AvaSure
AvaSure makes virtual care technology for hospitals, and launched virtual sitting (TeleSitter) in 2009. A camera in the patient room feeds a hospital employed observer who can talk to the patient through two way audio and summon staff, and the company says one observer can cover up to 16 patients. The platform now spans continuous observation, virtual nursing, specialty consults, virtual visits and workforce optimization, on fixed ceiling, dual camera and mobile devices, including ligature resistant units for behavioral health.
Computer vision runs alongside the observer. AvaSure's falls and elopement models detect people, classify posture relative to the bed and apply rules to changes over time, raising early warning and higher severity alerts at an operating point tuned for each unit; the models are the company's own, trained partly on real hospital frames after anonymization, and run on Oracle Cloud Infrastructure. The platform holds Epic Toolbox designation for inpatient virtual care, launches from Epic Hyperspace, works with Epic Monitor and receives orders and ADT data from Epic, and routes urgent alarms to nurse call with acknowledgments synced back.
More than 1,200 hospitals use AvaSure, which is based in Belmont, Michigan. It publishes customer results and a library of studies; no price, business associate position or video retention default appears on its public pages, and its trust center renders only in a browser.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Computer vision does consequential work on a platform that worked before it. AvaSure's virtual sitting, launched in 2009, ran for years on cameras, two way audio and a remote observer; the company now describes AI as an extension of that platform, which more than 1,200 hospitals use. Its January 2026 post on building AI for patient safety describes falls and elopement models that detect people, classify posture relative to the bed and combine posture over time with rules to raise early warning and higher severity alerts, with pressure injury, infection related behaviors and staff duress planned next.
The models are what let one observer watch many rooms, so they carry the economics of the observation product; cameras, audio, nurse call routing, Epic integration and the clinical education program would still have value without them.
What the system does, where it stops and who acts are all published. Computer vision raises alerts; a hospital employed virtual safety attendant assesses each one on video, talks to the patient and summons staff, and the system itself never intervenes. The January 2026 AI blog post sets out the escalation logic: a move from lying to sitting on the edge of the bed can raise an early warning, standing unassisted a higher severity alert, with the operating point tuned per unit and models run silently in background mode before alerts are switched on. Urgent alarms route to nurse call over HL7, APIs or middleware, caregiver acknowledgments sync back, and attendants can mark alerts as helpful, spurious or missed, which feeds model review and rollback.
The company states one observer can cover up to 16 patients, and one customer reported an 11 second average response time with a 38 percent reduction in urgent alarm rates. The detection sensitivity behind the thresholds is not published, by use case or as alert volume per observer shift.
The approach and training data are described; version history is not visible to a buyer. The January 2026 AI blog post describes the falls and elopement models in three layers (person detection and count, posture and position relative to the bed, and rules applied to posture over short time windows), trained on staged scenes, synthetic rendered scenes and real hospital frames that pass through a patent pending anonymization step and human review before labeling. It says the operating point on the precision and recall curve is tuned per unit at onboarding, and that new model versions run side by side with existing ones in the field and are rolled back if they miss defined criteria.
No model architecture, version number or change notice to customers is published, and no precision, recall or F1 figure is given for any detection. Hardware is specified in detail (1080p cameras with optical zoom, infrared night vision, ceiling, dual camera and mobile units).
The model layer is AvaSure's own and the host is named. The January 2026 AI blog post says the falls and elopement system is built on Oracle Cloud Infrastructure and describes AvaSure's own computer vision models, trained on staged, synthetic and anonymized real world frames with a patent pending anonymization system. No outside model provider sits in that chain.
The rest of the party list is not published: no subprocessor list, no camera or hardware manufacturer, and no statement of which third party applications run on the open platform or what they receive. The trust center at trust.avasure.com, hosted on Vanta, may list subprocessors.
Named deployments with dated figures and a library of studies, short of independent multi site evaluation. AvaSure's clinical evidence library lists customer and research items, including a February 2026 account of $3.2 million in measurable impact at one health system, a UCHealth virtual health center profile (2024), and 2023 studies on nurse burnout with video monitoring and in person sitters, fall prevention evidence and workforce safety. Customer results reported by the company include a 38 percent reduction in urgent alarm rates with an 11 second average response time at one health system, and a 16 percent cut in emergency department holding time with a 31 percent improvement in discharge turnaround at an urban safety net hospital. More than 1,200 hospitals use the platform.
The headline claim, that hospitals reduce adverse events by 50 percent or more, is a platform wide claim with no single stated method, and most of the evidence is customer generated and vendor hosted.
