RhythmX AI

The recommendations are the product. RhythmX AI's precision care platform reads the chart and the encounter against guidelines, payer and formulary rules, social context and care gaps, and returns patient-specific recommendations: labs, imaging, medications, follow-ups, coding and documentation opportunities, and routing to the right clinician. Its newest form, RhythmX Pulse, runs inside Microsoft Dragon Copilot and turns each captured encounter into recommendations.

The consolidated patient view is not sold apart from the recommendations. In a combination announced in November 2025 the company joined Get Well's patient engagement business as GW RhythmX, and nothing says whether the precision care platform is licensed on its own or bundled with engagement.

The review structure is published, and the point at which the system holds back is not. The privacy policy (version 2.0, 3 Jun 2026) says the AI takes no autonomous clinical action, issues no time-critical alerts, is not directed to patients, and that every recommendation is informational, carries inline attribution to its source guideline and requires clinician review before any action; the Dragon Copilot release says each recommendation can be accepted, edited or dismissed. Nothing says what the system does when the record is thin or the guidance conflicts, and the home page's routing of patients to a clinician type and channel is not described as reviewed.

The approach and the platform supplier are named without models or versions. The platform is built on SAI Group's Eureka generative AI platform and combines predictive models with generative recommendations; the privacy policy describes guideline retrieval from US and international medical societies, presented with the patient's record context and inline attribution. No model, training data, version or evaluation method is published.

The architecture is described and no model provider is named. The platform runs on SAI Group's Eureka generative AI platform, and the privacy policy lists cloud hosting, security, support, analytics and AI infrastructure vendors as subcontractors under the business associate agreement, affiliates SymphonyAI and ConcertAI as possible recipients, and Microsoft as a separate business associate where Dragon Copilot carries the product. None of the infrastructure or model vendors is named.

One named deployment with dated figures. Presbyterian Healthcare Services expanded the platform to its 200 primary care clinicians in February 2026; the release says the pilot surfaced more than 200 HCC and non-HCC conditions across more than 10,000 encounters within weeks and delivered 17,000 personalized treatment responses, and its chief medical information officer is quoted on consolidated information at the point of care. Sentara Health selected the platform for a pilot in September 2026. The figures are volumes, not measures of whether recommendations were correct or acted on, and there is no independent evaluation.

No training on customer data is committed, and the retention schedule sits in each contract. The privacy policy says customer PHI and content (transcripts, draft notes, prompts, outputs) are not used to train, fine-tune or improve general-purpose or shared models without the customer's written authorization, that any customer-specific adaptation stays in the customer's environment or under contract, that customer data is logically segregated, and that PHI is returned or destroyed at termination under the business associate agreement.

Retention periods themselves are left to each agreement. The company also markets access to its group's longitudinal data on 300 million patients and 4.4 billion annual claims without saying where it came from.

Business associate status is stated in a privacy policy that covers the product. The policy says RhythmX AI acts as a business associate of its health system customers under a business associate agreement that governs permitted uses, safeguards, breach notification (no later than 60 days) and termination, that it does not sell PHI or use it for marketing, and that where the product runs through a partner platform both companies are business associates under separate agreements. The agreement itself is not published.

Controls are described with nothing independent named behind them. The privacy policy lists encryption in transit and at rest, role-based access with logged administration, logical isolation between customers, key management, security logging, regular vulnerability scanning and penetration testing, and an incident response program, and says security and compliance documentation is available to customers and qualified prospects under confidentiality. Nowhere in its published material does the company name an attestation, framework, tester or cadence.

A published statement of intended purpose, beside marketing that goes further. FDA's 510(k) database lists no clearance. The privacy policy says the AI does not diagnose, treat, screen or predict in a manner intended to direct clinical care, takes no autonomous action, issues no time-critical alerts, does not process images or device signals, is used only by licensed professionals, and shows the source guideline for every recommendation so the clinician can review the basis. The home page still markets risk stratification and prediction, disease progression trajectories with predictions for care plans, and patient routing.

Physician review is claimed at scale without results. The company says it has completed more than 170,000 physician-reviewed clinical validations with physician-led governance and quality assurance built into every recommendation, and it names a clinical advisory board (executives of Sentara Health, Prime Healthcare, Mass General Brigham and Geisinger) and an expert panel. No validation method, acceptance rate or subgroup result is published. The recommendations draw on social, financial and lifestyle data, but nothing shows how they perform across patient groups or what the 170,000 validations measured.

A reviewer can trace each recommendation, and nothing stands behind it. Every recommendation cites its originating guideline inline, and clinicians can accept, edit or dismiss it. No error rate, acceptance rate by recommendation type, warranty, indemnity or remediation commitment is published. Recommendations span tests, imaging, medications and coding, which carry different costs when wrong, and performance and override rates are not broken out by recommendation type.

Epic in production and Dragon Copilot integration, with the write path undescribed. Presbyterian's clinicians use the platform in a consolidated workflow inside Epic, and RhythmX Pulse consumes the encounter captured by Dragon Copilot; the privacy policy says the AI reads structured and unstructured record data through standard interfaces. The company calls the platform EHR agnostic without naming another EHR, and nothing says whether accepted recommendations are written back to the chart.

Delivery routes, country and isolation are stated; the hosting path is not. The platform runs inside the customer's EHR (Epic at Presbyterian) and, for RhythmX Pulse, inside Dragon Copilot through Microsoft Marketplace. The privacy policy says the services are operated in the United States and that customer data is logically isolated, and it names cloud hosting and AI infrastructure only as categories of service provider.

The purchase routes are discoverable; the price is not. The platform is sold to health systems, and since July 2026 RhythmX Pulse can be added through Microsoft Marketplace by health systems already using Dragon Copilot, with no separate implementation. The company names no unit of charge or tier.

Primary care, with one named system behind it. The platform targets primary care physicians, advanced practice providers and care teams managing chronic conditions, and routes to specialists; Presbyterian's 200 primary care clinicians are the named deployment and Sentara Health selected it for a primary care pilot in September 2026. Validation in specialty or inpatient settings is not published.