LLM Tool
The LLM Tool integrates Large Language Models (LLMs) directly into a Dataworkz Agent, enabling it to generate content, analyse and summarise data, extract structured information, transform text, or provide automated responses — all driven by prompts you define at configuration time.
When to Use
Use the LLM Tool when an Agent needs to:
Generate or rewrite content based on dynamic inputs
Summarise long-form text into concise outputs
Extract specific information from unstructured text
Transform data from one format to another
Answer questions using reasoning rather than data retrieval
Provide automated responses with a defined tone or persona
Prerequisites
Before creating an LLM Tool, ensure at least one LLM has been configured on the Dataworkz platform via the LLM Configuration screen. The Tool cannot be saved without selecting a configured LLM.
Creating an LLM Tool
Go to AI Agents → Create Tool and select LLM from the Tool type selector.
The creation flow has four sections:
About Tool
Tool name
Yes
A descriptive name that conveys what the Tool generates or processes — for example, summariseDocument or generateTripPlan
Description
Yes
A clear explanation of what the Tool does and when the Agent should use it. The Agent uses this description to decide when to invoke the Tool during reasoning
Notes
No
Internal notes for documentation purposes — not used by the Agent during reasoning
💡 Note: A specific and well-defined description improves AI responses in the Agent. Include what type of content the Tool generates and the conditions under which it should be called.
Configuration
The Configuration section defines which LLM to use and how it should behave through prompts.
LLM Selection
Select the language model from the dropdown. Only LLMs that have been configured in the LLM Configurator will appear here.
Prompts
The LLM Tool requires at least one type of prompt to function. Two prompt types are available and can be used together:
System prompt
Sets the rules, context, tone, or personality for the LLM. Defines how the model should behave across all requests to this Tool
"You are a helpful assistant that explains technical terms in simple language."
User prompt
Defines the specific task or query the Tool handles. This is typically where dynamic inputs are injected via parameter placeholders
"Plan a 3-day trip to ${cityname}."
Prompts can include parameter placeholders using the format ${parameterName} — for example, ${text}, ${cityname}, or ${input_url}. Each placeholder automatically becomes an input parameter in the Input Parameters section.
Examples from the UI:
System prompt:
Generate a summary for ${text}User prompt:
Plan a 3-day trip to ${cityname}.
💡 Note: Combining a system prompt with a user prompt gives you precise control over the LLM's behaviour. Use the system prompt to define the model's role and constraints, and the user prompt to define the dynamic task it performs.
💡 Note: The LLM Tool requires at least one prompt — either a system prompt or a user prompt — to function correctly. Both can be used together for more controlled outputs.
Input Parameters
Input parameters are detected automatically from the placeholder variables defined in your prompts. Each ${parameterName} in the system or user prompt becomes an input parameter the Agent populates at runtime.
For each detected parameter, configure:
Name
Automatically populated from the placeholder in the prompt
Type
The data type — such as String, Integer, or Boolean
Description
A clear explanation of what the parameter represents, including expected formats or example values
💡 Note: If no parameters are detected, the message "No parameters detected. Use ${parameterName} in the prompt to add parameters" is displayed. Add placeholders to your system or user prompt to define the inputs the Agent needs to supply at runtime.
Output
The Output section defines the structure of the content the LLM generates in response to the prompts.
Output type name
The type of output returned — typically Object for a structured response
Description
A plain language description of what the output represents — for example, "the generated summary" or "the trip itinerary"
Object structure
A JSON definition describing the structure of the generated output
Example output structure:
Mark output as artifact — Enable this toggle when the LLM generates a file or document output — such as a formatted report, document, or audio transcript — rather than plain structured text.
💡 Note: If you are unsure of the exact output structure, run the Tool using the Test panel first and apply the returned structure automatically. This ensures the output definition accurately reflects what the LLM generates.
Testing the Tool
Use the Test panel on the right side of the page to validate the LLM Tool before adding it to an Agent Scenario:
Enter sample values for each input parameter
Click Test → to run the Tool
Review the generated output to confirm the system and user prompts work together to produce the expected result
Adjust the prompts and retest as needed until the output is consistent and accurate
💡 Note: Always test the LLM Tool with representative input values before connecting it to an Agent. Prompt quality directly determines output quality — iterate on your prompts until the Tool produces reliable, consistent results.
Saving the Tool
Once all sections are complete and the Tool has been tested, click Save & Close to save and return to the My Tools list, or Save to save without closing.
💡 Note: Once saved, the LLM Tool is immediately available to be referenced in any Agent Scenario within your Dataworkz environment.
⚠️ Warning: Avoid using vague or ambiguous prompts. Unclear instructions lead to inconsistent outputs, which can cause the Agent to produce incorrect or unhelpful responses. Always be explicit about the format, scope, and constraints of the expected output in your prompts.
Last updated