mdhub

The agents are the engine of the platform, and the platform is sold as a system of record that carries value without them.

Three named AI workers do the work the company is bought for. Sarah runs admissions, answering inbound calls, texts and faxes around the clock, qualifying, matching and scheduling patients and verifying insurance. Emma listens during sessions and generates structured behavioral health notes with CPT coding. Eric reviews, codes and submits claims, manages prior authorizations and works denials. Each is model driven end to end, and none of them exists without the models.

What sits underneath is sold separately and on its own merits. The company publishes product pages for a behavioral health EHR, a healthcare CRM and a revenue cycle management system, describes the EHR as built around behavioral health workflows rather than retrofitted from a general medical system, and offers the platform as the clinic's system of record with clinical data migrated in at implementation. E-prescribing and telehealth sit inside it. Strip every model out and a behavioral health practice management system remains, with the record of account, scheduling, billing infrastructure and prescribing. That is a product a buyer would pay for, and one the company describes in its own right.

The company frames it the other way. Its own positioning is that the surrounding system exists to give the agents somewhere to act, which is a true statement of intent and not a statement about what a buyer would still be able to use.

The other side is that nothing distinctive about this company is in the record system. The reason a clinic chooses it over an established behavioral health EHR is the three workers.

The boundary is published in both directions, plainly, and the routing inside it is not.

What the system may not do is stated where it matters most. Every clinical note requires clinician review and sign off before it enters the patient record, output is described as a structured first draft or a flag rather than a final authority, and workflows are said to carry mandatory human checkpoints. The company's guidance is explicit that for high acuity situations including suicidality, self harm and crisis, a qualified provider must evaluate any flag independently and never act on model output alone. The billing sequence follows from the signature rather than bypassing it: the billing agent pulls the signed note, then codes and submits, which puts a licensed human upstream of the claim.

The company goes further on its community clinic page, under a heading about being straight on what it does and does not do, stating that crisis response sits outside what it promises. Publishing a scope exclusion on the sales page for the buyer most likely to need it counts in its favor. It also tells the buyer rather than the caller, and it cannot supply a threshold.

Two reservations remain.

The first is that the review is sold on its speed. The stated target is a clinician signing in seconds rather than after hours. Sign off is the control that makes the whole design safe, and compressing it is presented as the benefit, so the marketing optimizes the thing the safety case depends on.

The second is the admissions agent, and this is the part of the structure that is missing. The agent answers every call, text and fax around the clock, which is sold as the point of it, and answering every channel at every hour means taking on the caller in crisis at two in the morning. Its product page carries no escalation criterion, no crisis behavior, no human handoff path and no transfer condition of any kind, while emphasizing that no inquiry goes unanswered whatever the hour. Nothing published gives the hours a person is actually reachable, or what a caller in distress hears before reaching one.

Nothing published about the models, by a vendor that tells buyers this is the thing to demand.

No foundation model is named, no architecture described, no evaluation methodology published, no versioning or update policy, and no accuracy figure for any of the three agents. That includes the two places a number would matter most: the fidelity of a generated session note to what was actually said, and the correctness of the diagnostic and procedure codes assigned from it.

Coding accuracy is the sharper gap because of what happens next. The billing agent takes the signed note, assigns codes and submits the claim. There is no published confidence threshold, no accuracy figure and no statement of what happens to a low confidence code, so nothing describes whether a doubtful code is flagged, held or simply submitted.

The asymmetry matters because the vendor supplied the test itself. Its published buyer guidance argues that a vendor unable to articulate clearly how its model generates documentation output is asking the buyer to stake clinical and legal liability on a process they cannot verify. That guidance is accurate, useful and well put. Applied to its own published material, no such articulation exists.

One claim on the home page describes the models rather than the platform: the clinical assistant is stated to be trained on behavioral health documentation standards. That says something about the training objective and nothing about what performs it or how well.

Not one party in the chain is identified, while the legal instrument discloses what happens to the content.

No model or model family, no foundation model provider, no hosting arrangement and no sub processor list is published anywhere. A platform running three agents across voice telephony, ambient session capture and claims processing has a chain, and none of it is named.

One retention commitment is real, specific and covers a single artifact: session audio is discarded once the note has been generated and not retained after that. For an ambient system in a therapy room that is the most consequential retention decision available.

What the audio commitment leaves open, the business associate agreement partly closes and in the vendor's favor. That agreement permits mdhub to create de identified information from protected health information under the regulatory de identification standard, and states that mdhub owns all such de identified data. So the transcript and the generated note, de identified, are the company's asset by contract. No page states whether models are trained on that corpus, and no clause forbids it. A buyer reading only the privacy language would not find this; a buyer reading the agreement would.

