Electronic Caregiver

The avatar is the product's reason for existing, inside a business that is substantially services, hardware and reimbursement plumbing. Addison Care is a three dimensional animated caregiver on a tablet that holds spoken conversations with an older adult, prompts medications, runs cognitive engagement exercises and captures vitals from connected devices. The company's November 2025 post on its own site states that the platform is "integrated with OpenAI's ChatGPT", and its launch announcements describe an AI powered virtual care ecosystem, so a learned component is present: a language model bought in from OpenAI, alongside the speech recognition a spoken conversation needs. The company's thesis is that a face and a dialogue drive adherence where a portal does not, which makes the conversational model the mechanism of the avatar rather than decoration.

What surrounds it is large and mostly not inference. The company's home page sells remote patient monitoring, emergency response and a round the clock physician consult without mentioning the avatar at all. Connected devices, tablets and national fulfillment are hardware and logistics; emergency response, nursing support and physician consults are staffed services; billing and payer approvals are commercial infrastructure. The chief executive has described that operational depth, not the AI alone, as the barrier to entry.

The model is the engine of the avatar, a core module, on a platform that carries other value.

Human services stand behind the avatar, and no boundary on what the avatar may say is published. The home page lists emergency response and a physician consult around the clock, and the company's announcements describe telecare integration with human telehealth support; clinicians manage patients through a provider portal. For an emergency response product that staffing is the safety net.

What is undefined is the conversational boundary and the handoff. The avatar is integrated with OpenAI's ChatGPT and holds daily open ended conversations with older adults about their health, medications and wellbeing. Nothing published describes what it is permitted to say, what happens when a user asks a medical question or describes symptoms or distress, which statements trigger a handoff to the telecare team, how quickly a person picks up, or what testing bounds the model. The "24/7" physician consult is a claim of presence rather than correctness, and it sets the scope the oversight is judged against: an always available channel that has taken on the user in crisis.

The population makes the gap matter more. For people with cognitive impairment or developmental disabilities an authoritative animated character is unusually persuasive, and the line between companionship and advice is easy to cross. No escalation timing, alert volume or emergency response time is published.

Oversight named as services, with no mechanism published for the avatar channel.

Components are named, including the language model supplier, and none carries a performance figure. The company's November 2025 post states the platform is "integrated with OpenAI's ChatGPT" and runs on AWS. Company material lists the components: a three dimensional avatar, conversational speech, machine learning, visual sensing, biomechanical analytics developed with a state university's health and aging laboratories, medication management, record system integration and connected hardware, with large language model integration, data architecture and longitudinal engagement analytics among its technical disciplines.

What is missing is version and measurement. No model version, update practice or change notice is published for the ChatGPT integration, and nothing says which parts of the avatar's conversation the model generates. No speech recognition accuracy, adherence effect attributable to the avatar, emergency detection performance, biomechanical validation or description of what the visual sensing computes is published. The biomechanical analytics are the least specified: gait and movement assessment developed with a university group implies a measurement claim with no published data behind it.

The model provider and the host are both named, and no register sits behind them. The company's November 2025 post states the platform is integrated with OpenAI's ChatGPT and runs on AWS, where it is listed on AWS Marketplace under a co-sell agreement with Amazon. The company also says each activation orchestrates dozens of services and vendors spanning connected devices, cloud services, billing engines, record connections and human support, which states that the chain is long without listing it.

No subprocessor list is published, no connected device manufacturer is named although the kit depends on third party instruments, no speech recognition component is identified, and nothing states what conversation data flows to OpenAI or under what retention and training terms. The Notice of Privacy Practices requires business associates to safeguard information without naming them. Training provenance is unaddressed: biomechanical analytics developed with a university research group imply a research dataset, and no consent basis is stated.

Named early customers and payer routes, and no outcome figure with a method or a source on the company's own pages. The company's 30 June 2025 launch announcement names Capstone Clinic, described as one of the largest primary care physician groups in Alaska, as the first customer, with four clinical locations live in New Mexico and Alaska and rollouts planned for Colorado and Arkansas, billing under remote patient monitoring and chronic care management codes; Pinnacle Integrated Medicine is named as the athenaOne integration partner, and the company says more than 100 clinical organizations use its broader services. It cites increased engagement and adherence improvements without figures, and a fall prediction study with UMass, the National Institute on Aging and NIH without results. The November 2025 post says the platform is deployed across private pay, Medicare, Medicaid and commercial programs, and a New Mexico developmental disability waiver program is a named route.

Outcome figures circulating about the company, including a 47 percent reduction in readmissions, an HbA1c reduction in type 2 diabetes and a 70 percent expansion of inpatient capacity through hospital at home, appear in directory listings rather than on the company's own pages, with no study, sample, site or comparison period attached. More than 100 pilot deployments are claimed, a count of trials begun rather than programs sustained. The home page, the November 2025 post and the launch announcement cite no peer reviewed publication.

Named customers without measured outcomes.

