Octagos Health

Models drive the platform's first read, and the service around them carries value of its own. Atlas AI reads incoming transmissions from implanted cardiac devices and ambulatory monitors, links device alerts with record data, generates structured reports and archives what it judges non actionable; the company says more than half of transmissions are archived so that only actionable alerts reach the clinic. Ask Atlas adds a natural language layer for querying a clinic's own monitoring data.

Around the models sit specialists certified by the International Board of Heart Rhythm Examiners who validate every transmission, integrations with more than 20 record systems, billing support and data migration, so a clinic buying the service without the models would still be buying a reading service. Neither the triage nor Ask Atlas exists without the models.

The oversight structure is described and the forwarding criteria are not. The company's Two Brain Approach pairs an AI first read with validation of every transmission by specialists certified by the International Board of Heart Rhythm Examiners, and the resulting reports go to the clinic's own clinicians for decisions. The company says more than half of transmissions are archived so that only actionable, billable alerts reach the clinic.

What is not published is the threshold: the criteria that make a transmission actionable, and what happens when a specialist disagrees with the model, including whether overrides are recorded and fed back.

Function and results are described and nothing under the hood is identified. Atlas AI is named and its role is stated: an initial read of device transmissions that links alerts with record data, produces structured reports and archives non actionable transmissions, with results from a peer reviewed study (99.1 percent sensitivity; 22.8 percent of transmissions forwarded against 38.5 percent for technicians).

Ask Atlas is described with worked examples of plain language questions over a clinic's monitoring data, such as which defibrillator patients have an ejection fraction below 30 percent. No architecture, model type, training data, versioning or update practice is published for either, and nothing says whether Ask Atlas runs on a model built in house or supplied by another company.

The host is named and the model layer is not. The company says the platform is built on Microsoft Azure, and its privacy policy lists categories of recipients (record system providers, healthcare organizations, analytics and service providers, affiliates) without naming them.

Nothing says whether Atlas AI is built in house or licensed, and Ask Atlas, a natural language interface over a clinic's monitoring data, is offered without naming the language model or the provider that receives clinic queries. No subprocessor list is published. The device side is named: Abbott, Medtronic, Boston Scientific, BIOTRONIK and ambulatory monitor makers supply the data.

Published research with a large multi site population, and named customers. The company cites a peer reviewed study in JACC: Advances (2025), Role of Artificial Intelligence in Reducing Data Deluge From Cardiac Implantable Electronic Devices, covering 690,673 transmissions from 74,217 devices at 78 US clinics: the AI forwarded 22.8 percent of transmissions against 38.5 percent for device technicians, with 99.1 percent sensitivity.

The study's design, including its reference standard and whether it was prospective, is set out in the paper rather than on the company's pages. Named customers with attributed quotes include the University of Pennsylvania, Chicago Cardiology Institute, Cardiology Consultants of Philadelphia and Atlanta Heart Specialists.

Other home page figures, including more than 300 customers, more than a billion transmission pages processed, a 300 percent average increase in return on investment and 94 percent billing efficiency in 90 days, are stated without method.

General assurances are published and the questions the models raise are not answered. The home page states HIPAA compliance and SOC 2 Type 2 and SOC 3 reports and says the platform is built on Microsoft Azure; the privacy policy (Octagos Inc., effective 17 December 2024) covers the company's websites, excludes protected health information, and mentions anonymized, aggregate analysis without saying what data it covers.

Nothing states whether customer transmissions train Atlas AI, whether learning is pooled across clinics or confined to one, how long transmissions and reports are retained, or what de-identification applies to data used in the company's research. Ask Atlas sends clinic questions over patient data through a language capability whose provider and retention terms are not named. Safety engineering for the natural language layer, such as how a wrong answer is caught, is not published.

HIPAA compliance is stated and no document behind it is published. The home page states HIPAA compliance alongside SOC 2 Type 2 and SOC 3; the privacy policy (Octagos Inc., effective 17 December 2024) covers the company's websites, says it does not apply to protected health information, and refers to a separate HIPAA policy that the site footer does not link. No business associate agreement is offered or described. The service widens the data flows beyond software: vendor employed specialists review every patient transmission, and Ask Atlas processes clinic queries over patient data.

A recognized attestation is named in the company's own material without the report. The home page lists SOC 2 Type 2 and SOC 3, and the privacy policy (effective 17 December 2024) states the company is SOC 2 Type 2 certified and describes encrypted transmission, access restrictions and breach notification procedures.