Training use and de identification are stated; video retention is not. The January 2026 AI blog post says real world hospital frames enter training and evaluation only after a patent pending anonymization system removes or obscures identifying features and a human reviewer confirms anonymization is complete, and that no identifiable visual information is retained for that purpose; frames are selected around ambiguous or clinically relevant moments. The devices carry visible power, privacy and wireless indicators, privacy modes blank the feed during exams or personal care, and a federal hardware variant uses validated cryptography.
Whether live observation video is recorded by default, how long any footage is kept, and who can retrieve it are not published. The website privacy policy covers the website only, and the trust center may say more. Nothing is published on safety engineering beyond model monitoring.
No business associate position is stated on AvaSure's public pages. The privacy policy linked in the footer applies only to information collected by the website. The footer's Security link opens trust.avasure.com, a Vanta hosted trust center that may hold the HIPAA position and agreement terms. Every customer is a hospital, so the question is how, not whether, the product handles protected health information.
The question that matters most for this product is video: whether observation is live only or recorded, for how long, and who may retrieve footage of patients in bed, including behavioral health patients under ligature risk observation.
One outside check is visible. A United States Department of Veterans Affairs solicitation specifies a federal variant of AvaSure's hardware with cryptography validated under the federal standard for cryptographic modules, which is an independently tested credential for that component, not an audit of how AvaSure runs security. The footer's Security link opens trust.avasure.com, hosted on Vanta, which may list attestations such as SOC 2.
The surface is large: network connected cameras with microphones in patient rooms, tied into nurse call systems that are life safety infrastructure, with third party applications on the same platform.
The product is scoped as observation with a person deciding, and makes no device claim. Virtual observation, virtual nursing and consults connect staff to the room; the falls and elopement AI raises alerts that a virtual safety attendant assesses before anyone acts. The planned uses (pressure injury, infection related behaviors) move toward clinical decision support, and no position is published for them.
For behavioral health, the governing framework is hospital accreditation guidance on ligature risk and continuous observation for patients at risk of suicide; AvaSure sells ligature resistant ceiling units for that setting.
AvaSure describes how its models are validated and monitored and publishes no results by group. The January 2026 AI blog post says models run in background mode on live video before alerts are enabled, that the operating point is set per unit with clinical and operational leaders, that production performance is monitored by sampling outputs, and that new versions are compared with established baselines and rolled back when they miss defined criteria. It acknowledges that patients vary widely in demographics, mobility and behavior, and that lighting, occlusion and camera position affect detection, and it builds those variations into training.
No detection results by skin tone, body size, lighting (including infrared at night), mobility or unit type are published. Patients placed under observation are often confused, agitated or at high fall risk, so the deployed population is the atypical one.
Every alert is checked by a person, and nothing contractual stands behind the output. Alerts go to a hospital's virtual safety attendant, who sees the room on video, speaks to the patient and summons staff; urgent alarms route to nurse call and caregiver acknowledgments sync back to the platform, so response is auditable. The January 2026 AI blog post adds structured attendant feedback on whether an alert was helpful, spurious or missed, and rollback of model versions that miss defined criteria.
No precision, recall or missed event rate is published, and neither is any warranty, service level, indemnity or remediation commitment. A hospital moving from one to one sitters to one observer for many rooms is relying on performance it cannot see.
A named record system integration, verifiable in Epic's marketplace, with data moving both ways. AvaSure holds Epic Toolbox designation for the 2026 Inpatient Virtual Care category, listed in Epic Showroom. Its Epic integration page describes four patterns: virtual visits launched from within Epic Hyperspace, AvaSure devices used with Epic Monitor for virtual observation, patient access through MyChart Bedside, and orders and ADT messages sent from Epic into the AvaSure platform. Urgent alarms route to nurse call systems over HL7, APIs or middleware, and caregiver acknowledgments sync back into AvaSure. More than 1,200 hospitals run the platform.
Whether observation events and interventions are written to the chart as discrete entries is not published.
The system is hybrid, the AI host is named, and no location is committed. Cameras and monitor stations are installed in the hospital, and a federal solicitation documents an on premise server component; the January 2026 AI blog post says the falls and elopement AI is cloud based and built on Oracle Cloud Infrastructure. No region, residency commitment or tenancy model is published, and nothing says which video or derived data leave the facility for the cloud models.
No rate card is published, and a federal buyer has published the commercial structure. A United States Department of Veterans Affairs solicitation lists what a hospital buys: software licensed per hardware device on an annual term at a named tier, support tiered separately, with an analytics reporting tool, pre recorded announcements in 30 languages and clinical program materials included; hardware as a separate line, with a federal variant carrying validated cryptography; clinical services (policy planning, program development, go live support) as their own line; a separately priced monitor station; and installation within 180 days of order. AvaSure's own site offers a cost calculator for the benefit side.
No figure appears for the license, hardware, support or clinical services.