A HIPAA compliance claim, a named cloud provider and a covered entity privacy notice, with nothing on what the avatar retains or sends to its model provider. The company's November 2025 post describes a HIPAA compliant architecture on AWS. Its Notice of Privacy Practices (Electronic Caregiver, Inc., updated 29 July 2024) lists the health information it holds (health conditions, medications, allergies, vital sign data, emergency call activity and telehealth encounters), limits research use to de-identified information under a special approval process, and says records of care are retained without stating a period. The terms (modified 1 May 2025) add Google Analytics and audience data on the website and state that SMS consent and mobile numbers are never shared for marketing.

What is collected raises the stakes. The device sits in a living room with a screen, a microphone and visual sensing, engaging an older adult in daily conversation, cognitive exercises and medication prompts while capturing vitals; cognitive engagement data in particular tracks decline. No retention schedule, deletion process or description of what the visual sensing keeps is published.

The language model integration is the sharpest open question. The company states the platform is integrated with OpenAI's ChatGPT, and nothing says what conversation content is sent to OpenAI, what is retained there, or whether it is used for training. Safety engineering, meaning guardrails on the model's health statements and how a harmful answer is handled, is not published. Consent needs attention too, since people with cognitive impairment and developmental disabilities are served.

The company states its position as a covered entity and publishes the notice that goes with it; its position toward provider customers is not stated. The footer links a Notice of Privacy Practices (Electronic Caregiver, Inc., updated 29 July 2024) covering the health information the company holds from the care it provides: health conditions, medications, allergies, vital sign data, emergency call activity and telehealth encounters. It sets out uses for treatment, payment and operations, limits research to de-identified information under an approval process, requires business associates to safeguard information, and gives the patient rights to request amendment and to complain to the company's chief clinical officer or to the Secretary of Health and Human Services. The company's November 2025 post also describes the architecture as HIPAA compliant.

What is not published is the agreement layer. When a primary care group bills remote monitoring through the platform, the company handles that group's patients' data as its business associate, and no business associate agreement, its tier or its subcontractor terms are published. That matters because each activation is described as orchestrating dozens of services, from connected devices and cloud services to billing engines and record connections, and the platform is integrated with OpenAI's ChatGPT, so conversation content may leave the company's environment. A consumer buying privately may sit outside HIPAA entirely, and nothing distinguishes that case.

A stated covered entity position backed by its notice, with the business associate side unpublished.

The credentials published belong to the host, not to the company. The company's November 2025 post describes an "AWS-certified, enterprise-grade architecture", AWS Well-Architected certification, graduation from the AWS Partner Accelerate program and a submission for AWS's healthcare competency. Those are assurances about the underlying infrastructure and the company's standing with Amazon; they are not an audit of this company's own controls, staff, development practice or handling of patient data. No SOC 2 report, ISO/IEC 27001 certificate, HITRUST certification, trust center, penetration testing statement or vulnerability disclosure policy appears on the home page, the November 2025 post, the Notice of Privacy Practices or the terms.

The surface is substantial and physical. Tablets with cameras and microphones sit in private homes, connected instruments pair to them, devices are fulfilled and retrieved nationally, and a staffed telecare operation holds access to patient records. Device authentication, firmware update practice, and what happens to a returned device that has recorded someone's daily life at home are unaddressed, and the ChatGPT integration adds an outbound path with no described controls.

No device claim appears anywhere, and the product is framed accordingly. The home page, the November 2025 post and the terms make no FDA, cleared or medical device claim, and the framing throughout is engagement, adherence support, monitoring, companionship and emergency response rather than diagnosis; the company does not describe the avatar as detecting or diagnosing a condition.

Three regimes sit around the product without being addressed. Remote patient monitoring billing generally requires physiological data collected by a device meeting the regulatory definition, so the connected instruments in the kit carry regulatory status of their own, and nothing published names those devices or what each is cleared to measure; a provider billing under those codes relies on that chain and cannot inspect it. Remote support technology under a state developmental disability waiver substitutes technology for staffed supervision of a vulnerable population and carries state oversight, consent and safeguarding requirements distinct from device law. And a general purpose language model conversing about health and medications can drift toward statements that would amount to clinical advice, with no published position on where that boundary sits.

Nothing is published on how the avatar's models are tested or governed. The home page, the November 2025 post on the company's site, the Notice of Privacy Practices and the terms carry no fairness testing, subgroup performance, calibration data, model evaluation or governance framework.

The exposures are broad because the product engages through several channels at once. Speech recognition degrades with accent, dialect, hearing impairment, dysarthria and the slower or less fluent speech common after stroke or with dementia, and each is overrepresented in the population this device is built for. A virtual caregiver that understands some older adults less well delivers less care to them, and the failure is quiet because the user simply stops engaging. Camera based analysis has documented differences in performance across skin tones and lighting, and home lighting is uncontrolled.

The language model layer adds a third surface. The company states the platform is integrated with OpenAI's ChatGPT, and no guardrails, evaluation or monitoring are published for conversations with people who may not be placed to recognize a wrong or invented answer.

