Written for the person
who has to approve
the install.
This page describes how the system is built rather than what we promise. If something here would fail your review, tell us which line and we will answer it directly.
Inference runs on the device in the room
Cameras terminate on an NVIDIA Jetson class device installed inside the operating room. Every model runs on that device. Raw video is decoded, processed and discarded in memory on the same machine that captured it.
There is no cloud inference path, and the device does not require outbound internet access to do its job. On a site that permits it, an outbound channel carries only event records and health telemetry.
The only thing that leaves the room
Anonymised event records: a timestamp, an event type, the entity classes and roles involved, and a confidence. No frames, no crops, no audio.
Where a site enables review clips for a flagged event, those clips are held on the device and expire on a retention window agreed before install. The default is short and the expiry is enforced by the device, not by a policy document.
Faces, names and audio are out of scope
The state model works on body landmarks, entity classes and roles. It does not run face recognition and it does not build identity across cases. Faces are blurred at the sensor before a frame reaches a model.
No patient identifiers, no names, no audio, no linkage to the medical record. The system is designed so that a record leaving the device cannot identify a patient or a member of staff.
Access and audit
Role-based access, scoped to the roles a hospital already has: infection control and the OR director by default. Every access to a record is itself logged.
The audit log is append-only from the perspective of the application. If an event was seen, the record of who looked at it and when cannot be quietly removed.
The limits, stated before you ask
Gray-1 is advisory. It produces operational and compliance analytics and makes no diagnostic determination. A person reviews every flag, and clinical decisions stay with the surgical team.
We do not describe the system as certified for anything it is not. Where a deployment would require a regulatory pathway, that gets settled before an install, in writing.