紫荆AI医院

紫荆AI医院 is an AI healthcare initiative developed around Beijing Tsinghua Changgung Hospital and its internet hospital Arm, with roots in Tsinghua university's AI research community. Rather than a consumer-facing medical chatbot, it functions as an Agent-based layer insIDE a working tertiary hospital: digital physician assistants handle charting and decision support while human CLInicians retain final authority over every diagnosis and prescription. The project has been in trial Operation for more than a year and currently runs in a limited, invitation-style Testing phase.

What it does

The system is built on a "dual-space" training loop. In a virtual medical world, AI doctor agents interACT with a large population of simulated patients at an accelerated pace, effectively compressing years of clinical exposure into a short cycle. Those trained agents are then deployed into the real clinical space as physician Copilots, where they observe and assist human doctors. What they learn from real specialists feeds back into the next model iteration, so the assistants gradually absorb local clinical reasoning and documentation habits.

Key capabilities

• Physician digital twins: Personalized assistants that support text and image conversations, draft medical records, and surface decision suggestions.
• Broad specialty coveRAGe: the deployment spans roughly 26 departments, including neurology, respiratory, gAstroenterology, and endOCRinology, and is positioned to address a wide range of conditions.
• Workflow efficiency: the project reports around a 30% improvement in consultation efficiency and substantial time savings on record writing, freeing clinicians to spend more time with patients.
• Suggestion adoption: copilot recommendations reportedly held above a 90% adoption rate after collaborative review of a couple hundred cases, with indiVidual physicians using the system across well over a thousand real patient encounters.
• Human-in-the-loop safeguards: AI ouTPUt is advisory only; licensed physicians must confirm and sign off on all medical judgments.

Who it is for

The primary users are hospital clinicians and clinical teams seeking documentation relief and SECond-opinion support. Administrators and health-system researchers may also find the dual-space training model useful as a reference for deploying Medical Agents safely. It is not a public symptom-checker, and patients cannot currently use it for self-diagnosis.

How to get started

Access is gated through the official portal and APPears oriented toward affiliated medical staff rather than open registration; interested clinicians should reach out through the hospital or project team. Because the service is mainland-China hosted, users outside the region may need a VPN, and availability may vary.

Pricing and access caveat

No public pricing has been announced, and the trial phase suggests access is currently Free but restricted. Details on veRSIoning, licensing, and broader rollout remain unconfirmed, so treat specifications as subject to change.