Broad acute care coverage, engineered by setting, with results behind part of it. AvaSure sells continuous observation, virtual nursing, specialty consults, virtual visits and workforce optimization across medical surgical, intensive care, emergency, behavioral health, long term acute care and pediatric settings, with ligature resistant ceiling units for behavioral health and mobile carts for the emergency department. It also sells observation staffing ratios and flow results: one customer reported emergency department holding time down 16 percent and discharge turnaround up 31 percent.
The company does not say how much of its installed base runs each application, results come mostly from individual customers, and the AI uses are newer than the platform.
Alternatives to AvaSure
Vendors in the same categories as AvaSure whose graded profiles look most alike, with a line on what each product does.
Similar is not the same, so check that each one does the job you need. No vendor pays for placement.
CLEW Medical, based in Netanya, Israel, with US operations in Boston, sells FDA cleared machine learning models that predict clinical deterioration in hospital critical care, delivered…
Artisight sells a smart hospital platform built around one ceiling or wall mounted device that combines dual 4K cameras, a microphone array, radar, indoor positioning and an NVIDIA Jetson…
Etiometry, based in Dedham, Massachusetts, sells a clinical intelligence platform for critical care that combines vendor neutral ICU data aggregation with four FDA cleared physiologic risk indices.
care.ai makes virtual care and ambient monitoring software for hospital inpatient units.
Caregility, based in Wall, New Jersey, runs a virtual care platform for hospitals and other care settings.
Sickbay, from Medical Informatics Corp (MIC) in Houston, is a vendor neutral platform that captures physiologic data and waveforms from bedside monitors, ventilators and other devices…
Ordered by how closely the graded profiles match across the fifteen axes, inside the same categories. See how each axis is graded in the methodology.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Not published; a federal solicitation discloses the structure as annual software licensed per hardware device at a named tier, with support, hardware, clinical services, monitor stations and installation carried as separate line items
|
Annual software license per hardware device, disclosed through federal procurement rather than published by the vendor. The unit is the device rather than the bed, the user or the monitored patient, which matters because cost then scales with rooms equipped rather than with patients observed, and the observation ratio of up to 16 patients per observer does not reduce it. Support is tiered separately from the software tier. Hardware, clinical services, monitor stations and installation are all distinct line items. Nothing indicates volume discounting or how the newer AI enabled applications are licensed relative to the base platform. | Not published on the pages read. The website privacy policy covers the website only; the trust center at trust.avasure.com renders only in a browser. Ask for the agreement template, the recording default and retention period for patient video, whether recording can be disabled at organization level, and how behavioral health observation footage is handled. | Charged separately and identified as such, which is itself the useful disclosure. A federal solicitation lists clinical services as a discrete line item covering policy and procedure planning, resource sharing and guidance, program development and best practices, and onsite go live support, alongside a separately priced monitor station and an installation period of 180 days from receipt of order. That establishes implementation as a distinct purchase rather than a bundled inclusion. The company also maintains a team of registered nurses running a clinical education program that accompanies the technology, which is consistent with a substantive services component. No figures are published for any of it, and nothing indicates whether clinical services are optional, mandatory in year one, or recurring. | Third Party Estimated |
No rate card is published, and the commercial structure is nonetheless visible in the open because a federal buyer published it, which places this materially above a quote only position.
A United States Department of Veterans Affairs solicitation enumerates the line items a hospital actually purchases. Software is licensed per hardware device on an annual term at a named tier, with a support level carried separately and inclusions listed: an analytics reporting tool, pre recorded announcements in 30 languages, clinical program materials and eLearning modules. Hardware is a separate line, with a federal variant carrying validated cryptography. Clinical services form their own line covering policy and procedure planning, program development, best practice guidance and onsite go live support. A large format monitor station is priced separately again. Installation is specified at 180 days from receipt of order.
That gives a buyer the unit, the term, the tiering and the unbundling without a single figure, and it identifies three separate negotiations rather than one: software tier, support tier and clinical services scope. It also reveals that clinical services are a distinct purchase rather than an included implementation, which is the kind of detail that surprises buyers late in a procurement.
The return side is unusually well served. Customer results carry dollar figures at named institutions, including $3.2 million of measurable impact at one health system and more than $1 million in a year at an academic medical center, and the cost of falls basis used in those calculations is attributed to a cited external source rather than asserted. A buyer can substitute their own fall cost and recompute rather than accepting the vendor's arithmetic.
What is entirely missing is price. No figure appears for the per device annual license, the hardware, either support tier or clinical services, and nothing indicates how cost behaves at scale across a large deployment, where per device licensing compounds directly with room count.
Ask for the per device annual license, what separates the support tiers, whether clinical services are mandatory in year one, and volume behavior above 100 devices.