You text your number, a trained human triage responder replies in the same thread in minutes, and the AI consultation begins there: your history, follow-up questions, then what it thinks and what it suggests.
Under the hood, it is a medical AI system with strict safety rules: it knows its limits, defaults to caution, relies on evidence, applies the same rules every time, puts emergencies first, and shows its work. Its evaluations are published, not just claimed. It is not a licensed doctor, and it says so.
What to do
Common questions
How does the Pymander AI doctor work?
Four steps. One: you text your number, and a trained human triage responder replies in the same thread in minutes. Two: the AI consultation begins right there - it takes your history, asks the follow-up questions a clinician would ask, and works through the possibilities. Three: it tells you what it thinks and what it suggests, in plain English, including self-care when that is the right answer. Four: when something needs an exam, a lab, or imaging, it says so plainly and points you to the right level of care. The safety rules run underneath all of this: the system knows its limits, defaults to caution, relies on evidence, applies the same rules every time, treats emergencies as emergencies first (chest pain, trouble breathing, stroke signs - 911, not a text), and shows its work. The honesty part matters too: it is a medical AI system, not a licensed doctor; it cannot prescribe or order tests, and it does not pretend otherwise. And the performance claims are public: 98.2% on USMLE-style questions and a published HealthBench Hard evaluation, with the safety architecture written up openly - all linked below, because "trust us" is not a safety argument.