OneChart

Capture, transcription, note structuring, extraction from charts and referrals, coding suggestions and patient intake are all model output. No human scribe layer, no services tier and no underlying practice management product that the AI is a feature of. The company sells the documentation engine itself and also licenses it as embeddable widgets and white label components, which is the AI being the product rather than a feature attached to one.

A review gate is described explicitly and sits before anything leaves the product: the clinician edits or approves the note, then exports it to the record. Coding output is framed as suggestion rather than assignment, and is delivered with supporting context and confidence levels so the clinician has something to evaluate rather than a bare code to accept. The company states in its own materials that control stays with the physician.

No confidence threshold, abstention behavior or defined escalation path is published, and the product also prefills pathology requisitions and schedules postoperative protocols from spoken input, which reaches beyond note text into clinical artifacts without a separately described review step for those outputs.

The strongest feature here is a product property rather than disclosure about the technology.

The feature is real: code suggestions are presented with supporting context and confidence levels, so a per suggestion confidence signal reaches the clinician at the point of decision. But surfacing a confidence number is a property of the interface, not a statement about what produces the output. Neither the approach nor the suppliers are named.

What is published about the technology itself is general. No model or model family is identified, no foundation model provider is named, no architecture is described beyond the word artificial intelligence, no training data is characterized, and no versioning or update practice is published, so nothing shows when the system changed or what changed with it. Proprietary capability is asserted rather than explained.

The confidence signal raises a further question. No calibration basis is published, so a clinician sees a confidence number without knowing what it was measured against, what a high one has historically meant, or whether any threshold alters the product's behavior. A confidence figure that is not calibrated against anything carries the appearance of measurement without the substance, which on a coding suggestion is the place it will be relied on hardest.

Nothing identifies any party in the chain. No model or model family, no foundation model provider, no hosting arrangement and no subprocessor list is published, and no training position is stated in either direction.

Three categories of content flow through that unnamed chain, and each probably follows its own rules, so no single retention answer fits them all.

The product ingests prior charts, laboratory results, imaging and referrals to build summaries, so it holds historical record content that was never part of the encounter it is documenting.

It captures patient-recorded audio directly, through a link sent to the patient that lets them describe symptoms by voice before the visit.

And it operates across dentistry, psychiatry and behavioral health alongside general practice, so some of what it holds carries the heightened protections attaching to mental health records.

The patient-recorded audio is the category with no obvious owner in the usual framing and it deserves its own answer. It was created by the patient, at the vendor's invitation, before any clinician was involved, and it sits with the vendor rather than in a chart. Who holds it, under what basis, for how long, whether it is used for improvement, and what the patient is told when they record it are all unanswered, and the ordinary analysis of a scribe does not reach it because the ordinary scribe never has a recording the patient made alone.

No study, controlled evaluation, accuracy figure, third party rating or verified deployment count is published.

The performance claims are all vendor stated and carry no denominator, method or period: a 50 percent reduction in manual documentation time, 20 hours saved weekly on charting, more than $30,000 in lost revenue recovered per year, and 90 percent or better completion on automated form filling. The last of those is the most checkable in principle and the most consequential in practice, since a form completed to 90 percent is one a clinician must still inspect in full to find the missing tenth.

The named organization claims need care, and the vendor's own wording is narrower than it first appears. Its site displays a logo wall under the heading that it is trusted by clinicians and healthcare staff from Kaiser Permanente, the University of British Columbia, McMaster University, Canada Basketball and others. Clinicians from an organization is a materially weaker claim than the organization as a customer: it is consistent with individual practitioners signing up on a free tier. For a nine person seed stage company those names are large, and nothing published gives the contracting entity, the scope or the number of clinicians, so the wall does not show an institutional deployment.

The three testimonials are labeled by specialty rather than by named clinician or practice, so none is independently checkable.

