Caregility

A clinical video communications platform that is genuinely adding intelligence, where the intelligence is not yet what the platform is.

The honest description of the core is connectivity. Secure two way audio and video into patient rooms, device fleet management across more than 30,000 endpoints, and over six million virtual sessions a year. The overwhelming majority of that volume is a clinician talking to a patient through a screen, which is transport and workflow rather than inference. The company's own earlier self description, a clinical collaboration and communications company, was accurate and the lineage still shows.

What has been added is real and specific rather than decorative. Edge based computer vision runs on the devices, audio sensing detects room duress, ambient listening and sensor capabilities exist, and an AI enhanced monitoring capability was added to the coordination application in 2026. A funding round in September 2025 was explicitly directed at computer vision, ambient listening and sensor technologies, and a named executive holds product, technology and artificial intelligence together, which is a structural commitment rather than a press release.

The comparison that clarifies the grade is the nearest competitor in this index. There, computer vision is what makes a single observer covering sixteen patients defensible, so the models buy the economics. Here the platform delivers value on video alone and the models improve it.

Graded C rather than lower because the intelligence is shipping, proprietary and running at the edge, and rather than higher because a buyer purchasing this today is buying a virtual care platform with AI capabilities, not an AI product. Ask what proportion of deployed rooms run the computer vision applications.

A sound human in the loop design with none of the measurement that would let a buyer evaluate it.

The model is right in shape. Virtual observation places a human observer between the system and any action, virtual nursing places a licensed remote nurse in the workflow, and the coordination application is explicitly framed around bringing bedside clinicians, remote caregivers and AI tools together rather than around automation. The company's own language describes augmented observation, which correctly positions the models as assisting a person rather than replacing one, and change management is named by its chief executive as the real difficulty, which is a more honest read of the deployment problem than most vendors offer.

What is missing is every number that would make the model assessable. No observer to patient ratio is published, so the supervision density being purchased is unknown. No response or notification time is published. No acknowledgement or closed loop mechanism is described, so whether a hospital can audit that alerts were received and acted on is unclear. No detection sensitivity or false alarm rate exists for the computer vision or the duress detection.

The comparison in this segment is direct and unflattering on this axis specifically. The nearest competitor publishes its observer ratio, routes alarms to nurse call, syncs caregiver acknowledgements back to close the loop, and has a customer reporting an eleven second average response. Those are the artefacts this axis exists to reward and none has an equivalent here.

Ask for the observer ratio, whether acknowledgements return to the platform, and the measured response time in a live observation programme.

Capabilities are named with useful architectural specificity, and the models themselves are described only by label.

The architecture disclosure is genuinely informative in two places. Computer vision is stated to run at the edge, which tells a buyer where inference happens and materially changes both the latency and the data movement picture. And connectivity options are enumerated down to 5G and satellite links, which is the sort of concrete detail that determines whether a rural site can deploy at all.

Device capability is described at a working level, including dual camera edge devices supporting concurrent workflows, so one unit can run ambient capture and an active consult at the same time.

Beyond that the model layer is opaque. Computer vision, audio sensing duress detection, ambient listening, sensor based technologies and AI enhanced monitoring are named as capabilities with no accompanying description of what each detects, how, or how well. No accuracy figure, no sensitivity, no false alarm rate, no operating point, no model card and no architecture description was located for any of them.

Duress detection is the least specified and the most consequential. Nothing states whether it classifies acoustic events, analyses speech content, or reads volume and tone, and those are very different systems with very different privacy and accuracy profiles.

One pre emptive note: further capability announcements cannot move this grade. Only published detection performance at a stated operating point will.

Ask what each model detects and at what sensitivity, and specifically what signal duress detection uses.

One ecosystem partner named publicly, and no chain disclosed behind it.

A dedicated pass located no cloud or infrastructure provider, no base or foundation model, no computer vision or speech framework, no sub processor register, no device manufacturer and no position on whether patient video or audio contributes to model development.

The partial credit is real and narrow. The platform is openly positioned as an ecosystem hosting third party clinical applications, and at least one such partner, a clinical documentation vendor, has been named publicly. That is one identified party processing on the platform, which is more than the nearest competitor discloses about its own open architecture, and it establishes that the ecosystem is real rather than aspirational.

It is not a register. A hospital cannot obtain a list of which third parties may access video, audio or device data, what each receives, where each processes it, or how they are vetted before admission. For an architecture whose stated value is that every new integration extends the room, the set of possible integrators is the supply chain, and it is undisclosed.

