TransformativeMed

A learned component is now in the product, sitting on top of a workflow platform that works without it. The CORES platform page describes two model driven layers: a semantic AI data platform that extends the record's data model for large language models and turns free text clinical notes into structured items such as expected discharge dates, durable medical equipment needs, consult statuses and care plans; and multi agent orchestration, supplied by Vantiq, in which agents perceive live record signals, apply clinical logic and launch coordinated actions across care teams. Language models reading notes are a learned component, and the orchestration comes from a partner.

The value underneath stands without them. The company's long standing asset is more than fifty prebuilt workflow modules embedded in the record: specialty patient lists, rounding and handoff views, task management, alert routing through Core Notify, secure messaging, a mobile companion and a diabetes module, built through what the company describes as knowledge engineering and configuration. A hospital would still hold those workspaces with every model removed, and the platform page itself describes hybrid workflows that combine proven applications with agents.

Autonomy is claimed in strong terms and no oversight structure is published. The platform page describes an agentic clinical operating system in which multiple AI agents perceive, reason, plan and autonomously execute multi step clinical and administrative workflows, contrasting it with passive tools that only make suggestions; agents launch coordinated actions across care teams; and the integration reads and writes across orders, documentation, reference data and workflows inside the record.

The discharge product states it outright. Core Discharge describes autonomous chart review in which agents extract discharge tasks, equipment requirements, barriers and readiness indicators from documentation as it is written, and says no manual review is required; service agents then question the rehab, primary team, disposition and consult services by mobile and update the plan themselves. The team dashboard shows AI summaries with their source documentation on hover, which lets a reader check a summary they choose to open. A statement that review is unnecessary tells a buyer not to build the step that would catch a wrong estimated discharge date or a missed equipment need.

Nothing published says what the agents may not do, which actions need a clinician's approval before they take effect, what happens when an agent is unsure, or how a clinician sees and reverses an action an agent took. The only nod to people is that workflows are hybrid, combining the existing applications with agents, which describes the product mix rather than a control.

The older product was a different proposition: workspaces that assemble record data and route tasks and alerts, with clinicians doing the rest. That description no longer covers what is sold.

The architecture is named layer by layer and one supplier is identified; the version discipline is not. The platform page describes four layers: multi agent orchestration, stated to be powered by Vantiq, in which agents read live record signals and launch actions; a semantic AI data platform that extends the record data model for large language models and extracts structured items from clinical notes; more than fifty prebuilt workflow modules; and an MCP based interface layer for agent integration. The record integration is named precisely: SMART, FHIR and MPages, embedded in PowerChart. Naming the orchestration supplier shows where part of the intelligence comes from.

What is missing is the model layer and its change practice. No language model or provider is named, the page describes the platform as built to adopt quickly evolving language models without saying which it uses now or how a change is tested, and no accuracy figure is published for the note extraction, which drives discharge dates and equipment needs.

The older half of the product remains inspectable in a way learned systems are not: specialty workspaces built as configuration on the record's own extension toolkit can be read by the institution running them.

One party in the chain is named, Vantiq, and the model layer is not. The platform page states that multi agent orchestration is powered by Vantiq, so a buyer knows that a separate company's real time platform receives live record signals and runs the agents. It also describes a semantic AI data platform that prepares clinical notes for large language models, without naming which model or provider sits behind it, and says the platform is built to adopt quickly evolving models.

The embedded workspaces keep a short chain for the older product, since they run inside the customer's record through its own extension framework. The newer chain is the one that matters and it is undescribed: nothing says what Vantiq receives or retains, where note text is sent for the language model, whether any model provider sees identifiable notes, or whether those parties are bound by the same terms as the company. No subprocessor list exists, and the only privacy document on the site covers website visitors.

Long lived deployments, customer quotes with figures, and no measured outcome with a method behind it. The company has been named Best in KLAS for clinician digital workflow in 2026, and deployments described in customer announcements run for years at academic medical centers: University of Tennessee Medical Center since 2013, University of Washington Medicine including Harborview, University of Alabama at Birmingham, a Veterans Affairs division and King Faisal Specialist Hospital. KLAS measures customer satisfaction and product experience, not outcomes.

The figures published are customer quotes. On the Core Diabetes page an unnamed manager reports insulin errors down 20 percent and nurse error risk down 33 percent, and time to treatment for out of range glucose down 73 percent at one facility and 28 percent at another; a regional director is quoted on observation time falling from 26 to 24 hours; and a figure of up to 45 minutes saved per provider per day has been claimed. None names the site, period, method or denominator.

