axon-ivy-smart-workflow

name	axon-ivy-smart-workflow
description	Create AI-powered data extraction and processing in Axon Ivy workflows using the Smart Workflow AgenticProcessCall element.

Axon Ivy Smart Workflow - AI Extraction Skill

Use this skill when creating AI-powered data extraction or processing in Axon Ivy workflows.

Prerequisites Check

Step 1: Check pom.xml for Smart Workflow Dependency

Before creating Smart Workflow elements, verify the project's pom.xml contains the required dependencies:

<dependency>
  <groupId>com.axonivy.utils.ai</groupId>
  <artifactId>smart-workflow</artifactId>
  <version>13.2.0-SNAPSHOT</version>
  <type>iar</type>
</dependency>
<dependency>
  <groupId>com.axonivy.utils.ai</groupId>
  <artifactId>smart-workflow-openai</artifactId>
  <version>13.2.0-SNAPSHOT</version>
  <type>iar</type>
</dependency>

Step 2: If Dependencies Missing

1. Search for smart-workflow project in workspace:

   • Look for smart-workflow/pom.xml or smart-workflow-openai/pom.xml in the workspace
   • Use glob pattern: **/smart-workflow/pom.xml

2. If smart-workflow project EXISTS in workspace:

   • Add the dependencies to the target project's pom.xml
   • Use the version from the smart-workflow project's pom.xml

3. If smart-workflow project DOES NOT EXIST in workspace (no pom.xml found AND no smart-workflow-*.iar file found):

   • Use the ivy-market-import skill to import the smart-workflow library into the codebase before doing anything else.
   • After import completes, continue with the rest of this skill.

Creating Smart Workflow Elements

ProgramInterface Element for AI Extraction

Use ProgramInterface with AgenticProcessCall to call AI for structured data extraction.

{
  "id" : "f1",
  "type" : "ProgramInterface",
  "name" : "Extract Data with AI",
  "config" : {
    "javaClass" : "com.axonivy.utils.smart.workflow.AgenticProcessCall",
    "userConfig" : {
      "system" : "You are an AI assistant. Extract the requested information from the provided text.",
      "tools" : "[]",
      "resultType" : "package.model.OutputClass.class",
      "resultMapping" : "in.outputVariable",
      "query" : "Extract information from this text:\n\n<TEXT>\n<%= in.inputText %>\n</TEXT>"
    }
  },
  "visual" : {
    "at" : { "x" :     
          
  
     
        
        
        
          
          
            
              
              
          
        
      
        
                 
      
        
               
      
     
    
           
  

userConfig Parameters

Parameter	Required	Description
system	Yes	System prompt instructing the AI what to extract
query	Yes	User query with input data (use <%= in.variable %> for template expansion)
resultType	Yes	MUST end with .class - The Java class for structured output
resultMapping	Yes	Variable to store the result (e.g., in.result)
tools	No	List of callable sub-processes as tools (default: "[]")
model	No	AI model to use (e.g., "gpt-4.1")

IMPORTANT NOTES

1. resultType MUST End with .class

CORRECT:

"resultType" : "hr.onboarding.model.EmployeeInfo.class"

WRONG:

"resultType" : "hr.onboarding.model.EmployeeInfo"

2. Output Type MUST Be a Single Object (NOT a List)

The AI extraction cannot return a List directly. Always use a wrapper object.

WRONG - Will not work:

"resultType" : "java.util.List.class"
"resultType" : "List<hr.model.Item>.class"

CORRECT - Use a wrapper class:

// Create a wrapper class
public class ExtractionResult {
  private List<Item> items;
  // getters and setters
}

"resultType" : "hr.model.ExtractionResult.class"

3. Use @Description Annotations for Better Extraction

When creating model classes for AI extraction, use LangChain4j @Description annotations:

package hr.onboarding.model;

import dev.langchain4j.model.output.structured.Description;

@Description("Employee identification information")
public class EmployeeInfo {

  @Description("System username for the employee")
  private String employeeUsername;

  @Description("Employee's first/given name")
  private String employeeFirstName;

  @Description("Employee's last/family name")
  private String employeeLastName;

  // getters and setters
}

Complete Example: Employee Data Extraction Subprocess

Data Class: ExtractEmployeeData.d.json

{
  "$schema" : "https://json-schema.axonivy.com/14.0-dev/project/data-class.json",
  "simpleName" : "ExtractEmployeeData",
  "namespace" : "hr.onboarding.agent",
  "fields" : [ {
    "name" : "inputText",
    "type" : "String",
    "comment" : "Raw input text"
  }, {
    "name" : "employeeInfo",
    "type" : "hr.onboarding.model.EmployeeInfo",
    "comment" : "Extracted employee information"
  }, {
    "name" : "error",
    "type" : "ch.ivyteam.ivy.bpm.error.BpmError"
  }, {
      
      
   

