Eleos Health

The AI is the product. Eleos drafts notes from session audio or a short summary, scans notes for documentation errors, and surfaces clinical insights from the session, all from its customized natural language processing engine, which Eleos says was developed by AWS and adapted for behavioral health using its own data set of real treatment sessions. There is no record system or services layer underneath; it runs inside the provider's EHR. Remove the models and nothing is left to sell.

The oversight structure is described. Eleos frames its product as augmented rather than autonomous intelligence: it drafts note content, which Eleos says covers about 80 percent of a typical progress note, and the provider reviews, tweaks and submits. Its Compliance product checks notes for documentation errors before claim submission, and Eleos states that it is not there to replace a provider's judgment about clinical care. What is missing is what happens after an output is wrong and where the system stops: no edit rate, threshold or escalation path is published for notes or for identified interventions.

The approach and the supplier are both named: a natural language processing engine developed by AWS, customized for behavioral health and trained on Eleos's own data set of real behavioral health sessions, which drafts note content and identifies evidence based interventions from the session.

The company states a transcription accuracy of around 93 to 94 percent, which it calls its word error rate, without a method or test set, and says its clinical team reviews the data for ongoing improvement. Studies listed on its science page describe parts of the method. No model version, update discipline or error profile by client population is published.

AWS is both the model supplier and the host. Eleos's AI FAQ says its natural language processing engine was developed by AWS and customized for behavioral health, and all customer data sits on HIPAA eligible AWS servers in the continental United States. Residency is stated as a commitment: Eleos supports deployment to designated regions and guarantees that data will not move outside the region originally designated.

What persists is stated too: audio is deleted within one to two minutes, transcripts shortly after, and AI-generated suggestions and engagement data are kept to refine the models. No sub-processor list is published; the privacy policy describes categories of service provider only.

Eleos publishes research rather than only asserting outcomes. Its strongest item is a randomized clinical trial, "Effects of an Artificial Intelligence Platform for Behavioral Interventions on Depression and Anxiety Symptoms", published in the Journal of Medical Internet Research and run at Ozark Center in Missouri, which compared the platform with treatment as usual in 47 adults in outpatient therapy for depression or anxiety and reported better attendance and greater symptom improvement on the PHQ-9 and GAD-7.

The design is prospective and randomized; it is also small and single site, so it is preliminary evidence, and Eleos's claim to be proven to deliver significantly higher results than leading alternatives reaches further than one site can carry. Its science page lists further peer reviewed papers and studies with academic collaborators, including a machine learning model predicting therapy homework assignment.

Retention and training use are stated specifically, and the training answer matters most. Session audio is processed in real time and deleted within one to two minutes after the session ends, and transcripts are deleted shortly after. Eleos keeps the AI-generated note suggestions and user engagement data and states that it retains them for refining its models, together with minimal session identifying information: clinician ID, session date and time, client unique ID and name.

No de-identification step is described for the retained suggestions, no retention period is stated for them, and nothing says a customer can decline their use for model refinement. Nothing is published on safety engineering: no guardrails, hallucination checks, or handling of a note containing something never said.

Eleos states that it is HIPAA compliant and processes customer information on behalf of the provider organizations it serves, and its privacy policy says customers can ask Eleos to execute a business associate agreement. Its security page states that the agreement can be found there alongside the privacy policy, but the agreement's terms, including whether it binds subcontractors and applies to every customer, are not reproduced on the pages linked from the main site.

Eleos also states FERPA compliance, reflecting school based work, which puts some deployments under a second federal confidentiality regime. Its substance use disorder page states no position on 42 CFR Part 2.

By its own account, Eleos has passed an external SOC 2 Type II plus HITRUST audit, run through the combined AICPA and HITRUST approach, and it lists ISO 27001, ISO 27799 and NIST CSF alongside it, with an independent agency contracted for annual penetration testing and continuous vulnerability monitoring. The certifications are presented as its own.

