How NOX Leo Screens Health Conversations
Leo screens every Nox health conversation for acute medical red flags before an AI-generated answer appears. It uses a deterministic first check for urgent warning signs, adds further screening based on your chosen safety setting, and surfaces clear guidance to seek care when a likely emergency is recognized.
How Leo screens a health message
When you send a message in Nox, Leo runs before the conversational AI responds. Leo is Nox’s user-facing medical-safety system, designed to identify signs of acute red-flag conditions in free-form health conversations.
That order matters. A health conversation can include an everyday question, a vague symptom, or a description that may need immediate attention. Leo’s role is not to diagnose the cause or decide what condition someone has. It is to recognize potentially serious patterns and make sure safety guidance appears before an AI answer could distract from the need for care.
The first layer is independent and deterministic. In practical terms, it uses fixed rules to screen each message before any AI is called. If one of those rules is triggered, Nox displays an emergency or urgent-care banner before showing AI-generated content.
For a closer look at the design behind that sequence, read How Leo Works.
What health red flags does Leo look for?
Leo’s deterministic triage layer covers more than 100 rules across over 70 acute red-flag categories. These include signs associated with stroke, chest pain and cardiac symptoms, severe breathing difficulty, severe allergic reactions, trauma and severe bleeding, mental-health crisis, pregnancy warning signs, concerning symptoms in children, environmental emergencies, and toxicology or overdose.
These categories are not diagnoses. A red flag is a feature of a message that can signal a need for urgent evaluation, such as a sudden severe symptom, a dangerous exposure, or a change that could be time-sensitive. It is appropriate to treat a red-flag screen as a prompt to seek care—not as proof that a particular emergency is happening.
The system is designed for people speaking naturally rather than filling in a rigid questionnaire. That matters because real health messages are often incomplete: “My chest feels weird,” “I can’t catch my breath,” or “Something is very wrong.” People may also describe symptoms in a different language, use slang, or refer to an older term for a condition.
For more context on the difference between a warning sign and a diagnosis, see What Are Medical Red Flags?.
What happens when a rule is triggered?
When Leo’s deterministic layer identifies a relevant red flag, Nox shows clear emergency or urgent-care guidance before any AI-generated response. If the situation appears to be an emergency, the right next step is to contact local emergency services.
Nox can show a local emergency number when you have set your region in Settings. If the region is not known, it falls back to universal emergency numbers—911, 999, or 112—so the guidance is not left blank.
This approach aims to make the safety message prominent rather than burying it at the end of a long explanation. In a time-sensitive situation, the most helpful response is not a detailed theory about symptoms. It is a clear prompt to get appropriate human help.
Leo can support a safety-first interaction, but it cannot replace emergency services, a clinician, or professional assessment. If you believe you or someone else may be experiencing an emergency, contact local emergency services right away.
How Leo handles slang, indirect language, and multiple languages
Fixed rules are useful because they are consistent and measurable. But people do not always describe urgent symptoms in a predictable way. They may use slang, indirect wording, an older disease name, or another language.
When Leo’s deterministic layer finds no match, a lightweight AI classifier performs a second check in the background. It re-reads recent messages to look for dangerous descriptions that fixed patterns may miss. If it recognizes a likely emergency, Nox surfaces the same seek-care guidance used for a pattern match.
This second layer is a backstop, not a replacement for the deterministic screen. It can only add a safety note; it cannot remove or soften a warning already produced by the deterministic layer.
Leo also includes multilingual coverage. Its deterministic rules cover multiple languages, and the additional classifier is intended to help with messages where wording is indirect or does not match a fixed phrase cleanly. Still, no automated system can reliably catch every emergency in every form of language. If symptoms are concerning, severe, sudden, or persistent, contact a qualified clinician; for emergencies, contact local emergency services.
Can you choose how much screening Leo uses?
Yes. Leo provides three safety levels directly in the Nox chat box: Relaxed, Standard, and Strict. They run from least to most protective.
Relaxed warns about clear emergencies only. Standard also warns on situations that may be urgent. Strict adds the AI backstop that re-reads recent messages for potentially dangerous descriptions expressed indirectly or in another language.
Whatever setting you choose, Leo always screens for true emergencies. The setting changes which optional layers run on top of that core emergency screen, not whether Nox looks for clear emergencies at all.
Agent and voice conversations always use the strictest setting. That creates an added margin of caution in interactions where the conversation may move quickly or contain more context across messages.
Why Leo uses more than one safety layer
Health AI safety is not the same question as whether an answer sounds accurate or helpful. A response can be well-written and still fail at the most important task if it overlooks a message that needs urgent action.
That is why Leo separates its deterministic red-flag detector from the conversational response. The first screen is fixed and runs before any AI answer. The second classifier is there to broaden coverage for language that rules may not anticipate, but it cannot override the first layer.
The deterministic detector’s overall recall and false-positive rate are measured against a maintained test set. Nox publishes a high-level summary of those metrics on its Trust & Transparency page. The classifier’s results are kept separate because it is not deterministic.
Measurement is important, but it should not be confused with a guarantee. Every screening system has limitations, including the risk of missed cases and false alarms. Leo is designed as a safety net, and Nox does not claim that it catches every emergency.
For a deeper explanation of why strong answer quality alone does not settle the safety question, see Why High Model Accuracy Is Not Enough for Health AI.
What Leo does not do
Leo does not diagnose conditions, prescribe treatment, or determine that a person is safe. It screens messages for acute red flags and surfaces guidance when the language suggests an emergency or urgent concern.
A person can have a serious health issue without using words that trigger a screen. A message can also trigger a precautionary warning without confirming an emergency. That uncertainty is one reason conversations about symptoms should remain grounded in professional care when symptoms are severe, sudden, worsening, persistent, or otherwise concerning.
Nox is an educational health and wellness companion, not a medical device or a substitute for a clinician. Leo’s purpose is narrower and critical: make a potential emergency harder to miss in a health conversation.
Common questions
Does Leo read every message before Nox answers?
Yes. Leo’s deterministic red-flag layer screens each message before any AI model is called. Depending on the selected safety level, additional screening may also run.
Will Leo always catch an emergency?
No. Leo is a safety net, not a guarantee. No automated system catches every emergency, so anyone who may be experiencing an emergency should contact local emergency services.
Does Leo diagnose the cause of symptoms?
No. Leo screens for acute red flags and surfaces guidance to seek appropriate care. It does not diagnose, treat, or prescribe.
Why might Leo show a warning when I am not having an emergency?
Safety screens may use precautionary warnings when a message resembles a potentially urgent pattern. A warning is not a diagnosis; it is a prompt to consider timely professional care.