AI-powered ServiceNow chat triage with GPT-4 — Incident & request router

Short description

Automatically triage incoming chat messages into Incidents, Service Requests, or Other using an LLM-powered classifier; create Incidents in ServiceNow, submit Service Catalog requests (HTTP), and route everything else to an AI Agent with web search + memory. Includes an optional summarization step for ticket context.

Full description

This n8n template wires a chat trigger to an LLM-based Text Classifier and then routes messages to the appropriate downstream action:

1. Trigger: When chat message received — incoming messages from your chat channel.

2. Text Classifier: small LLM prompt/classifier that returns one of three labels: Incident, Request, or Everything Else.

3. Create Incident (ServiceNow connector): when labeled Incident, the workflow creates a Servicenow Incident record (short fields: short_description, description, priority, caller).

4. Submit General Request (HTTP Request): when labeled Request, the workflow calls your Service Catalog API (POST) to place a catalog item / submit a request.

5. AI Agent: when labeled Everything Else, route to an AI Agent node that:

   • uses an OpenAI chat model for contextual replies,
   • can consult SerpAPI (web search) as a tool,
   • saves relevant context to Simple Memory for future conversations.

6. Summarization Chain: optional chain to summarize long chat threads into concise ticket descriptions before creating incidents/requests.

This template is ideal for support desks that want automated triage with human-quality context and searchable memory.

Key highlights (what to call out)

• Three-way LLM triage: ensures messages are routed automatically to the correct backend action (Incident vs Service Request vs AI handling).
• ServiceNow native connector: uses the ServiceNow node to create Incidents (safer than raw HTTP for incidents).
• Service Catalog via HTTP: flexible — supports organizations using RESTful catalog endpoints.
• Summarization before ticket creation: produces concise, high-quality short_description and description fields.
• AI Agent + Memory + Web Search: handles non-ticket queries with web-augmented answers and stores context for follow-ups.
• Failover & logging: include a catch node (optional) that logs failures and notifies admins.

Required credentials & inputs (must configure)

• ServiceNow: Instance URL + API user (must have rights to create incidents).
• Service Catalog HTTP endpoint: URL + API key / auth header (for POST).
• OpenAI API key (or other LLM provider): for Text Classifier, Summarization Chain, and AI Agent.
• SerpAPI key (optional): for web search tools inside the AI Agent.
• Memory store: Simple Memory node (or external DB) for conversation history.

Nodes included (quick map)

• Trigger: When chat message received
• Processor: Text Classifier (OpenAI/LLM)
• Branch A: ServiceNow (Create Incident)
• Branch B: HTTP Request (Service Catalog POST)
• Branch C: AI Agent (OpenAI + SerpAPI + Simple Memory)
• Shared: Summarization Chain (used before A or B where enabled)
• Optional: Error / Audit logging node, Slack/email notifications

Recommended n8n settings & tips

• Use structured outputs from classifier ({ label: "Incident", confidence: 0.92 }) so you can implement confidence thresholds.
• If confidence < 0.7, route to a human review queue instead of auto-creating a ticket.
• Sanitize user PII before storing in memory or sending to external APIs.
• Rate-limit OpenAI/SerpAPI calls to avoid unexpected bills.
• Test the Service Catalog POST body in Postman first — include sample variables JSON.

Short sample variables JSON (Service Catalog POST)

{
  "sysparm_quantity": 1,
  "variables": {
    
    "description": "User reports VPN timeout on Windows machine; error code 1234"
  }
}