.create, .update, .delete, .move, .rename, .duplicate, .restore), every opening (dashboard.open) and every document (dashboard.render, with its format, its parameters, its clicks and where it came from: interface, api or flow_group). A click and an automatic refresh are not in it: they would drown the log.

Limits and good to know

• Aggregate in the dataset where you can. The rows of a request stay in the memory of the preview worker: see DASHBOARDS__MAX_ROWS.
• Without Valkey or the bucket a dashboard works, without the cache.
• Administrators change what the cache keeps, live, in Administration → Settings (the settings); DASHBOARDS__MAX_ROWS stays in the environment, as REPORTS__MAX_ROWS does.
• The screen draws its charts in your browser. The PDF and the image draw them with the chart service, as a report does: when it does not answer, a document with charts stops and says so.

What is not in this version

• No custom tiles and no code of your own in a tile.
• No drill-down from a tile to another dashboard.
• A click filters its own dataset only: across datasets, use a parameter.
• No comments by people and no alerts on a number.
• No dashboard public without signing in: a link needs an account with the rights.
• No CSV or Excel of a tile. A dashboard goes out as a PDF or an image.
• The assistant and MCP do not create or change dashboards, and have no tools on them; the search finds a dashboard by name and gives its link.

Related

• Reports and Report functions: datasets, parameters, expressions
• Pushmail: sending a dashboard from a flow group
• API: dashboards
• Permissions
• Configuration: dashboards, Settings from the panel