A second content category sits outside the usual framing entirely. The admissions agent gathers intake detail, screens and qualifies people who have not yet entered a treatment relationship. Highly sensitive mental health information is therefore created at a point where the protections attaching to a clinical record may not yet apply, about people who may never become patients, and the business associate agreement by its own terms applies only where the provider is a covered entity and mdhub is acting as a business associate.

Named deployments with figures attached, and no method behind the headline numbers.

The named evidence is specific. University Health Kansas City reports clinicians saving between six and twenty minutes per note with behavioral health notes needing minimal editing. Family Care Center reports more than 600 clinicians on the platform and past 50,000 sessions a month. Radial appears twice, once on admissions qualification and once from a practicing psychiatrist on clinician adoption. A small practice serving neurodivergent adults and couples is named. Scale is stated at more than 10,000 clinicians nationwide, and listing on the athenaOne marketplace is modest independent validation of integration quality, since the partner states that listing requires passing its technical and security vetting.

The headline claims are the ones with nothing behind them: 30 percent more patients scheduled, denials cut by 70 percent, more than two hours given back to clinicians daily, operational costs down by half, and 98 percent of clinics reporting improved efficiency. None carries a site, a period, a denominator, a baseline or a method. The named testimonials speak to documentation time, which supports one of the five; the denial reduction and the cost halving, which are the strongest commercial claims made, are supported by nothing published.

The satisfaction figure and the efficiency figure also measure reported perception rather than outcome.

No study, controlled evaluation or third party assessment is published, and none would be expected for an operations platform. The arithmetic behind the denial and cost figures is what is missing.

One strong retention commitment, one answer buried in the legal instrument, and one question the product's own design raises and does not answer.

The commitment first, because it is specific and checkable: session audio is discarded once the note has been generated and not retained afterward. For an ambient system in a therapy room that is the single most consequential retention decision available, and making it plainly is better practice than a general assurance. Encryption at rest and in transit, role based access and audit logging are stated, alongside a commitment not to store data beyond what treatment requires.

The training question is not unanswered, but the answer is in the business associate agreement rather than on any privacy page. That agreement permits mdhub to create deidentified information from protected health information under the regulatory deidentification standard, and states that mdhub owns all such deidentified data. So the route by which session content becomes a corporate asset is published, lawfully and explicitly, while no page states whether models are in fact trained on it. The company reports supporting millions of patient sessions, so the corpus is large and it is psychotherapy content, and reidentification risk in behaviorally rich text is well documented. The ownership clause is the operative disclosure.

The second open question is data collected before anyone is a patient. The admissions agent gathers intake detail, screens and qualifies people who have not yet entered a treatment relationship, and operates around the clock by voice, text and fax. Highly sensitive mental health information is therefore created at a point where the protections attaching to a clinical record may not yet apply, about people who may never become patients, and where the caller may not understand they are talking to a system that records.

The agreement is published in full, which is rare, and reading it answers a question the marketing does not and raises one the marketing does not either.

The instrument is on the site, dated 25 June 2025, and a buyer can read it before contracting. It is available to every customer rather than to enterprise accounts only, which removes the usual smaller buyer gap, and the company states it works directly with a customer's security team through vendor review and shares audit documentation. Its published guidance on the health privacy rule is accurate, including the covered entity and business associate distinction and the point that layering several tools each under its own agreement creates uncovered gaps at the handoffs, which is a fair argument for a single platform and a real risk. Breach notification without unreasonable delay is stated, unsuccessful security incidents are defined and carved out by deemed notice, and return or destruction of protected health information at termination is covered.

One clause is easy to miss. The agreement permits mdhub to use protected health information to create de identified information under the de identification standard, and states that mdhub owns all such de identified data. That is lawful and common, and it is also the mechanism by which therapy session content can become a corporate asset.

What is absent from the published agreement is the subcontractor flow down. No clause requires subcontractors who create, receive, maintain or transmit protected health information on mdhub's behalf to agree to the same restrictions, which is the standard obligation. The general assurances clause covering disclosures for mdhub's own management is not a substitute for it.

The substance use deferral sits alongside. Asked whether the platform handles records under the separate federal confidentiality regime, the published answer recommends that practices discuss the requirements with their own compliance team. That is not a statement that the platform segments those records from automated processing, and for a practice serving that population the question stays open.

A real posture with real third party involvement, carrying a discrepancy and a wording error the vendor should fix.

What is claimed and supported: an attestation named with its type as SOC 2 Type II, encryption in transit and at rest, role based access, audit logging, single sign on through Microsoft and Google, a business associate agreement published in full and available to every customer, and a stated willingness to share audit documentation with a buyer's security team. Listing on the athenahealth marketplace carries independent weight, since the partner states that listing requires passing its technical and security vetting rather than resting on vendor assertion.