Its trust center at trust.eleos.health is where the reports sit; the main site gives neither their periods nor whether they can be downloaded without a request. Its FAQ calls the SOC 2 a certification and lists a Deloitte certification, loose wording for an attestation and an auditor, and which firm issued each report is not stated. Stated controls include TLS 1.2 or higher in transit, AES-256 at rest on AWS, host based intrusion detection, endpoint detection and response, SSO and MFA, and least privilege access.

Eleos makes no medical device claim and states its own scope. Its AI FAQ says Eleos is not currently used for diagnostic purposes and is not a replacement for clinicians, would likely be treated as administrative or supplementary support under state AI laws, and does not replace providers' judgment about clinical care.

Its marketed capabilities fit that scope: drafted note content, documentation compliance checks, and clinical insights such as interventions identified from the session, delivered to the clinician. The rules that govern it are payer documentation and audit requirements, which its Compliance product addresses before claim submission; state Medicaid oversight of its community behavioral health buyers; state AI laws, which its FAQ reviews state by state while advising client notice and written consent; FERPA in school settings; and 42 CFR Part 2 for substance use disorder records, on which Eleos states no position.

ISO/IEC 42001:2023 certification covers Eleos's AI management system. It was announced in November 2025 after Mastermind, an accredited certification body, audited the company's technology infrastructure and governance program, and Eleos states that it has also earned Drummond's predictive decision support interventions risk certification.

Both are organization level instruments that reach the product Eleos sells under its own name; neither is a product specific evaluation, and no bias evaluation or subgroup result is published. The groups where performance is plausibly at risk follow from the product's inputs: accent, dialect and the more than 150 languages it transcribes before writing notes in English, and client presentations that change speech, such as thought disorder, medication effects or intoxication.

The mechanism is review before commit: Eleos drafts note content, providers review, edit and submit, and its Compliance product scans notes for documentation errors before a claim is submitted. What stands behind the output is thin. Eleos states a transcription accuracy of around 93 to 94 percent with no method or test set, and publishes no warranty, indemnity or remediation commitment for a wrong note or a wrong intervention identification. Its science page lists peer reviewed studies, some with academic collaborators; none is presented there as measuring the error rate of the product's own output with a stated denominator.

Rather than integrating with record systems in the usual sense, Eleos says, it embeds in the provider's web based EHR through a browser extension. Its AI FAQ names the systems it works with: Carelogic, Credible, InSync, SmartCare, My Avatar NX, MyEvolv, EchoVantage, PCE, Kipu, Welligent, eHana and Exym, specialty record systems that community behavioral health runs on. That is one directional writing into named systems. Eleos also describes integrations with client systems over its own VPN without naming them. No bidirectional exchange, FHIR API or network connection is named.

Deployment, residency and isolation are each documented. Eleos runs as browser embedded software over the provider's web based EHR, with an Outreach product for field based care. All customer data is stored and processed on HIPAA eligible AWS servers in the continental United States, and Eleos supports deployment to designated geographic regions with a stated guarantee that data will not move outside the region originally designated. Data is logically separated between clients, and integrations with client systems run over an Eleos-managed VPN linking the customer's data center to Eleos's cloud.

There is no published price and no pricing basis. Eleos sells to provider organizations, community behavioral health, substance use disorder, home health and hospice agencies, through a demo and contract, and its site describes outcomes and time saved but not what drives the cost, whether clinicians, sessions, notes or modules. The cost cannot be sized before a sales conversation.

Coverage is named broadly within one vertical: community behavioral health, substance use disorder treatment, and since 2025 home health, hospice and palliative care, across individual and group therapy, case management, peer support and prescriber workflows, in the office, by telehealth and in the field, including rural and offline settings, with notes produced from sessions in more than 150 languages. It also serves supervisors, leadership and compliance teams.

Validation sits behind part of that: the published randomized trial ran in outpatient therapy for depression and anxiety at one community clinic. Nothing published validates the substance use, case management, home health or hospice settings.