One publishing practice matters. The company publishes ranked listings of the best AI scribes on its own domain, including comparisons naming competitors, with itself placed favorably. That is marketing content on the vendor's own estate rather than third party assessment, whatever its position in a search result.

Health privacy compliance is asserted repeatedly across the product pages, and none of the three lifecycle commitments is published. No retention schedule, no audio deletion commitment, no de-identification practice, and no statement in either direction on whether customer content is used to train or improve models. That combination is the largest gap in what the company publishes.

The exposure is wider than a scribe's in three directions, and they are unlikely to be governed identically, so a single retention answer would not cover all three.

The product ingests prior charts, laboratory results, imaging and referrals to build summaries, so it holds historical record content that was never part of the encounter being documented.

It captures patient-recorded audio directly, through a link sent by text or email that lets a patient describe symptoms by voice before the visit. That is patient-generated content held by the vendor outside any clinical encounter, created at the vendor's invitation before a clinician was involved, and the ordinary analysis of a scribe does not reach it.

And it operates across dentistry, psychiatry and behavioral health alongside general practice, so some of what it holds carries the heightened protections attaching to mental health records.

The training silence weighs most on that last category. A behavioral health session transcript is among the most sensitive text a documentation product will ever hold, and a buyer in that setting is entitled to a stated position rather than an absence.

What is published is transport and access rather than lifecycle: encryption at rest and in transit, optional two-factor authentication and administrative controls. Those are real and they do not answer any of the three questions above.

What exists is a compliance claim. The product and specialty pages describe leading compliance standards, leading privacy standards and industry standard encryption at rest and in transit. What the company does not publish is a statement of business associate status, an offer of a business associate agreement, its scope, or any tier at which one is executed. The site footer carries a privacy page and a frequently asked questions page; there is no terms of service, no security page and no legal section beyond that.

Two features of how the product is sold make that gap operational rather than theoretical. It is offered on a free tier with self serve sign up and no sales conversation, so a covered entity can begin putting patient information through it before any agreement exists, which under the United States rule is the violation rather than a technicality. And the company is Canadian, based in Hamilton, Ontario, handling the Ontario provincial billing schedule alongside United States code sets, so it serves both markets. Which regime governs a given account is not stated, and the obligations differ: a business associate agreement is a United States instrument, and Ontario health information custodians operate under provincial law with its own requirements for agents and service providers.

No named or dated certification, audit report, penetration testing statement or trust center exists. The public site is thin: its footer carries a privacy page and a frequently asked questions page, with no terms of service and no security page. Security is described through phrases rather than artifacts, leading compliance standards and leading privacy standards, alongside industry-standard encryption at rest and in transit and options for two-factor authentication and administrative controls.

Two features of how this product is sold make the absence more consequential than the size of the company would suggest.

It is offered on a free tier aimed at solo practitioners and small practices. That buyer will never run a vendor security review and has no procurement function, yet carries the regulatory obligation personally. A vendor reaching clinicians who cannot assess it has a stronger reason to publish independent assurance than one selling to a hospital security team, not a weaker one.

And the company offers application interfaces and custom integrations at its enterprise tier and markets a partner program, so other software can place its documentation and coding functions inside their own products. A developer buying that does not merely trust this vendor; it inherits this vendor's posture into its own product and passes it downstream to practices and patients who will never see the name. Subprocessors, retention and access controls are more load-bearing for a component sold to builders than for a product sold to one clinic.

The breadth of the surface compounds both. The platform ingests prior charts, laboratory results, imaging and referrals, captures patient-recorded audio before the visit, and operates across general practice, dentistry, psychiatry and behavioral health.

The company is at an early stage, and it offers that free tier.

No clearance is claimed and none is required for documentation. Even so, the product's scope reaches well beyond a scribe, and several features prepare or generate clinical artifacts rather than record them.