The discrepancy concerns the attestation. The company's audit partner published a case study describing completion of health privacy and SOC 2 Type I work and describing the company as preparing for the Type II audit. The vendor's own material states Type II. The case study carries an April 2026 date, so a Type II report may well have completed since and both statements may have been true when written. A buyer cannot tell from published material which is current, and no report, period or date of issue is published.

The wording error is live on the home page, which describes the company as SOC 2 certified. SOC 2 produces an attestation report from an accounting firm. There is no certifying body, no certificate and no registry, so certified is the wrong word, and it invites a buyer to credit an external verification of a kind that does not exist in this framework.

No trust center, penetration testing statement or vulnerability disclosure policy is published, and the footer's security documentation reference does not resolve to a reachable document.

No clearance, almost certainly none needed, and a regulatory surface unusually wide for a single product.

Documentation, scheduling and billing software sits outside the device definition. That part is simple. The harder part is that this platform runs three agents across admissions, clinical documentation and billing, plus eprescribing and telehealth, and each function lands in a different regime. A buyer inherits all of them at once.

Substance use disorder records are governed by 42 CFR Part 2, whose modernized framework reached its enforcement date on 16 February 2026. Part 2 is stricter than the health privacy rule: it requires specific patient consent for disclosures the privacy rule would permit for payment and operations without one, and it restricts redisclosure downstream. The technical requirement it creates is concrete, namely that automated systems must segment Part 2 records out of processing unless consent is on file.

That requirement points directly at the admissions agent. An artificial intelligence answering an inbound call and reading appointment history to give context can disclose that a caller has a substance use appointment, to a person not authorized to receive it. It also points at the company's buyer targeting: certified community behavioral health clinics are federally required to provide substance use treatment, and the company sells to them by name.

Three further regimes apply. Outbound patient engagement engages the Telephone Consumer Protection Act, under which artificial voice calls need prior express consent and automated texts are separately covered; inbound contact raises none of this, so direction is the distinction, and the admissions agent sends appointment reminders. Claim submission carries no vendor level regulator at all, with liability sitting on the clinic under the False Claims Act. And many states impose mental health confidentiality rules stricter than the federal floor.

Nothing published states the company's position on any of them, and its answer on Part 2 refers practices to their own compliance team.

A published ethics position and no published evidence about its own models.

The company writes about ethical artificial intelligence in behavioral health, including bias auditing and clinician oversight, and the writing is sound. What is absent is any performance data broken out by group, any bias testing statement about its own systems, any governance or model update policy, and any third party audit of the models.

Two concerns are specific to this product rather than generic.

The first is speech, and it applies to two of the three agents. The clinical assistant transcribes therapy sessions and the admissions agent conducts live calls. Speech recognition accuracy varies measurably across accent, dialect and first language. The company markets directly to community and certified community behavioral health clinics, which serve a disproportionately multilingual and dialect diverse population, and publishes no language coverage at all. A note generated from a degraded transcript is a clinical record that misstates what a patient said, and it is signed and billed.

The second is gating, and it is the sharper of the two. The admissions agent qualifies prospective patients, matches them to a clinician and verifies eligibility so that, in the company's own words, the team only sees qualified patients. That is an access decision made before anyone becomes a patient, executed by a model, with no clinician in the loop. If qualification, scheduling priority or clinician matching varies systematically by how someone speaks or by their coverage, the effect falls on who receives care at all. Nothing published describes what qualification means operationally, what criteria are applied, or what is measured.

Nothing stands behind the output, and the billing agent is where that matters most.

No accuracy figure, error rate, published limitation, evaluation methodology, warranty, indemnity or remediation commitment is published for any of the three agents. Two gaps sit where a number would matter most: the fidelity of a generated session note to what was actually said, and the correctness of the diagnostic and procedure codes assigned from it.

Coding is the sharper one because of what happens next. The billing agent takes the signed note, assigns codes and submits the claim, so an error leaves the building as a representation to a payer, with false claims consequences landing on the practice rather than on the vendor. No threshold is published, no accuracy figure, and no statement of what happens to a low confidence code. The clinician signature sits upstream of the claim, which is the right architecture, and a signature on a note is not a review of the codes later derived from it.

The vendor's own buyer guidance argues that a vendor unable to articulate how its model generates documentation is asking the buyer to stake clinical and legal liability on a process they cannot verify. That is a correct test and its own material does not meet it. A buyer can act on that directly rather than reading it as a criticism.

One mechanism does exist: mandatory clinician sign off before a note enters the record, with output framed as a draft rather than an authority. That is a control a clinic can exercise, and it is the reason this is a considered absence of commitment rather than an absence of any structure at all.

Interoperability reads differently here, because this vendor is also an electronic health record.

