Theator

Everything Theator sells is produced by its models from raw surgical video. A proprietary vision language model reads the camera feed and identifies steps, instruments, anatomical landmarks, safety milestones, events and case complexity, and that structured record becomes the operative report, the quality dashboards, the billing detail, the teaching library and the research datasets.

Capture is automatic, with no annotation or dictation by the surgical team, so the video unit in the room has nothing to offer on its own without the analysis behind it.

Surgery-to-Text drafts the operative report and the surgeon reviews, edits and finalizes it inside Epic or Oracle Health before it becomes the record, and clinical documentation specialists and coders then work from the signed report. A Miami pilot reported a median of 6 edits per case. Quality, safety and benchmarking outputs go to dashboards for surgeons, department chairs and quality officers, who decide what to act on.

Nothing explains how a surgeon or department disputes an automatically tagged event or complication before it reaches the benchmarks.

The engine is a proprietary vision language model trained on more than 600,000 curated procedures across 9 specialties and 180+ procedure types. It reads the whole case rather than single frames, identifying steps, instruments, anatomy, safety milestones, best practices, events, complexity and idle time, and patient context from the EHR arrives over HL7 and FHIR to complete each report. Published papers describe earlier models in detail, such as step recognition at 88.8 percent accuracy in inguinal hernia repair.

Model versions and the schedule on which the model changes are not published.

The model is Theator's own, a proprietary vision language model trained on its curated library of more than 600,000 procedures and covered by more than 30 patents. The cloud layer is AWS, with Oracle Cloud Infrastructure for Oracle Health customers, and the trust center carries a subprocessor list and a data processing agreement.

Whether an outside language model writes any part of the report text is not stated.

Peer reviewed studies written with hospital surgeons carry the evidence. A Mayo Clinic study in the Journal of the American College of Surgeons (2025) compared AI operative reports with surgeon written ones in 158 robotic prostatectomies, using expert review of the video as ground truth, and found overall accuracy of 87.3 percent against 72.8 percent. In Annals of Surgery (2024), a quality initiative in two general surgery departments used the platform to monitor the critical view of safety in 279 cholecystectomies, and monthly adoption rose from 33 percent to 76 percent. Other papers report 88.8 percent step recognition in inguinal hernia repair and warm ischemia times within seconds of ground truth in 61 partial nephrectomies.

The two headline studies ran at one and two sites, with Theator staff among the authors.

Identifying content is removed at capture, before video leaves the hospital. The video unit strips patient and staff identifiers and metadata, removes frames recorded outside the body at 99.3 percent accuracy and applies k anonymity of at least 3, a method Theator developed with former Mayo Clinic chief information officer Cris Ross and published in JAMA Surgery in 2023. De identified video then travels over a VPN to a HITRUST certified cloud, and Theator says every new procedure deepens the model.

Retention periods sit in a media data retention policy in the trust center rather than on the public site.

Business associate agreements are signed with every partner, Theator says, and its deployment page describes how protected information moves through the product: removed at capture before transmission, encrypted end to end, and held behind role based access with audit logging. Its trust center lists HIPAA among its compliance items alongside a data processing agreement, and the security package it offers hospitals includes a BAA template and data flow diagrams.

The agreement itself is not posted on the public site.

A SafeBase trust center lists SOC 2 Type 2, HITRUST, ISO/IEC 27001 and ISO 27799:2016, and the AWS partnership page names the HITRUST certification as r2. Documents in the trust center include a SOC 2 report, a penetration test report, a network diagram and a security whitepaper, and the controls it covers include application and network penetration testing, audit logging, incident response, business continuity and disaster recovery, alongside annual third party audits with continuous compliance monitoring.

The reports and certificates are released through an access request.

Surgery-to-Text and the quality, teaching and research tools are sold as documentation and analytics software, with no device claim. The terms of service describe content management and a visual operative report that surgeons use to review prior surgeries for educational purposes, and they rule out real time feedback during surgery and use for treatment decisions.

That description fits an earlier review tool better than Surgery-to-Text, which now writes the operative record and its billing detail into the EHR.

Model output is measured against expert video review in peer reviewed studies, and Theator reports results from the SAGES critical view of safety challenge, a benchmark run across several institutions. Its quality dashboards compare technique, timing and outcomes across surgeons and sites, with the same model applied to every case so that comparisons rest on one method.

Results are not broken out by procedure difficulty, patient anatomy or video quality.

Report errors are measured in a Mayo Clinic study that Theator cites on its own site. In Khanna et al., Journal of the American College of Surgeons (2025), 20 of 158 AI reports for robotic prostatectomy (12.7 percent) carried a clinically significant discrepancy against expert review of the video, and 46 (29.1 percent) carried at least one discrepancy of any kind, against 43 and 84 of the surgeon written reports. Every report is reviewed and signed by the surgeon before it enters the record.

The terms disclaim any warranty on reports and content, so no remedy attaches to report accuracy.

Reports land directly in the patient record in Epic and Oracle Health, with no separate portal or login. Patient context flows in over HL7 and FHIR and structured reports and data flow back out, and Epic's Showroom lists two Theator Connection Hub products, Theator EHR Connector and Theator Surgery to Text Launch. Oracle announced the Oracle Health collaboration in June 2026, with reports delivered into Oracle Health EHR for care teams, quality reviewers and billing staff, and UHealth Miami's Epic integrated pilot cut time to the final operative note by about 70 percent.

No hospital is named yet as running the Oracle Health connection.

A small video unit connects to the existing laparoscopic or robotic tower through DVI, SDI or HDMI, and de identified video travels over a VPN, with TLS 1.2 or later, to a HITRUST certified cloud where analysis, storage and delivery are fully managed. AWS is Theator's main technology partner, and for Oracle Health customers Surgery-to-Text runs on Oracle Cloud Infrastructure. Reports return into the EHR, while the quality, teaching and research tools run in a separate web app.

The countries or regions where video is stored are not stated.

The platform is sold as an all inclusive subscription, with no separate charges for storage, hardware or updates and deployment measured in weeks. Its buyer pages for surgeons, executives and revenue cycle leaders model the return: a documentation ROI calculator, at least $186 per case in documentation value from deployment data, and tiered estimates running from a 10 OR community hospital to a 60 OR academic center.

No subscription price or unit of charge is published.

Coverage follows the camera: laparoscopic and robotic procedures on any manufacturer's system, with 180+ procedure types across general surgery, colorectal, bariatric, urology, gynecology, thoracic, ophthalmology, ENT and gastroenterology. Named deployments include Mayo Clinic, Cleveland Clinic, McGill University Health Centre, UHealth Miami and Assuta Medical Centers in Israel, and the case studies come from academic centers and an eight hospital system.

Published studies cover robotic prostatectomy and partial nephrectomy in urology, cholecystectomy, inguinal hernia repair and gynecology, with no results yet for thoracic, ophthalmic or ENT cases.