The sharpest is the dermatology workflow. The clinician describes a lesion aloud, the system tags its location on a body map, suggests diagnostic codes and pre-fills the pathology requisition. Two things happen there that documentation does not do. A requisition is an order, so the system is preparing an action taken on the patient rather than recording one. And the anatomical site is being derived from speech and placed on the form that will accompany a specimen. A site recorded incorrectly on a requisition is a wrong-site error in the specimen chain, and it is the kind that surfaces late, after a result has been attributed to the wrong lesion. Nothing published says what the clinician confirms about the site before the requisition is generated, or whether the body map tag is presented for confirmation separately from the note.

Three further surfaces sit beyond documentation. The pediatric workflow auto-updates growth charts, which writes structured measurements, and inserts age-specific anticipatory guidance, which is clinical advice generated rather than recorded. The inpatient workflow produces shift handover summaries, and handover is a recognized high-risk communication point where omission causes harm. And contextual prompts surfacing medication adherence and history to guide decisions at the point of care are decision support rather than documentation.

A patient-facing element is also present: patients receive a link and record their symptoms by voice before the visit, with no clinician in that exchange. The vendor separately markets voice agents and clinical care agents alongside the scribe, which widens the surface again.

The revenue framing reaches into the maximize or optimize language that marks the aggressive end of coding products. The vendor's own homepage uses it twice: suggested billing codes are said to ensure optimized claims, and the billing feature is described as optimizing code management. Alongside that sits a headline figure of $30,000 of lost revenue regenerated per year.

One counterweight stands out, because it is a control rather than a slogan. The plastic surgery documentation preserves the cosmetic versus medically necessary distinction, which is a guard against precisely the misclassification a payer would challenge. Code suggestions are also delivered with confidence levels rather than as bare numbers, which leaves the clinician room to weigh them.

On fairness itself, nothing is published. There is no fairness statement, subgroup analysis, accent or dialect disclosure, or performance breakdown of any kind, and no account of how the models are evaluated or monitored.

The multilingual gap is the sharpest version of that. Multilingual support is sold as a paid tier feature, so the vendor is charging for performance in languages whose performance it has never characterized, and the languages themselves are not named. A clinician documenting in a second language has no way to know whether the product is better or worse at it than the one it was built in.

One mechanism here is well placed: code suggestions are presented with supporting context and confidence levels, so a per suggestion confidence signal reaches the clinician at the point of decision rather than sitting in an interface only an integrator sees. Surfacing confidence where the decision is made is the useful placement, because it lets the person accepting or rejecting a code weight it, which is exactly the discretion a coding suggestion should preserve.

The signal has no published basis. No model card, accuracy methodology or calibration reference is published, so a clinician sees a confidence number without knowing what it was measured against, what a high one has historically meant, or whether any threshold changes the product's behavior. A confidence figure that is not calibrated against anything trains users to trust it at whatever level feels right, which can be worse than publishing no figure at all, because it carries the appearance of measurement.

Nothing stands behind the output. No warranty, indemnity or remediation commitment is published, and no route is described for a patient who finds an error.

The published performance claims are unfalsifiable as written and sit beside the confidence feature awkwardly: a 50 percent reduction in documentation time, 20 hours saved per week, and $30,000 of revenue regenerated per year, none with a denominator, method or period.

Named integrations rather than implied ones. Browser extensions and open application interfaces, with Jane, WebPT and Empower PT named as confirmed connections and a strategic partnership with C3 EMR enabling native ambient documentation and telehealth workflows. Notes export to PDF or text, or copy and paste as a fallback, and the enterprise tier adds application interface access and custom integrations.

The candor about the ceiling matters as much as the integrations. The company states in its own materials that for hospital grade systems the depth of bidirectional sync requires validation, and that some workflows still rely on structured export rather than full native write back. Naming the limit up front marks exactly where the integration stops.

The named connections are practice management and allied health systems rather than major record platforms, nothing is verifiable in a marketplace listing or integration documentation, and no deployment is named against any specific system. The partnership with C3 EMR is described as enabling native documentation but no live site is cited.

