A deterministic crisis gate runs in code before any AI model sees a message, answering emergencies with fixed safety instructions instead of AI-generated text.
A deterministic crisis gate runs in code before any model sees the message. That order matters. An emergency isn't left for a language model to recognize and respond to correctly in the moment; the code checks first, and only routine messages ever reach the model that generates an answer.
Deterministic means the check isn't a judgment call made fresh each time. It's a fixed piece of code, running the same way on every message, catching crisis and emergency situations before anything else happens. That's different from hoping a model interprets a concerning message correctly based on its training. The gate doesn't have an off day.
Safety guardrails are tuned per field, because an emergency looks different depending on the business. A gym's standing safety rules decline to advise on an injury and offer a person, passing contact to staff only with permission — a genuine emergency is the only case the gate catches, where the fixed reply tells the visitor to call 911. A med spa's assistant carries a no-medical-advice line with a handoff to a clinician for anything that needs clinical judgment. The gate isn't one generic rule bolted onto every business; it's shaped around what "urgent" actually means for that kind of business.
Once the gate identifies a message as urgent, the reply is fixed text written in advance — it tells the person exactly what to do right now, like leaving the area and calling 911 — instead of letting a model improvise. It deliberately does not continue the conversation. Ordinary requests to reach a person take the normal path instead: the assistant asks permission, captures the visitor's name and contact, and sends that to the owner's inbox. The point of the gate isn't to have the AI handle the emergency; it's to make sure the AI never tries.
A visitor on an HVAC company's chat writes: "I smell gas in my house right now, what do I do?"
The gate catches this before any model generates a response, and the reply is the fixed instruction written for exactly this: "Leave the area immediately and call 911 or your local fire department from a safe location. Do not use switches, plugs, flames, or anything that could create a spark, and do not go back inside until emergency responders say it is safe."
A sore shoulder or a routine injury question, by contrast, is not a gate case — it's handled by the assistant's standing rules, which decline to give medical advice and point the visitor to staff. The intercepted reply is fixed text — it gives the visitor immediate instructions rather than continuing the conversation.
No. It runs in code before any model sees the message. The model never gets the first look at an urgent message; the gate does.
The gate itself runs the same way for everyone, but the guardrails behind it are tuned per field. A gym routes injuries to staff and 911. A med spa hands medical questions to a clinician. The routing matches the business.
The assistant hands off to a human. Their name and contact are captured with their permission and sent to the owner's inbox right away, regardless of the hour. Try it free: a private preview, built in under 24 hours, no card required, at askmy.business.
Give us your website and we build your private assistant — free, usually within a day, no card. Nothing goes live without your approval.
Build my free previewUpdated 2026-09-01 · AskMyBusiness by Vision Genesis · Knoxville, TN