Skip to main content

Generating AI Dashboards with Prompts

Describe the dashboard you want and TagoAI builds the layout, then ask for changes the same way. This page covers where to prompt, how to write a good prompt, generator rules, restoring an earlier version, and what to do when a generation fails.

Admin view of the Cold Rooms AI Dashboard with two temperature cards, a temperature chart and a Door Alarms list. A follow-up prompt asks to drop the alarms list and make the chart full width, and the layout updates to match.

Where to prompt​

You can generate and change the dashboard from three places:

  • Prompt bar, at the bottom of the dashboard. Best when you know what you want.
  • AI Chat, in the sidebar that the AI button in the dashboard header opens. Best when you only have a rough idea.
  • API, from your own code or an Analysis. Best for building dashboards automatically.

Prompt bar​

The prompt bar floating at the bottom of an AI Dashboard, reading Describe dashboard changes..., with the model button set to Default and a send button.

Type in the prompt bar to generate the first layout, then to change layout, charts, labels, colors, or which data appears. While a generation runs, you can cancel it at any moment. A prompt can be up to 4,000 characters in English.

Pick the model and reasoning effort from the model button in the prompt bar. The choice applies to every AI Dashboard in the profile, and Run users' prompts in TagoRUN run on it too.

Conversation history

The prompt bar remembers the last 10 messages while the dashboard stays open, so a follow-up such as "make it wider" knows what "it" is. The history lives in the page: reloading, closing the tab, or signing out drops it, and the server keeps no copy, unlike the TagoAI Chat history. The saved layout and its versions stay.

AI Chat​

The AI button in the dashboard header opens the TagoAI Chat in a sidebar. The chat looks up your devices and variables first, then writes the full request for the generator. For how a chat change counts toward your AI usage, see Quotas and billing.

API​

Generate AI Dashboard runs the same generation as the prompt bar, from a profile token or an Analysis. Send the prompt, and optionally history with earlier turns so the generator can follow up on them. The API keeps no conversation of its own, so send the turns you want it to see on every call.

Writing a good prompt​

Name the devices or variables, the chart types, the time range, and any layout constraints:

Fleet overview for devices tagged type=sensor: last temperature and battery
as cards, a 24h line chart of temperature, and a table of devices with
battery below 20%. Dark theme, compact header.

Use variable and tag names exactly as they appear on your devices, so the generator picks the right data. If you keep repeating the same conventions in every prompt, move them into the rules and write them once.

Refining​

After the first layout exists, keep prompting in the prompt bar or the AI Chat:

  • "Make the temperature chart full width."
  • "Add humidity next to temperature on the same line chart."
  • "Use a lighter background and larger titles."
  • "Drop the map; keep the table and KPIs only."

Each refine that changes the layout replaces the previous one, and the data reloads for the new design. Refine in small steps rather than in one long prompt, and check that the numbers still line up with your devices' current data after a big structural change. Generated output can be wrong for your data. Correct it with another prompt, or restore an earlier version.

Generator rules​

Generator rules let you set instructions that every prompt on this dashboard follows, so you don't have to repeat them each time. They apply to every generation, whether you, a Run user, or an Analysis starts it. Write them in Dashboard settings, up to 10,000 characters.

Use the rules for conventions the generator can't guess from your data:

The device name is stored in the device tag key device_name; use it instead of the real name.
Sites are tagged site=<code>. Group charts by site.
Temperatures are in Celsius. Show one decimal place.
Run users can see the rules

A Run user with AI dashboard generation can get the rules repeated back by asking. Write naming and tagging conventions there, never secrets, tokens, or other customers' names.

Rules shape the layout and do not widen access. Whatever the generator writes, a user's data comes only from the devices that user can access.

Restore an earlier version​

Each generate or refine that changes the layout saves a version. One that changes nothing saves no version but still counts as a generation. To go back to a version:

  1. Open the dashboard menu and pick Restore version.
  2. Pick a version from the list.
  3. Click Restore.

Restoring replaces the whole layout with the saved one. When a Run user is picked in the User picker, the list and the restore apply to that user's personal dashboard.

Troubleshooting​

Generation depends on TagoAI for the profile. When something goes wrong, the prompt bar or the widget shows one of these messages:

MessageWhat to do
Out of AI creditsA paid model ran with no credits left. Open AI Wallet to add credits, or pick the free model with the model button. See AI Credits.
Monthly spend cap reachedThe profile hit its monthly AI spend cap. Adjust cap to raise or remove it, or wait for the next month.
Free model limit reachedGenerations on the free model count against the monthly free model allowance, even on a profile with its own provider. Wait for it to renew, or pick a paid model. See TagoAI Service.
Dashboard model unavailableThe model set for AI Dashboards is not available. Fix in AI models to pick another.
AI dashboards are disabled for this applicationAI is turned off for the profile, so AI Dashboards stop loading data in Admin and TagoRUN. Turn AI back on for the profile. See AI availability.
Still generatingAnother generation is running on this dashboard. Try again when it finishes.
Generation didn't finishThe AI service did not respond or could not build a dashboard from the prompt. Your prompt is kept: Retry it as it is, or edit it first.
This took longer than expectedThe generation may have finished on the server after Admin stopped waiting. Check the dashboard again in a moment before you retry.
This dashboard changed elsewhereSomeone else updated the dashboard while you had it open. Reload dashboard to see the current version.
No data in this rangeWiden the period and check that the device is sending. If it stays empty, refine the prompt with the exact variable, tag, or time window.
Couldn't load this widgetThe widget tries again on the next reload.
Couldn't load the dataClick Refresh data in the prompt bar.

A failed generation leaves the previous layout in place.