Weight loss members on Hims can now work with a care engine that keeps a running picture of their treatment. The platform, already live inside the company’s labs business and the Hers app, replaces message threads that began with no memory of the person writing them.
The platform keeps a four-tier memory of the patient: intake history and stated goals, live program data such as dose schedules and weigh-in trends, lifestyle context gathered across earlier conversations, and the session in progress. When someone logs nausea, the engine checks the treatment timeline against that history and either offers non-clinical guidance or hands the case to a clinician with the context attached.
Hims describes the design around what it calls a 4G safety protocol. Inbound questions are sorted by intent, and anything requiring a clinical decision is blocked from autonomous handling. Clinician-written protocols become programmatic rules that outputs must trace back to, and responses are scored by a separate AI evaluator alongside human audit sampling.
Before release, model checkpoints go through offline red-teaming and full multi-turn simulations judged on clinical pass-fail benchmarks rather than language-model fluency metrics.
The commercial argument is data. Hims says the loop tightens with every titration and side-effect report, producing an asset that generic chatbots and ordinary fee-for-service telehealth networks cannot copy.