The platform includes a behavioral health native record system designed around its own agents, described as built for behavioral health workflows rather than retrofitted from a general medical system, so for a clinic adopting the full stack the usual integration question dissolves. What remains is what it connects to for everyone else, and there the position is flexible: the three agents are described as operating independently or together, so a clinic may keep an existing record system and run agents alongside it before committing to the whole platform.

The strongest external evidence is the athenahealth relationship. Listing on the athenaOne marketplace is corroborated outside the vendor's own site and the partner states that listing requires passing its technical and security vetting. That is a check on integration quality rather than a logo.

An open interface is offered for integration and data access, described as used to connect analytics and workflow systems, and single sign on is supported through Microsoft and Google. For a buyer, an open interface on the record of account matters more than any single connector, because it is what makes reporting and eventual migration possible. E-prescribing sits inside the platform, which moves part of the clinical record through a national network.

The qualifications are real. Only one major record system is named specifically, and integration with others is asserted without naming them. No standards based interoperability is described, no support for the usual clinical data exchange standards is stated, and no interface documentation, uptime commitment or export specification is published; the footer's integrations reference does not resolve to a reachable document.

Nothing published on where any of it runs, under a purchase that is hard to reverse.

No hosting provider, region, tenancy model, subprocessor list, backup or recovery posture or data residency commitment is published. The service is evidently cloud delivered and vendor operated, with implementation managed end to end and clinics live within days for a single agent or about four weeks for the full platform including configuration, integration, data migration and training.

The weightiest deployment fact is not technical. This is offered as a full platform including the record of account, so adopting it at full scope is a system of record replacement rather than a tool addition. Clinical records migrate in at implementation. That raises the exit question sharply: what leaves, in what format, on what timetable, and who bears the cost. An open interface for integration and data access is stated, which helps, and no export specification or retention and return terms are published. The business associate agreement does commit mdhub to return or destroy protected health information at termination where feasible, which is the floor rather than an export plan.

The adoption path is better designed than the exit path. The three agents are described as operating independently or together, so a clinic can run one alongside an existing record system and move across later, and the initial commitment stays partial and reversible. The commitment becomes hard only at the point the record itself moves.

No pricing published, detailed returns published, and a switching position that changes what the price means.

Nothing is disclosed: no rate card, no basis, no statement of whether pricing is per clinician, per agent, per session or per site, no term, no minimum and no volume banding. Sales are demo led with implementation managed by the vendor, and practices are stated to go live within days on a single agent and about four weeks for the full platform.

The returns are specific by contrast: around 30 percent more patients scheduled, denials cut by 70 percent, more than two hours saved per clinician per day, operational costs down by up to half, and 98 percent of clinics reporting improved efficiency. A buyer is given a precise numerator and no denominator. The figures may be sound and they cannot be weighed against anything until a price sits beside them.

What is particular here is the position the buyer occupies after signing. This is offered as a full platform including the record of account, so at full adoption the vendor is not a tool the clinic uses but the system the clinic runs on. Pricing power at renewal is different when leaving means migrating the clinical record. That is not a criticism of the model, which has real advantages the vendor argues well, but it is a commercial fact.

The modular path is a real mitigation. The three agents are described as operating independently or together, so a clinic can adopt one alongside its existing record system and move fully later, which keeps early commitment reversible. The moment that changes is the record migration.

Tightly and deliberately bounded, named at a usable level, with one part of the named scope unaddressed.

The naming is specific. Settings: community and certified community behavioral health clinics, group practices, and enterprise multi site organizations. Clinician types: psychiatrists, therapists and psychiatric mental health nurse practitioners, each with its own page. Use cases are enumerated rather than implied, covering scheduling and intake, medication refill lines, telehealth, chart audit, referral management, practice management, billing and documentation. Support behind it is real: more than 10,000 clinicians, named deployments including a 600 clinician organization running past 50,000 sessions a month, and a university health system.

The company also publishes what it does not do, on the page selling to the segment most likely to need it. Its community clinic page states plainly that mdhub is not a certified community behavioral health clinic compliance system, that prospective payment system billing, quality measure and cost reporting sit outside what it promises, and that crisis response does too.

One gap sits inside the scope the company does claim. Certified community behavioral health clinics are federally required to provide substance use disorder treatment, and the company markets to them directly. Substance use records are governed by a separate federal confidentiality regime with its own consent and redisclosure rules, and the company's published answer on whether the platform handles those records recommends that practices discuss the requirements with their own compliance team. That is a deferral rather than a scope statement, and it leaves a named setting only partly covered.

No language coverage is published, which matters because community mental health serves a disproportionately multilingual population and two of the three agents depend on understanding speech. No pediatric or adolescent position is stated.