All demos

Support Ticket Triage

5 agents · ~2 min

Automates the confident calls, pauses for a human on the risky one

1 · Brief
2 · Live run
3 · Results

Support inbox

Nothing to fill in — kick it off and take the reviewer's seat.

The inbox is empty — for now

A ticket-writer agent invents a realistic, multi-issue customer ticket the moment you start. No two runs are alike.

What happens to the ticket

  1. Triage classifies severity and splits multi-issue tickets into parts
  2. Each part is grounded in the support knowledge base — scores shown
  3. Replies are drafted and routed by confidence
  4. Low-confidence branches pause for your sign-off
  5. Approved branches merge into one final resolution

You are the human reviewer. When a branch isn't confident enough to auto-send, the run pauses and waits for your decision — up to 3 minutes, then the recommended option applies automatically.

Waiting for human verification…

Meet the panel

  • Ticket WriterInvents the ticketstandbyTicket Writer: standby
  • TriageSeverity + categorystandbyTriage: standby
  • KB LibrarianKnowledge-base groundingstandbyKB Librarian: standby
  • The DrafterReply draftsstandbyThe Drafter: standby
  • EscalationConfidence routingstandbyEscalation: standby

What you'll get

  • Resolved ticket thread
  • Auto-resolved + human-approved branches
  • Escalation reasoning, kept visible
  1. 1A writer agent invents a realistic two-issue ticket — the demo self-starts.
  2. 2Triage splits it; every part is searched against a real policy knowledge base, retrieval scores shown openly.
  3. 3Replies are drafted per part; the least-grounded part always routes to a human — closest match isn't precedent.
  4. 4A real LangGraph interrupt() checkpoints the run while you approve one of three options (3-min window, then the recommended one applies).
  5. 5The graph resumes with your decision and merges everything into one resolution, escalation reasoning kept visible.