Sage is the AI companion inside Salve. It reads the health picture you have entered and helps you see patterns and prepare for appointments. This page explains the rules Sage operates under, so you know exactly what it will and will not do.
Sage never diagnoses a condition, and never recommends, changes, or tells you to stop a medication or a dose. If you ask it to, it will name that limit warmly and point you back to your care team. Every AI surface in Salve carries the same line: AI suggestions are not medical advice. Always consult your healthcare providers.
Sage describes what it observes ("pain tends to be higher on days with less sleep") and does not claim cause ("less sleep causes your pain"). When a pattern is based on fewer than about two weeks of data, or the signal is weak, Sage says so plainly rather than overstating it. Findings are framed for your next visit, never as something your doctors missed.
The correlations Salve surfaces are produced by a deterministic engine that runs in your browser over your own records. The AI is used to put those findings into plain language, not to invent them. It looks for recurring symptoms, relationships between things you track, and triggers, and it presents them as observations to bring to a professional.
If you describe symptoms consistent with a medical emergency (for example chest pressure, trouble breathing, one-sided weakness or slurred speech, the worst headache of your life, throat swelling, fainting, or severe bleeding), Sage stops and tells you to call 911 or get to the nearest emergency room now. It does not try to triage or talk you through it. This is enforced in two independent places: a deterministic on-device detector that opens an emergency-resources screen, and a hard instruction in the AI's own system rules. The same applies to mental-health crisis, which routes to 988 and the Crisis Text Line.
Sage is instructed never to fabricate citations, study names, statistics, guideline names or version numbers, dates, or links. It states general medical facts as general knowledge and points to reputable sources or your care team rather than naming a source it does not actually have. For answers backed by a web search, it only cites sources that were actually in the results.
Your health, symptoms, medications, conditions, labs, vitals, cycle, genetics, and mental health; warm support for what you are going through; how to use Salve or move data in from another app; and patient self-advocacy in the US system (decoding denial letters, drafting appeals, preparing for difficult appointments), always grounded in your own record. It declines requests outside that scope rather than guessing.
Salve routes simpler tasks to lighter models and complex analysis to stronger ones, on either the free (Google Gemini) or premium (Anthropic Claude) tier you have selected. AI runs only with your explicit, revocable consent, and your profile text is pseudonymized by default before it is sent. See Transparency for who processes what, and Security for how data is protected.