User Manual

Use AI Mode Safely

What NDCTerm's AI mode is for, how approval gates protect high-impact actions, and when the typed commands are the better tool.

What AI mode is

AI mode takes a request in plain language and turns it into the same commands this manual teaches. "Cancel this booking if the refund is over $500" becomes a tool plan: retrieve the order, obtain the carrier's cancellation quote where available, compare, draft the traveler email — and then stop and show you. The AI works against your real session and real orders: fare data and order state are live, and a reversal amount is identified as either carrier-quoted or a stored-fare estimate. It is not a simulation — which is exactly why it's gated.

Approval before commitment

High-impact actions — booking, ticketing, cancelling, refunding, exchanging, charging a paid seat or service — require your approval before they execute. The AI can retrieve, quote, reprice, and explain freely, because none of that commits anything; the moment a step would touch money or a live booking, it presents what it intends to do and waits.

This mirrors how the commands themselves behave — CANCEL quotes and prompts Y/N before acting — so the AI having quoted something never means it happened. Read what it's asking to commit exactly as you'd read a quote before your own Y.

When to use it — and when not to

For routine work, the typed commands are faster and exact. If you already know the next keystroke — *ABC123, $MAP, REFUND — typing it beats describing it, every time. AI mode earns its place on intent you'd otherwise have to decompose yourself:

  • "Is it cheaper to refund this or take the credit?" — two quotes, compared, neither committed.
  • "Find replacement flights for the cancelled segment and quote the change."
  • "Summarize this booking and draft an update to the traveler."

The pattern: use AI mode when the question is the work; use commands when the action is the work.

AI mode is still evolving — treat this page as its contract (real data, approval before impact) rather than a tour of its features. For the product-level story, see Agentic AI.