🤖 Tutorial: How to use Gemini AI to Debug SAP CPI Errors Automatically

[#1]

🤖 Tutorial: How to use Gemini AI to Debug SAP CPI Errors Automatically 🤖

Body: Why settle for cryptic stack traces? I recently built a custom Exception Subprocess in SAP Integration Suite that catches failures and asks Google's Gemini AI to explain them in plain English.

Here is the exact step-by-step configuration I used to build the "CPI-GEMINI" flow:

Step 1: The Exception Subprocess Strategy I moved the error handling logic out of the main process into a dedicated Exception Subprocess. This ensures that any failure in the main "Happy Path" immediately triggers this recovery logic.

Step 2: Constructing the AI Prompt (Content Modifier) To talk to the API, we need to format the request perfectly. I used a Content Modifier with two key parts:

Headers: Set Content-Type to application/json and Authorization to your Google API Key.

Body: This is where the magic happens. I wrapped the standard ${exception.message} variable inside the required Gemini JSON schema: {"contents": [{"parts":[{"text": "In cpi iflow following error encounter:${exception.message}"}]}]}.

Step 3: Calling the Model (HTTP Adapter) I used the standard HTTP Receiver adapter to POST the payload to the Gemini 2.5 Flash endpoint.

Address: https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent

Method: POST.

Step 4: Handling the Response (JSON to XML & Filter) The AI response is a large JSON object. To extract just the answer:

JSON to XML Converter: I converted the response and added a custom root element named root.

Filter: I used an XPath expression to drill down to the exact text node: /root/candidates/content/parts/text.

Step 5: Logging the Insight (Groovy Script) Finally, I used a simple Groovy script to grab the filtered AI explanation and attach it to the Message Processing Log (MPL).

Code: messageLog.addAttachmentAsString("Error_Body_Payload", body, "text/plain").

The Result: Instead of just seeing "System Error," the monitoring log now contains an attachment where Gemini explains the likely cause of the failure based on the error context.

https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key={YOUR_A...}

<xs:schema attributeFormDefault="unqualified" elementFormDefault="qualified" xmlns:xs="http://www.w3.org/2001/XMLSchema">
  <xs:element name="root">
    <xs:complexType>
      <xs:sequence>
        <xs:element name="candidates">
          <xs:complexType>
            <xs:sequence>
              <xs:element name="content">
                <xs:complexType>
                  <xs:sequence>
                    <xs:element name="parts">
                      <xs:complexType>
                        <xs:sequence>
                          <xs:element type="xs:string" name="text"/>
                        </xs:sequence>
                      </xs:complexType>
                    </xs:element>
                    <xs:element type="xs:string" name="role"/>
                  </xs:sequence>
                </xs:complexType>
              </xs:element>
              <xs:element type="xs:string" name="finishReason"/>
              <xs:element type="xs:float" name="avgLogprobs"/>
            </xs:sequence>
          </xs:complexType>
        </xs:element>
        <xs:element name="usageMetadata">
          <xs:complexType>
            <xs:sequence>
              <xs:element type="xs:byte" name="promptTokenCount"/>
              <xs:element type="xs:short" name="candidatesTokenCount"/>
              <xs:element type="xs:short" name="totalTokenCount"/>
              <xs:element name="promptTokensDetails">
                <xs:complexType>
                  <xs:sequence>
                    <xs:element type="xs:string" name="modality"/>
                    <xs:element type="xs:byte" name="tokenCount"/>
                  </xs:sequence>
                </xs:complexType>
              </xs:element>
              <xs:element name="candidatesTokensDetails">
                <xs:complexType>
                  <xs:sequence>
                    <xs:element type="xs:string" name="modality"/>
                    <xs:element type="xs:short" name="tokenCount"/>
                  </xs:sequence>
                </xs:complexType>
              </xs:element>
            </xs:sequence>
          </xs:complexType>
        </xs:element>
        <xs:element type="xs:string" name="modelVersion"/>
      </xs:sequence>
    </xs:complexType>
  </xs:element>
</xs:schema>

import com.sap.gateway.ip.core.customdev.util.Message
import com.sap.it.api.msglog.MessageLogFactory

def Message processData(Message message) {
    // 1. Access the Message Log
    def messageLog = messageLogFactory.getMessageLog(message)
    
    // 2. Get the current Body as a String
    // This captures whatever payload is currently in the flow (Error response or Request)
    def body = message.getBody(String.class)
    
    if (messageLog != null) {
        // 3. Attach it to the Monitoring Log
        // "Error_Body_Payload" is the name of the link that will appear in the Monitor
        if (body != null && !body.isEmpty()) {
            messageLog.addAttachmentAsString("Error_Body_Payload", body, "text/plain")
        } else {
            messageLog.addAttachmentAsString("Error_Body_Payload", "Body was empty.", "text/plain")
        }
    }
    
    return message
}