The edge computer vision runs on hardware whose manufacturer is unnamed, and the audio sensing capability implies a speech or acoustic processing component that is likewise unidentified.

The training question is unaddressed in both directions. Six million sessions a year of patient video and room audio is an extraordinary corpus, and nothing states whether any of it has been used to develop the models now being sold.

Ask for the sub processor register, the ecosystem vetting process and what partners receive, and whether session data trains models.

The largest deployment footprint in this segment and the thinnest published evidence relative to it, which is a combination worth stating plainly.

Scale is not in doubt and is independently plausible. More than 1,500 hospitals, 75 health systems, deployment across multiple continents, more than 30,000 connected devices and over six million virtual sessions a year. A third party analyst firm named the platform the leading virtual care platform outside record system vendors in three consecutive years, which is external assessment rather than self report, and listing in a major record vendor's validated integration programme for two consecutive years is a further independent check.

What is missing is measurement. No peer reviewed publication concerning this platform was located. Named customer references are testimonial in form, describing enthusiasm and adoption rather than reporting figures, and no quantified outcome for falls, length of stay, response time, readmission or cost was located at any named institution.

The contrast with the nearest competitor is instructive and belongs on the record. That vendor has a smaller footprint, roughly 1,200 hospitals against 1,500 here, and publishes a 40 month peer reviewed study, response times in seconds, alarm reduction percentages, throughput improvements and dollar figures at named hospitals. Deployment breadth and demonstrated outcome are different claims, and this index has already recorded one vendor whose 80 site footprint produced no outcome data at all.

Six million sessions a year is an enormous evidence base sitting unanalysed in the open.

Ask for any peer reviewed work, and for quantified outcomes from a named virtual nursing or observation programme.

Compliance posture is well stated and handling detail is not, with one capability raising a question nothing addresses.

What supports the grade is genuine. Compliance across three privacy regimes is claimed, certification against information security standards is asserted, and edge based computer vision means at least some inference occurs on the device rather than requiring a continuous stream to a central service. The platform is positioned as purpose built for clinical environments rather than a general video conferencing tool adapted to healthcare, which is a real architectural distinction when the alternative is consumer software handling patient encounters.

Handling specifics are absent. No encryption statement, no retention schedule, no access control model, no deletion process and no position on whether recorded sessions or captured video contribute to model development.

The retention question is the central one in this category and is unanswered here as it is for the nearest competitor. Whether patient room video is observed live and discarded or recorded and stored determines the storage, access logging and discovery obligations a hospital inherits.

Audio sensing deserves separate attention because it is a distinct exposure. Duress detection implies microphones active in patient rooms continuously rather than only during a session, and a hospital room contains clinical conversations, family discussions and disclosures made in the belief that nobody is listening. Whether the system processes acoustic features only or parses speech, whether anything is retained, and whether patients or staff are informed, are all unstated.

Ask what audio sensing captures and retains, whether video is recorded by default, and whether either trains models.

Multi jurisdiction compliance stated explicitly, with an independent assessor named for the health privacy position, which is more than any comparable vendor in this index provides.

Three regimes are claimed rather than one: United States health privacy law, European data protection law and Californian consumer privacy law. That combination is not decorative for a vendor operating across multiple continents, and stating it invites scrutiny a domestic only vendor avoids. A buyer in Europe has an answer to a question most vendors here cannot address at all.

The assessor naming is the part that distinguishes this record. Company material cites a named independent security firm as having verified the health privacy compliance position. Health privacy law has no certifying body, so this is an assessment rather than a certification and should be described that way, and it is still meaningfully stronger than the unattributed badge that constitutes the entire posture of several other vendors in this segment.

What remains absent is the contractual layer. No business associate agreement template, negotiation stance, execution requirement or subcontractor flow down position was located.

One question specific to this product is unaddressed and matters more than the template. Continuous audio sensing for room duress detection means microphones processing sound in patient rooms outside of any active video session, and nothing states what that processing captures, whether speech content is analysed or only acoustic characteristics, or whether anything is retained.

Ask for the agreement template, the scope of the assessor's verification, and precisely what audio sensing captures and retains.

More externally assessed credentials than anything else in this segment, stated without the qualifiers that would make them assessable.

The claims are substantive. Certification against information security standards and a service organisation controls report are both asserted, alongside compliance with United States health privacy law, European data protection law and Californian consumer privacy law. A named independent security firm is cited as having verified the health privacy position, and naming an assessor is a meaningful step beyond an unattributed badge.

