medOS AI
AI Coworkers
Named agents with a license to recommend
Coder, nurse, pharmacist, RCM, admissions and marketing coworkers run as first-class identities with plane-scoped DB grants. Every proposal lands in the Acknowledgement Inbox; accept/edit/reject is the training label.
Suggest ICD-10 I50.0 — heart failure
Chart review found CHF documentation. Adds 2 secondary codes (E11.9, I10) for DRG specificity.
Interaction flagged — warfarin + NSAID
Active warfarin order conflicts with new ibuprofen Rx. Bleeding risk — recommends paracetamol switch.
Draft appeal — denied claim REP-1142
Denial code 042 (missing authen code). Appeal letter drafted with authen attachment, ready for review.
accept / edit / reject = the training label
Capabilities
cowork_agents identity registry with per-agent audit trail
Three data planes — Clinical / Operational / Growth — enforced at DB-grant level
Proposals → Inbox → Accept / Edit / Reject; the decision trains the model
Growth plane is consent-walled with zero clinical grant
Off by default, per-tenant kill switch