The practical consequence follows the vendor's own admission: on a hospital record system this is an export workflow until proven otherwise.

No hosting region, residency option, subprocessor list or on-premise path is published, and nothing establishes whether a third-party model service processes the encounter or what it retains. Delivery is a web platform with browser extensions and interfaces, signed into at a separate domain from the marketing site.

The gap is specific rather than generic here. This is a Canadian company, based in Hamilton, Ontario, serving clinicians on both sides of the border and handling the Ontario provincial billing schedule alongside United States code sets. Provincial health privacy law and the United States rule point at different answers, and at least one province expects health information to remain in country while the United States rule imposes no localization requirement at all. Where the audio and notes physically sit is therefore a question with a buyer on each side needing a different answer, and neither can infer theirs from the other's.

A second reason the answer matters follows from how the company sells. It offers interface access and custom integrations at the enterprise tier and runs a partner program, so other software can place its functions inside their own products. Where that happens the data path acquires another party and another set of terms, and the clinician using the host product may not know this vendor is involved at all. Nothing published says which organization holds what in such a deployment, or under whose agreement.

The ingestion surface makes residency a larger question than one conversation. The product pulls charts, laboratory results, imaging and referrals, and captures patient-recorded audio through links sent by text or email. All of that moves toward the vendor before any note exists.

A full published ladder with the gating disclosed, which is uncommon at this company size. Free at up to 40 notes a month with one template, Plus at $79 per user per month for up to 250 notes with unlimited templates and multilingual support, Pro at $125 for unlimited notes plus billing code suggestions, PDF form filling, voice intake and multi-user workflow, and custom enterprise terms for interface access and dedicated support. Sign-up is direct, so a buyer can start without a sales conversation. Only the enterprise tier is unpriced, which is the normal and defensible exception.

The tier structure is itself informative. Coding support, form filling and voice intake are Pro only, so the real entry price for a practice that wants the billing features is $125 rather than 79, and multilingual support starts at Plus. The true cost of each capability is visible before making contact.

The vendor's own site carries one defect that bears on every figure. The home page carries a second pricing block beneath the main ladder describing a different scheme: a starter plan free for the first 25 notes with seven-day access, a professional plan shown at $0 per month with unlimited notes and an annual discount, and an enterprise tier. That contradicts the ladder above it on both the free allowance and the paid price, and it displays a zero where a figure belongs. The main ladder is the one corroborated elsewhere on the vendor's own estate, and the second block reads as a stale template that was never removed.

Nothing on the site says which of the two schemes governs.

The specialty range is unusually broad for a company of this size. Coverage reaches dentistry, plastic surgery, dermatology, pediatrics, physiotherapy, sports medicine, family medicine, home care, emergency response teams, psychiatry and behavioral health, and the depth is described at instrument level rather than as a template count.

Plastic surgery preserves the cosmetic versus reconstructive distinction in the documentation. Dermatology tags a spoken lesion description onto a body map and pre-fills the pathology requisition. Pediatrics updates growth charts and inserts age specific anticipatory guidance. Behavioral health produces SOAP, DAP and BIRP with modality awareness across cognitive and dialectical behavioral therapy. Inpatient work covers admission checklists, vitals, medication reconciliation and shift handoff summaries. Home care documents visits for later integration with the record, and emergency response covers patient care reports.

It also handles the Ontario provincial billing schedule alongside ICD-10 and CPT, so it documents to a Canadian public payer as well as United States code sets. For a practice billing OHIP, that is the difference between a product that fits and one that does not.

Multilingual support is available from the paid tier upward, with the languages unnamed.

Two limits sit under that breadth. The breadth is described by the vendor and no deployment is named against any individual specialty, so for specialties outside the handful with published detail, nothing shows what has actually been built. And this is a company of roughly nine people; breadth of this kind at that size is a claim about roadmap as much as about shipped depth.