The credential test costs the grade the level above. Neither claim carries the qualifier this index consistently requires. The information security standard is referenced by family rather than by number, and the controls report is referenced without its type, which is the distinction between controls that were designed and controls that were tested over a period. Those qualifiers are the whole information content of such a claim, and this index has recorded several vendors blurring exactly this distinction.

What is missing beyond that is the supporting apparatus. No trust centre exists as a standing page, no report availability or request process was located, no audit period or assessor is named for the information security work, and no penetration testing disclosure or vulnerability disclosure policy was found.

The surface warrants it. Network connected cameras and microphones permanently installed in more than 1,500 hospitals, running an open ecosystem that admits third party applications to the same device and data layer, is a large and shared attack surface.

Ask for the standard number and the report type, the audit period, and how ecosystem applications are vetted.

A defensible non device position, undeclared, with two capabilities drifting toward the boundary and a second regulatory regime unaddressed.

No clearance, grant or approval was located and none is claimed. The base position is sound: secure video communication with clinicians on both ends neither diagnoses nor treats, and virtual observation with a human assessing every event is workflow rather than a medical device function.

Two newer capabilities complicate it. Augmented observation with vitals monitoring is described in company material, and contactless or camera derived vital signs is the point at which several vendors in this index approach device regulation without stating a determination. Audio based duress detection is a different kind of claim again, inferring a safety state from sound, and while its output is a staff alert rather than a clinical finding, no determination is published for either.

The unaddressed regime is the more interesting gap and follows from the company's own compliance claims. A vendor asserting European data protection compliance is operating in a jurisdiction that now regulates artificial intelligence systems directly, with obligations that turn on risk classification and apply to deployed systems rather than only to devices. Nothing published addresses how the computer vision and audio sensing capabilities are classified under that regime, and a multi continent vendor claiming European compliance will be asked.

Ask for the written device determination covering vitals monitoring and duress detection, and the European artificial intelligence risk classification.

Governance framing and a governance structure exist, and no measurement sits behind either.

What is present is more than nothing and worth crediting precisely. The company uses responsible artificial intelligence as an explicit product framing rather than leaving the models unlabelled, and it has appointed an executive holding product, technology and artificial intelligence together. A named accountable owner for the models is a governance artefact, and several vendors in this index have neither.

What is absent is all of the substance. No fairness or bias testing, no performance broken down by any subgroup, no calibration data, no drift monitoring, no model documentation and no external audit was located.

The risk profile here is broader than for a camera only vendor because two modalities are in play. Computer vision reading patients in beds carries documented differential error by skin tone, worsened under the infrared conditions night observation requires, and by body habitus, bedding and positioning aids. Audio sensing carries its own well documented pattern, since acoustic and speech models degrade by accent, dialect, pitch and language.

That second one deserves particular scrutiny given the application. Duress detection infers threat or distress from sound, and a system that misreads raised voices, unfamiliar languages or particular vocal registers as aggression would direct security attention unevenly across patients and families. Nothing published addresses it.

Ask for detection performance by skin tone and under infrared, and for duress detection performance across languages, accents and vocal registers.

Nothing contractual is published, for a platform carrying observation, duress detection and clinical communication.

A dedicated pass located no service level agreement, no accuracy warranty, no uptime commitment, no indemnity and no remediation position for any application.

Uptime is the omission that stands out here rather than detection accuracy, and it distinguishes this record from the sensing vendors. This is communications infrastructure carrying more than six million clinical sessions a year across 1,500 hospitals, and when it is unavailable a virtual nurse cannot reach a patient, an observer cannot see a room and a consult cannot happen. Availability commitments are standard in enterprise communications contracts and none is published here, nor any historical availability figure.

The detection side carries the same gap as its competitors. No sensitivity or false alarm rate is published for computer vision or duress detection, so the residual risk of a missed event cannot be characterised by a hospital deciding how many rooms one observer covers.

The ecosystem model introduces an allocation question the nearest competitor does not have to answer as sharply. When a third party application running on this platform produces a wrong output, responsibility divides between the application vendor, the platform and the hospital, and nothing describes how.

One pre emptive note: further certifications or integrations cannot move this grade. Only a contractual commitment, or published availability and detection performance, will.

Ask for the availability commitment and historical uptime, what is warranted on detection, and how liability is allocated for ecosystem applications.

Independently verifiable marketplace listing, named modules, named integration paths and a named ecosystem, which is the most completely evidenced interoperability position in this index.