Process: ExtractEmployeeData.p.json

{
  "$schema" : "https://json-schema.axonivy.com/14.0-dev/project/process.json",
  "id" : "19CF01A0E1B2C3D4",
  "kind" : "CALLABLE_SUB",
  "config" : {
    "data" : "hr.onboarding.agent.ExtractEmployeeData"
  },
  "elements" : [ {
      "id" : "f0",
      "type" : "CallSubStart",
      "name" : "extractEmployeeData(String)",
      "config" : {
        "signature" : "extractEmployeeData",
        "input" : {
          "params" : [
            { "name" : "inputText",       
          
            
              
              
          
        
          
            
                      
                      
          
            
              
              
          
        
      
                   
                 
     
        
        
        
        
          
          
            
            
            
            
            
        
      
                          
         
            
            
            
              
                         
          
                       
                                 
         
                 
     
        
        
                   
     
        
        
        
        
          
            
        
      
                   
                             
     

Using Tools (Callable Sub-Processes)

To give the AI access to tools (sub-processes it can call):

"userConfig" : {
  "system" : "You are an assistant with access to tools...",
  "tools" : "[\"searchDatabase\", \"createRecord\", \"sendNotification\"]",
  "resultType" : "hr.model.AgentResponse.class",
  "resultMapping" : "in.response",
  "query" : "<%= in.userRequest %>"
}

Each tool name must correspond to a callable sub-process signature in the project.

System Prompt Best Practices

1. Be specific about what to extract
2. List all fields the AI should populate
3. Provide rules for handling edge cases
4. Specify formats for dates, numbers, etc.
5. Use examples if the extraction is complex

"system" : "You are a data extraction assistant.\n\nEXTRACT:\n1. fieldName: Description of what to extract\n2. anotherField: Another description\n\nRULES:\n- Rule 1\n- Rule 2\n\nFORMATS:\n- Dates: yyyy-MM-dd\n- Phone: +1-XXX-XXX-XXXX"

Query Template Patterns

Simple Text Input

"query" : "Process this text:\n\n<%= in.inputText %>"

JSON Object Input

"query" : "Process this data:\n\n<%= dev.langchain4j.internal.Json.toJson(in.dataObject) %>"

Multiple Inputs

"query" : "Context: <%= in.context %>\n\nData to process:\n<%= in.inputData %>\n\nInstructions: <%= in.instructions %>"

Binary File Input (Image or PDF)

Pass binary files directly in the query — AgenticProcessCall understands InputStream and Binary objects natively. No base64 encoding needed.

"query" : "Extract all data from this invoice:\n<%= in.uploadedFile %>"

Supported data types for file variables:

Type	How to obtain	Notes
java.io.InputStream	From uploaded file or CMS	Preferred for images
ch.ivyteam.ivy.scripting.objects.Binary	From CMS binary resource	Preferred for PDFs
java.io.File	From file system path	Works for both
CMS binary directly	ivy.cms.co("/path/to/resource")	Embedded in template

Example: Extract invoice from user-uploaded file

Data class:

{
  "$schema" : "https://json-schema.axonivy.com/14.0-dev/project/data-class.json",
  "simpleName" : "ExtractInvoiceData",
  "namespace" : "invoice.extraction",
  "fields" : [ {
    "name" : "uploadedFile",
    "type" : "java.io.InputStream",
    "comment" : "Uploaded invoice file (PDF or image)"
  }, {
    "name" : "invoice",
    "type" : "invoice.model.Invoice",
    "comment" : "Extracted invoice data"
  }, {
    "name" : "error",
    "type" : "ch.ivyteam.ivy.bpm.error.BpmError"
  }, {
      
      
   

AgenticProcessCall query:

"query" : "Extract all invoice data from this document:\n<%= in.uploadedFile %>"

Type guidance for financial/document extraction:

Field type	Use	Reason
java.lang.Double	Prices, amounts, totals	Sufficient for display and calculation
java.lang.Integer	Quantities	Whole numbers only
String	Dates, invoice number, IBAN	Return as string, parse in Script if needed
List<T> (via wrapper)	Line items	Must be wrapped — see Note 2 above

System prompt example for invoice extraction:

"system" : "You are an invoice data extraction assistant.\n\nEXTRACT THE FOLLOWING:\n1. invoiceNumber: The invoice identifier\n2. invoiceDate: Issue date (format: yyyy-MM-dd)\n3. dueDate: Payment due date (format: yyyy-MM-dd)\n4. vendorName: Name of the issuing company\n5. customerName: Name of the billed customer\n6. items: List of line items (description, quantity, unitPrice, total)\n7. subtotal: Amount before tax\n8. vatRate: VAT percentage (e.g. 8.1)\n9. vatAmount: Tax amount\n10. totalDue: Final amount due\n\nRULES:\n- All amounts are numbers (no currency symbols)\n- Return null for fields not found in the document"