People stand behind the monitoring side of the product, and nothing stands behind what the avatar says. The company runs its own telecare operation: its home page lists emergency response and a physician consult around the clock, and its announcements describe human telehealth support behind the avatar, so an emergency call or an out of range reading reaches a person who acts on it. The company is a covered entity, and its Notice of Privacy Practices (updated 29 July 2024) gives a patient the right to request in writing that the company amend its records of the care it provided, with a written answer if it declines. That route runs to the company directly, and it covers records of care (vitals, emergency call activity, telehealth encounters) rather than the avatar's spoken statements.

No service level, response time commitment, warranty, indemnity or remediation position is published. The terms (modified 1 May 2025) cover the website and the ecgSMART application, provide content as is, disclaim all warranties and exclude liability for damages from use of the website. For a product carrying emergency response for older adults and people with developmental disabilities who often live alone, no detection rate, response time or availability figure is published. An avatar integrated with a general purpose language model, speaking daily with a cognitively impaired person about medications, can produce a wrong statement the user acts on, and nothing describes what is warranted or disclaimed about it.

Staffed review of emergencies and readings, with nothing attached to the output.

One record system is named with the data it receives, and nothing verifies the direction beyond the company's announcement. The company's 6 May 2025 announcement describes Addison Care as integrated into athenahealth's athenaOne record system: clinical alerts generated by the avatar platform are delivered into athenaOne, billing events trigger automated claims, and vitals, wellness check data and patient interaction records support remote patient monitoring and chronic care management programs. Its 30 June 2025 launch announcement names Pinnacle Integrated Medicine as the athenaOne integration partner. The November 2025 post on the company's site refers to EHR connections among the dozens of services each activation orchestrates, and the provider portal is described as integrating into record systems.

No other record system, interface standard or marketplace listing is named, and whether captured vitals and avatar conversations post to the athenaOne chart as discrete entries is not stated. That matters financially as well as clinically: reimbursement under remote monitoring and chronic care management codes depends on documented monitoring time and data, so what the integration writes decides whether the billing route works at scale. Connected instruments from other manufacturers feed the platform, and none is named.

A named record system receiving alerts and billing data.

The cloud provider is named and nothing else about where data rests is described. The company's November 2025 post describes an AWS architecture it calls AWS Well-Architected, with an AWS Marketplace listing and a co-sell agreement with Amazon. The physical deployment is described operationally: national device fulfillment and device lifecycle management are company functions, and the tablet and connected instruments sit in the patient's home.

No region, residency commitment, tenancy or segregation model, or continuity and recovery position is published, and naming a host is a description of infrastructure rather than a commitment of location.

Continuity matters because of what the product replaces. Where emergency response and daily medication prompting stand in for a staffed visit or a monitored pendant, an outage removes a safety arrangement rather than a convenience, and no availability figure or fallback is published. The device depends on a home internet connection the company does not control, and rural deployment in places like Alaska is part of the pitch where connectivity is least reliable; whether the platform works offline, buffers locally or fails silently is unstated.

A single hosted platform on AWS with its location implied rather than committed.

The payment route is stated clearly and the company's own price is not. Addison Care ships to Medicare patients under remote patient monitoring and chronic care management codes, which are established codes with published national payment amounts, so a provider can look up what it will be paid; those amounts are what Medicare pays the provider, not what the provider pays Electronic Caregiver. The company's November 2025 post names private pay, Medicare, Medicaid and commercial programs, and a New Mexico developmental disability waiver for remote support technology is another route. The company also states it is listed on AWS Marketplace.

That tells a clinic the service is billable through mechanisms already in its revenue cycle, which is the question that decides adoption for a primary care group.

The company's own economics are absent. No device cost, subscription rate, per enrolled patient fee, telecare service charge, share of reimbursement or contract term is published, and the offering bundles a tablet and connected instruments, national fulfillment, the software platform and a staffed telecare operation, each of which would ordinarily carry its own cost. A provider knows the revenue side and cannot compute the margin. The private pay price is also unpublished, for the buyer with no reimbursement to offset it.

The posture is discoverable and no figure is published.

Wide across payer types and living situations, with named programs behind part of it. The company's November 2025 post says the platform is deployed across private pay, Medicare, Medicaid and commercial programs, and a New Mexico developmental disability waiver program for remote support technology is a named route; the June 2025 Medicare launch began with four clinical locations and an Alaskan primary care group serving rural patients. Each payer route is a different buyer with its own eligibility and documentation rules.

Settings follow the person rather than an institution: private homes, rural communities, home care agency clients through a franchise partnership offering in person visits with continuous virtual care, hospital at home programs, and supported living for people with developmental disabilities. Remote support technology in that last setting substitutes for staffed supervision, which carries its own consent and autonomy questions.

Clinical coverage is broad and shallow by design, spanning chronic condition management, diabetes, behavioral health, medication adherence, cognitive engagement and emergency response, with the avatar as a general engagement layer rather than a condition specific tool. Nothing validates performance in any of these settings: no active patient count, split across payer routes, or measure of which settings carry sustained volume beyond pilots is published.

Named settings, with deployments behind the Medicare and waiver routes.