The listing is the part that cannot be self asserted. Caregility appears in the major record vendor's validated integration programme for inpatient virtual care, and in 2026 for the second consecutive year. Admission to that programme is granted by the record vendor rather than claimed by the participant, which converts an interoperability assertion into a third party fact and is exactly the check this axis should look for.

Beneath it the detail is specific. Named integration paths connect to the bedside television application, the monitoring module, the mobile clinical application and the main clinical desktop, through several distinct named integration mechanisms. That is four separate surfaces of one record system rather than a single connection, and each serves a different user in a different place.

Beyond the record system, integrations are described with nurse call systems, interactive patient consoles and digital bedside devices, which are the other systems that actually occupy a hospital room.

The ecosystem layer extends it further. Third party clinical applications run on the platform, with a clinical documentation vendor among those named publicly, so the integration story runs in both directions rather than only outward.

What is missing is thin by comparison: no public interface specification was located, and nothing states whether virtual encounters and observation events post to the record as discrete entries.

Ask for the interface specification and whether encounters write to the chart.

The topology is partly disclosed and the hosting behind it is not.

What can be established is useful. The platform is cloud delivered with cloud based device fleet management across more than 30,000 endpoints, while edge devices sit physically in patient rooms and run computer vision locally. That split matters: inference at the edge means video need not stream centrally to be analysed, which narrows the residency question for the most sensitive data even without a residency commitment. Connectivity is flexible down to cellular and satellite links.

What is absent is the centre. No cloud provider is named, no region is stated, no tenancy model is described and no explicit data residency commitment was located.

The international position makes this more consequential than a domestic omission would be. Deployment spans multiple continents and the company claims European data protection compliance, which carries requirements about where personal data is processed and how transfers are handled. Claiming that compliance implies regionalised processing exists; nothing published describes it, so a European buyer has an assertion rather than an architecture.

The ecosystem model adds a further unknown, since third party applications running on the platform may process data in infrastructure the vendor does not control and has not described.

Ask where the platform is hosted and in which regions, how European deployments are separated, what leaves the edge device, and where ecosystem partners process.

Cost is absent from every vendor surface. A dedicated pass located no pricing page, no unit of charge, no range, no tiering, no implementation fee position and no minimum commitment, with every route terminating in a demo request.

One third party analyst listing offers a figure in the region of 100 to 500 dollars and attaches no unit to it. Without knowing whether that is per device, per user, per month or per session, the number cannot be used for anything, and a buyer who anchored on it would be guessing. This record does not treat it as a disclosure.

The product structure makes the silence expensive. The platform spans virtual nursing, observation, rounding, consultation, operating room telehealth and hospital at home, across inpatient, ambulatory, post acute and home settings, on more than 30,000 devices. Those cannot plausibly share one rate, and nothing indicates whether charging follows the device, the room, the session, the concurrent user or the enterprise. Device fleet management as a described capability implies a device denominated model without confirming it.

No return proxy is published either. Staffing relief and capacity extension are the commercial argument and no cost per session, cost per device or avoided labour figure supports it.

The nearest competitor in this segment has its full commercial structure visible through a federal procurement record, which shows that this kind of disclosure is achievable in this category rather than impossible.

Ask for the unit of charge, how the applications price relative to the platform, the device fleet cost at scale, and the minimum term.

The widest span of care settings in this segment, with the distribution across it unstated.

Coverage runs the continuum rather than a slice of it: inpatient acute and intensive care, ambulatory, post acute, and the patient's home including hospital at home programmes. Application coverage is correspondingly wide, spanning virtual nursing, virtual observation and e sitting, virtual rounding, specialty and pediatric consultation, operating room telehealth and family engagement in care planning. Users are named across roles, from sitters through nurses to physicians and specialists, which is a meaningful distinction because those roles need different interfaces.

Geographic reach is genuine and evidenced by regulatory posture rather than assertion, since compliance is claimed against European and Californian regimes alongside United States health privacy law, which a purely domestic vendor would have no reason to state.

Connectivity options extend the reach further in a way specific to this vendor. Support for 5G and satellite links addresses rural and mobile deployment, which is the constraint that usually decides whether virtual care reaches critical access hospitals at all.

What is missing is the same gap the nearest competitor has. More than 1,500 hospitals are claimed and nothing states how many run observation only, how many have progressed to virtual nursing, or what proportion use the newer AI applications rather than video alone. With six million sessions a year, the mix is knowable internally and unpublished.

Ask for the session mix by application, and how many sites run beyond a single unit.