No magic. Just a careful chain of well-tuned parts.
Type whatever you'd actually want to know — symptoms, medication names, lab results, "is this normal?" Dr. ICU is built to handle the way humans actually ask, not the way medical textbooks want you to.
Your question is sent to Anthropic's Claude Sonnet — one of the strongest reasoning models available. We pair it with a system prompt that's been hand-tuned for medical conversations: cite your sources, distinguish strong evidence from weak, never invent statistics, never play doctor.
Conversational at the top, then the relevant evidence, then the practical "what should I actually do." When studies are referenced, they're cited inline. When the answer is uncertain, the answer says so.
If your question maps to a service (a lab test, a telehealth provider, an at-home device), Dr. ICU may suggest a vetted option. We may earn a commission if you use the link. This never changes what the AI tells you — the medical answer is generated before the pathway is selected.
Dr. ICU runs on Anthropic's Claude Sonnet, accessed via a server-side proxy that holds our API key (your browser never sees it). Streaming responses are rendered token-by-token via Server-Sent Events for a real-time feel. The 3D character is a custom GLB model rendered with Three.js, with idle animations driven by a small physics simulation.
Free: 5 questions per day. Plenty for casual use.
Pro ($9/month): Unlimited questions, deeper research mode (longer answers with more citations), and early access to new features.
No ads. No data resale. Pro just funds the API bills.
Dr. ICU is open about its limits. The disclaimer is one click away on every page. The AI itself will tell you when a question is outside its competence. We'd rather lose a query to "see a doctor" than confidently mislead you.