The Answers It Is Not Allowed to Give

How it works

You set the boundaries: topics Chatbot.OM must decline, figures it may never quote, discounts it may never offer, and categories that always go to a person. These are enforced before any reply is sent, in whichever language the conversation is happening in.

Where a rule fires, the chatbot says something useful rather than something evasive, and moves the conversation to whoever should own it.

• Topic bans applied across every language
• Limits on prices, discounts, and commitments
• Categories that always escalate, regardless of confidence
• Rules enforced before the reply is sent, not after

Why the limits are the product

Any modern language model can hold a plausible conversation. The commercially important question is what happens at the edges: when a customer pushes for a discount, asks for legal or medical advice, or tries to get a commitment your business cannot honour.

Guardrails are what make an automated conversation safe to put in front of customers without someone reading every message afterwards.

Sensitive categories in Oman

Some sectors have specific limits. Clinics should not be giving clinical advice in a chat window, law firms should not be giving legal opinions, and financial services have rules about what may be said about products before a customer is identified.

Those constraints are configured explicitly rather than hoped for, and the resulting behaviour is testable before you go live.