The newest product has no evidence of its own. Core Discharge cites one result, UW Medicine's discharge before noon rate rising from 12 to 17 percent in 2018. That predates the agentic module, which the page presents as amplifying those gains and which was announced in August 2026 with University of Tennessee Medical Center as a solution the two are building together. Nothing published measures the agents.

No peer reviewed publication or independent evaluation appears.

The data path is sketched and nothing is committed about it. The platform page describes the workspaces as embedded in PowerChart through SMART, FHIR and MPages, which keeps much of the handling where the record already lives, and then adds two layers that change the path: a semantic AI data platform that extends the record data model for large language models and reads clinical notes, and agent orchestration supplied by Vantiq, a separate company's platform. A mobile companion also reads and writes patient data on clinicians' devices.

No retention statement, de identification position, training position or safeguard description is published for any of it. The privacy statement and terms on the site cover website visitors only, so there is no document a buyer can read about patient data at all.

The language model layer is the open question. Clinical notes are the richest and most identifying text in the record, and the page says they are transformed for language models without saying where that processing happens, whether notes leave the customer's environment, what the partner platform retains, or whether any of it trains a model. Nor is it said what the mobile application caches on a device.

No statement of status and no privacy document that reaches the product. The footer links a privacy statement and terms of use, and both say in their opening lines that they govern the company's website: personal information from website users, newsletter and marketing contacts, and Google Analytics. Neither mentions protected health information, patients or business associate status, and no agreement, template or tier is published.

Given deployments inside large academic health systems and a Veterans Affairs division, agreements almost certainly exist in private; none is readable before contracting.

The product's current shape makes the missing instrument more consequential than it was. The platform now describes agents with read and write access to orders and documentation, a language model layer processing clinical notes, an external orchestration partner, and a mobile companion that reads and writes patient data on clinicians' devices. Each of those is a place a business associate agreement and its subcontractor terms would need to reach.

Nothing independent is published. The footer's privacy statement and terms cover the website, and no page names a SOC 2 report, HITRUST certification, ISO 27001, trust center, penetration testing statement or vulnerability disclosure policy.

One detail suggests more exists than is shown. The platform runs at a Veterans Affairs division, and federal deployment normally requires a product to clear a formal security authorization; no such assessment is mentioned anywhere on the site.

The product's current architecture raises the stakes. Agents with write access to orders and documentation, an external orchestration platform, a language model layer over clinical notes and a mobile companion carrying patient data are four surfaces a security review would want to scope, and nothing describes how any of them is controlled or tested.

Core Diabetes recommends insulin treatment to the ordering provider, and no regulatory position is stated for most of it.

The product page describes decision support for ordering subcutaneous insulin, decision support for nurses giving it, and an Advisor offering "automated treatment recommendations based on analysis of chart information". These ask a provider to start or change a treatment. No position is stated for them: no intended use, nothing on how they are regulated, and nothing showing the protocol rule behind a recommendation at the point of ordering. Chart data shown alongside is the input to a recommendation, not the reasoning that produced the dose.

The solutions page says Core Diabetes includes an FDA cleared algorithm. That clearance belongs to a partner: Pronia Medical Systems' GlucoCare, 510(k) K081388, cleared in August 2008 for calculating intravenous insulin doses to a protocol target range in hospitalized adults, excluding diabetic ketoacidosis. It does not reach the subcutaneous dosing or the Advisor, and the page does not say what it covers. The clearance names no version; the federal device identifier database lists GlucoCare 2.X. The home page names a 140B protocol for nurse IV titration, which Pronia describes as newer than the target ranges it has described as cleared.

The rest of the platform is operational: workspaces, task routing and discharge agents that forecast dates and extract equipment needs. The agents' write access to orders is the second regulatory question.

Nothing published on how the platform's judgments are governed or tested, and there are now two kinds of judgment to govern.

The older kind is editorial. Specialty workflow configuration decides what appears at the top of a nephrologist's patient list, which values surface on rounds and what counts as discharge readiness, and those clinical judgments are encoded by the vendor's knowledge engineers for every institution running the product. They age as practice, formularies and quality measures change, and a view that quietly stopped reflecting current practice keeps rendering with institutional authority. No versioning policy or review cadence for that content is published.

