Prompts

Prompts are reusable templates that combine your expert instructions with live telemetry pulled from ThingsBoard. When triggered, Trendz automatically injects the relevant device or asset metrics into the template, submits it to the configured LLM, and displays the response alongside your data — no manual data gathering or prompt rewriting required.

Once saved, a prompt can be applied in two places:

Where	How
Trendz widget	Select a prompt per view in View Settings → AI Assistant. See AI Widget Summary — Set the Prompt.
ThingsBoard widget action	Pass the prompt ID to trendz.getAnalytics() in a custom widget action. See AI Widget Summary — Customize with getAnalytics().

Click the Assistant icon in the left sidebar, then open the Prompts tab in the upper-left area of the workspace.

The Prompts table shows all available prompts with columns: Name, Updated, Created, Type, and Actions.

The Type column has two values:

Type	Description
System	Built-in prompts provided by Trendz. Cannot be renamed, edited, or deleted.
Custom	Prompts you create. Can be renamed, edited, and deleted.

1. Click New Prompt. The prompt editor opens.

2. Write the prompt content in the editor.

3. Enter a name for the prompt.

4. Click Save.

Click any prompt row to open it in the editor and read its content. To copy the prompt ID — used when referencing a prompt from a widget custom action — click Copy prompt ID in the upper-right corner.

1. Click the prompt row to open it.

2. Click Edit.

3. Update the content.

4. Click Save.

Two ways to rename a prompt:

From the Prompts table: Click the three-dot menu in the Actions column, select Rename, type the new name, and press Enter.

From the editor: Click the pencil icon next to the prompt name, type the new name, click any free space to confirm, then click Save to persist the change.

Click the three-dot menu in the Actions column, select Delete, and confirm by clicking Yes.

The Run test feature lets you preview how a prompt will analyze real data before applying it to production dashboards.

1. Open the prompt in the editor.

2. Click Run test in the upper-right corner.

3. In the test panel that opens on the right, choose one of your saved views from the Select View dropdown.

4. Click Test prompt to generate the summary.

5. Review the result.

The editor uses Markdown — a format that LLMs understand well and that keeps prompts readable for humans too.

Click Show Preview to split the screen: the left side shows the Markdown source and the right side shows the rendered output. Click Hide Preview to return to the single-pane view.

The toolbar in the upper-left of the editor provides one-click shortcuts for common Markdown blocks:

Shortcut	Block
B	Bold text
I	Italic text
H1	Heading
"	Blockquote
🔗	Link
🖼	Image
•	Unordered list
1.	Ordered list
<>	Code block

Click the pencil icon next to the prompt name, type the new name, then click any free space to confirm. The name is not saved until you click Save.

After editing, click Save to apply all changes or Cancel to discard them. Both buttons are enabled only after a change is made.

Trendz includes six built-in system prompts available on every installation:

Prompt	Purpose
Trendz System Default Summary Prompt	Concise 2–3 sentence executive summary covering performance, sentiment, and emerging trends
Trendz System Operational Efficiency Optimization Prompt	Detects throughput or resource-use improvement opportunities; focuses on utilization gaps and bottlenecks
Trendz System Data Quality Review Prompt	Highlights data inconsistencies, outliers, and anomalies; evaluates impact on analysis
Trendz System Positive Performance Prompt	Identifies areas of strong performance and growth potential; summarizes key positive trends
Trendz System Incident Identification And Prioritization Prompt	Detects anomalies, spikes, or rare behaviors in time-based or event-based datasets
Trendz System Root Cause Analysis Prompt	Explains the likely root cause of a performance issue with a confidence rating

Practice	Guidance
Be specific	Describe the exact question or analysis you want. Vague prompts produce vague responses.
Use Markdown structure	Use headers, bold, and lists to organize the prompt. LLMs interpret structured prompts more accurately.
Keep it concise	Long prompts can dilute focus. State the task clearly and avoid padding.
Test before deploying	Use the Run test feature with a real saved view to verify the output before attaching the prompt to a dashboard.
Use saved prompt IDs	When referencing prompts from widget actions, use the prompt ID instead of inline text so you can update the prompt without touching widget code. See AI Widget Summary.
Name prompts clearly	Use names that describe the analysis goal, e.g. Boiler Root Cause Analysis instead of Prompt 3.