n8n Auto-Triage Chat Messages to ServiceNow Incidents, Requests, and AI Agent

Chat input → LLM triage → ServiceNow incident / catalog request / AI agent with search & memory → optional summary

Who

IT service desks or ops teams already using n8n and ServiceNow

You get

Automatically classify chat messages into Incident, Service Request, or Other; create ServiceNow incidents, submit service catalog requests via HTTP, or hand off to an AI agent with web search and memory; optionally summarize results

You need

n8n

Steps

1. Configure chat trigger and classifier model

   Add a 'When chat message received' node as the workflow trigger. Add an 'OpenAI Chat Model' sub-node, select gpt-4.1-mini (or gpt-4.1 / gpt-4), and bind an openAiApi credential. This model powers the Text Classifier.

2. Set up the text classifier with three-way prompt

   Add a 'Text Classifier' node and connect the OpenAI Chat Model from the previous step. Keep the default input '={{ $json.chatInput }}'. Configure three categories: Incident (breakage/complaint), Request (access provisioning/resource order), and Everything Else (knowledge search, etc.). Set the systemPromptTemplate to the source prompt.

   Execution prompt · Source text。Text Classifier node options.systemPromptTemplate field

   Copy prompt

   Please classify the text provided by the user into one of the following categories: {categories}, and use the provided formatting instructions below. Don't explain, and only output the json.

3. Configure the ServiceNow incident creation branch

   From the Text Classifier's Incident output, add a 'Create an incident' node (ServiceNow type) and bind a serviceNowBasicApi credential. Keep operation as create, resource as incident, and short_description as the default expression '={{ $('When chat message received').item.json.chatInput }}'. Connect this branch output to the Summarization Chain.

4. Configure the service catalog HTTP request branch

   From the Text Classifier's Request output, add an 'HTTP Request' node. Set method to POST, authentication to predefinedCredentialType with serviceNowBasicApi, and add header Accept: application/json. Replace the URL with your actual ServiceNow service catalog order endpoint; the template's default URL is the author's demo address and must be changed. Use a body structure based on the source example.

   Execution prompt · Editorial suggestion。HTTP Request node jsonBody field; editorial suggestion based on source example, replace with your actual catalog item endpoint

   Copy prompt

   ={
     "sysparm_quantity": 1,
     "variables": {
       "description": "{{ $('When chat message received').item.json.chatInput }}"
     }
   }

5. Configure the AI agent branch with search and memory

   From the Text Classifier's Everything Else output, add an 'AI Agent' node. Connect 'OpenAI Chat Model1' (gpt-4.1-mini) as the language model, 'Simple Memory' as memory, and 'SerpAPI' as a tool. Set the AI Agent's systemMessage to the source prompt. Connect this branch output to the Summarization Chain as well.

   Execution prompt · Source text。AI Agent node options.systemMessage field

   Copy prompt

   You are a helpful assistant. Who creates requests based on the user query using the tool. Always create the request using the http tool attached to you

6. Add the optional summarization chain and final output

   At the convergence point after all three branches, add a 'Summarization Chain' node and bind 'OpenAI Chat Model3' (gpt-4.1-mini). Set summarizationMethod to stuff and set the prompt to the source template. This node receives upstream outputs and generates a CONCISE SUMMARY.

   Execution prompt · Source text。Summarization Chain node options.summarizationMethodAndPrompts.values.prompt field

   Copy prompt

   =Write a user friendly summary of the following:
   
   
   Agent Action"{text}" 
   User Query"{{ $('When chat message received').item.json.chatInput }}"
   
   
   CONCISE SUMMARY:

You are setting up the n8n template 7515 for ServiceNow chat triage. Confirm the following before importing or building: 1) n8n instance access; 2) OpenAI API key (for classifier, AI Agent, summarization); 3) ServiceNow instance URL and an API user with incident_create rights; 4) optional SerpAPI key; 5) your actual service catalog item order URL (replace the template's demo address). After confirmation, import the workflow JSON or configure nodes step by step. Pay special attention to the Text Classifier's three categories, whether the AI Agent systemMessage fits your 'Everything Else' use case, and replacing the HTTP Request URL and body variables.