The newer kind is learned. A language model layer extracts discharge dates, equipment needs and care plan items from clinical notes, and agents apply clinical logic and act. Note writing varies by specialty, author and institution, so extraction quality will vary with it, and a missed equipment need or a wrong discharge date falls unevenly on the patients whose notes the model reads least well. No evaluation, monitoring, drift detection or performance by specialty or patient group is published, and the platform is described as ready to swap in newer language models without saying how a swap is tested.

Nothing published says what happens when the system is wrong, and the product now acts rather than displays. The platform page describes agents that autonomously execute multi step clinical and administrative workflows, with read and write access to orders, documentation and reference data, and a language model layer that turns clinical notes into expected discharge dates, equipment needs and care plan items. A wrong extracted discharge date moves a discharge; a missed equipment need delays one; an action written into an order by an agent becomes part of the record.

No accuracy figure, evaluation method, warranty, service level, indemnity or remediation commitment is published for any of it, and no correction route is described: nothing says how a clinician sees what an agent did, how an agent's action is reversed, or who answers for a harm traced to one. The only legal documents on the site are website terms, which provide the site's materials as is and disclaim reliance on them; no customer agreement is published.

The older workspaces were inspectable configuration controlled by the institution, which put authorship where clinical accountability sits. That property does not carry over to agents and a language model layer, and the published material does not say how it is replaced.

Depth of record system integration is not a feature of this product; it is the product. The platform page names the mechanism precisely: SMART, FHIR and MPages integration, embedded in Oracle Health's PowerChart, with bidirectional read and write across orders, documentation, reference data and workflows. The workspaces run inside the record rather than beside it, so clinicians do not change application, sign in twice or reconcile two views of the same patient, which is the failure that kills most overlay products. A mobile companion reads from and writes back to the record, and task synchronization, alert routing and secure messaging run across the care team inside the same environment. An MCP based interface layer is described for agent integration.

Production evidence is long and specific: deployments described in customer announcements at University of Tennessee Medical Center since 2013, University of Washington Medicine, University of Alabama at Birmingham, a Veterans Affairs division and King Faisal Specialist Hospital, and a Best in KLAS recognition in 2026 for clinician digital workflow.

The limit is breadth of record systems: the company positions the product around Oracle Health, and nothing describes what is available on another record system.

Where the workspaces run is clear from the architecture; where the newer layers run is not stated at all, and nothing is committed about either. The platform page says the product is embedded in PowerChart through SMART, FHIR and MPages, which means the specialty workspaces largely extend software the hospital already hosts and data handling stays in the customer's record instance. That follows from the design rather than from any residency commitment, and no hosting provider, region or residency statement is published.

Two layers now sit outside that environment by description: agent orchestration supplied by Vantiq, a separate platform, and a semantic AI data platform that prepares clinical notes for large language models. Nothing says where either runs, in which region, what they store, or how one customer's data is separated from another's. The mobile companion moves patient data onto devices as well.

So a buyer evaluating the product today is evaluating two deployment models under one name: an embedded one that is favorable, and an agentic one that is undescribed. No availability commitment is published for workflows clinicians now depend on.

The product menu is laid out in full and carries no prices. It lists the workflow products (Core Work Manager, Core Discharge, Core Mobile and Core Notify), a disease product (Core Diabetes) and the CORES platform underneath, which the platform page describes as more than fifty prebuilt workflow modules plus agentic automation and an interface layer. The positioning is explicitly around getting more value from an existing Oracle Health investment at a fraction of the cost, and the route in is a demo request.

What no page answers is the unit of charge, and on an embedded clinical workflow product the candidates diverge: per bed, per facility, per clinician user, per module, or a platform fee, with the agentic layer possibly priced separately because it rests on a partner's platform. Nothing states implementation fees, which for configuration inside a hospital's record are usually substantial, or the term.

Specialty breadth inside acute inpatient care is the product thesis. The premise is that one record system cannot suit every acute care team, so the platform ships more than fifty prebuilt workflow modules for specialties and care team roles: specialty patient lists, rounding preparation and handoff, discharge planning, task management, alert routing and secure messaging, with a diabetes module layered on and a mobile companion for work away from a workstation. The agentic layer is described across clinical and administrative workflows, with discharge planning the named use case.

The setting is inpatient acute care across physicians, nursing and multidisciplinary teams on Oracle Health, with deployments described at academic medical centers in the United States, a Veterans Affairs division and a large academic hospital abroad, and a 2026 KLAS recognition for clinician digital workflow. That is validation behind the core.

Coverage is deep within acute inpatient care and absent outside it, with no ambulatory, post acute or community presence, and the agentic workflows are new and carry no named deployment.