AI support ticket triage

1. A trigger receives the ticket: customer, subject and message.
2. A Classify agent returns urgency and topic as structured output instead of free text.
3. A Switch fires urgent when urgency is high, and default otherwise.
4. The urgent branch builds an escalation with a Set node; the default branch drafts a reply with a second agent.
5. A Merge joins whichever branch ran, and the Output node marks the result.

Give the classify agent an output schema so later nodes branch on fields, not on parsed text. Novalink makes the schema strict for you.

{
  "type": "object",
  "properties": {
    "urgency": { "type": "string", "enum": ["low", "normal", "high"] },
    "topic": { "type": "string", "enum": ["billing", "bug", "account", "other"] }
  },
  "required": ["urgency", "topic"]
}

nodes.classify.output.structured.urgency == 'high'

When a ticket is misrouted, open the run: the skipped branch is dimmed, and the classify node shows the exact prompt it received and the fields it returned. Adjust the prompt and re-test just that node against the same ticket.

• Add Switch cases per topic to route billing and bugs to different queues.
• Post the result to your helpdesk with an HTTP Request node.
• Set the reply agent's error mode to error_port and route failures to a human review branch.