A SOC 3 report is a public summary by design, but none is linked, and there is no trust portal, penetration testing statement, vulnerability disclosure policy or subprocessor list. The service model widens the scope that matters: vendor employed specialists access patient transmissions routinely, so the attestation should reach the review operation as well as the platform.

The software triages device transmissions, yet the company makes no device claim and states no regulatory position. Atlas AI performs an initial read that decides which transmissions are forwarded as actionable and which are archived, and the company presents the platform as software for device management and clinical decision support without an intended use statement or regulatory analysis.

Specialists certified by the International Board of Heart Rhythm Examiners validate every transmission, so the basis of the model's call is reviewed by a credentialed person before a report reaches the clinic, which is the main argument that the software supports a determination rather than making one. The company's published regulatory attention is on reimbursement: it publishes guidance on the device monitoring procedure codes and the 2026 changes.

A peer reviewed evaluation of the model exists and only aggregate results are published. The company cites a study in JACC: Advances (2025) covering 690,673 transmissions from 74,217 devices at 78 US clinics, which compares the model's first read with device technicians, so the model's behavior has been through peer review rather than only described.

What the company publishes is the aggregate: no results by device type, manufacturer, age, sex or clinic, no model documentation, and no description of the training population. The exposure is specific: transmission formats and alert logic differ by manufacturer and device type, and patients who transmit irregularly are disproportionately older and rural, so an aggregate sensitivity can hide weaker performance in one group.

The company publishes the model's error rate with the study behind it, and nothing stands behind the output. The home page reports 99.1 percent sensitivity from a peer reviewed study of 690,673 transmissions from 74,217 devices at 78 US clinics, in which the AI forwarded 22.8 percent of transmissions against 38.5 percent for device technicians. That is a published error rate with its method and denominator, a figure a buyer can test in a pilot and hold the company to in negotiation.

Specialists certified by the International Board of Heart Rhythm Examiners validate every transmission, which puts a credentialed reviewer between the model and the clinic. No accuracy commitment, indemnity, service level for the specialist review, or remedy for a missed actionable transmission is published.

Named integrations with major record systems, a stated mechanism, and production scale behind them. The home page names Epic through a native SMART on FHIR application with single sign on, Oracle Health, athenahealth, NextGen, eClinicalWorks, MEDITECH, Greenway, AdvancedMD and Veradigm among others, and claims more than 200 integrations across more than 20 record systems, averaging 21 days to go live.

The company describes the integrations as bidirectional, with device alerts linked to record data in its reports and reports returned to the chart, and named customers confirm integrations delivered on schedule and at no added cost. Device data arrives from Abbott, Medtronic, Boston Scientific and BIOTRONIK implants and from ambulatory monitor makers.

A hosted platform on a named cloud, with the location of patient data implied rather than committed. The company says the platform is built on Microsoft Azure, and its specialists are described as US based. The privacy policy (effective 17 December 2024) says its servers operate in the United States and that data may be processed outside a user's state or country, but it covers the company's websites and excludes protected health information, so it commits no location for patient data. No region, tenancy model, deployment option, subprocessor list, availability commitment or exit terms are published, and Ask Atlas is a further processing path with no stated location.

How the service is sold is published, and no price is. The pricing page lists four packages: Pro, the platform; Premium, adding dedicated clinical readers and patient engagement services; Flex, with those services on demand; and Custom. Every package includes Atlas AI, automated alert settings, redundant transmission archiving, bidirectional record system integration at no cost, data migration, onboarding and a customer success contact. Three billing options are described: a billable transmissions plan whose charge adjusts to the clinic's billable monthly transmissions, a fixed monthly plan based on estimated services, and a custom plan. The home page adds that integration averages 21 days to go live, that there are no fees, and guidance on the reimbursement codes clinics bill, which describes what a clinic recovers from payers rather than what it pays Octagos.

No figure is attached to any package or to the billable transmission unit, and no contract term, minimum or setup charge is published, so a clinic can see what drives the cost but cannot size it.

The unit of charge and package structure are discoverable, with no figure.

Coverage is named broadly, with validation behind the implanted device part. The company lists pacemakers, defibrillators, resynchronization devices, implantable loop recorders, heart failure monitors, Holter and extended Holter, mobile cardiac telemetry and select consumer wearables, names Abbott, Medtronic, Boston Scientific and BIOTRONIK among the manufacturers, and serves device clinics, cardiology groups and health systems.

The peer reviewed study behind its performance figures covers implanted device transmissions at 78 US clinics; nothing published validates the model's first read on ambulatory monitors or consumer wearables.