Ultralight

The models run core modules of a full record system. Ultralight is an EHR and practice platform (charting, labs, prescribing, telehealth, billing, scheduling and a patient app), and its AI scribe, lab and wearable synthesis, and Copilot, which answers questions about a patient's chart and device trends, would not exist without models. Scheduling, billing, messaging and the record itself would still stand without them, which is why this is not an A. Ask which modules a practice can run with the AI switched off.

Control is asserted without a described mechanism. The platform page says charting, visit prep and data synthesis happen automatically while clinicians stay in control of final decisions, and the company's own AI guide tells clinicians to ask any vendor where the line sits between draft and wait and execute automatically.

Ultralight does not publish that line for its own product: nothing says whether AI-drafted notes, care plans or follow-ups need sign-off before they reach the chart or the patient app, and no confidence signal or abstention behavior is described. Ask exactly which outputs wait for approval.

The functions are described and nothing producing them is identified. The company describes an AI scribe, specialty lab parsing, trend synthesis across labs and wearables, and Copilot, and says plans draw on a proprietary knowledge base reviewed by functional medicine experts. No model or model family, training or fine-tuning data, version or update practice is published. Ask which models produce notes, insights and care plans, and how practices learn when they change.

The architecture is described and no model provider is named. The AI features read the chart, labs and device data inside the platform, and the privacy policies name Google Analytics and reserve disclosure to unnamed vendors and service providers. No hosting provider, model provider or subprocessor list is published, and nothing states the retention or training terms that apply once clinical content reaches one. Ask for the subprocessor list, including any model provider and scribe vendor, and the terms that bind them.

Named clinics and vendor figures with no method. Four clinics are quoted on the home page (D5 Health, Farm to Functional, Kuros Medical, Living Proof Medicine), and the company says 75 clinics are live. Its results claims, 10 to 15 hours reclaimed weekly per clinician, a 25 percent increase in patient capacity and 10 percent higher margins, and one clinician's 10 hours a week cut to 30 minutes, carry no method or denominator. Ask a named clinic for its before and after measures.

A retention schedule is published for the patient app and nothing about what reaches a model. The patient app privacy policy (18 Apr 2025) says the company processes information on behalf of the practice, hosts in the United States, keeps account data for 30 days after a verified deletion request, communications and security logs for six years and usage data for up to 12 months, and may use de-identified or aggregated data for any lawful purpose.

Nothing says whether clinical content reaches an outside model provider or trains any model. The App Store listing declares that the developer does not collect any data from the app, which the company's own policy contradicts. Ask whether patient data trains any model, which provider processes it, and for the terms that govern provider-controlled records.

Agreement status is stated and a service privacy document stands behind it. The terms of use (9 Jun 2025) permit production access only after a master subscription agreement and, where applicable, a business associate agreement are signed, and the patient app privacy policy describes the company as processing information on behalf of the practice. The agreement itself is not published, and the clinical app privacy policy is a short website notice from July 2024. Many practices in this market are cash pay, so ask when the company treats the agreement as applicable, and for the agreement and its subcontractor terms.

Nothing independent is published. The company's security page is published with a generic heading and no security content, the privacy policies promise reasonable safeguards, and no attestation, penetration test, trust center or report request path appears on the pages read or in search. For a product that holds a practice's entire record, prescribing and messaging, ask for any independent assessment, the backup and recovery arrangements, and the breach process.

No device claim is made and no regulatory position is published, beside functions that press on one. FDA's 510(k) database lists no clearance for the company. The platform offers evidence-based insights to guide recommendations, predictive insights on what each patient needs next, continuously updated therapy guidance, and personalized care plans covering supplements, IVs, peptides, nutrition and lifestyle that are shared with the patient. Ask for the company's regulatory analysis of the therapy guidance and predictive insights, and whether a clinician approves each plan before a patient sees it.

Expert review is claimed without an evaluation. Treatment plans and protocols draw on a proprietary knowledge base the company says is reviewed by functional medicine experts, and practices can add their own; advisors with functional medicine credentials are named, and the company's AI guide warns that general models were trained on conventional medicine.

No evaluation of the scribe, the synthesis or the care plan output is published, by patient group or otherwise, and nothing says what sources the knowledge base draws on or how it is updated. Ask for the knowledge base's sources and review process, and for any accuracy check on generated care plans.

Nothing published says what happens when the output is wrong. No error rate, accuracy figure for the scribe or the lab synthesis, warranty, indemnity or remediation commitment is published, and no mechanism for challenging an output (citations to sources, review before commit) is described for the product. The patient app privacy policy lists audit trails only as a record type it retains for six years.

The sandbox is provided as is, and production use requires a master subscription agreement that is not published. Ask what the agreement says about errors in AI-drafted notes and care plans, and how a clinician traces an insight to its source.

Named device connections read into the record; lab, exchange and export routes are not named. The company's wearable integration guide documents direct connections to Dexcom, FreeStyle Libre and Oura, and Apple Health and Android Health Connect for other devices, with Whoop in progress; the data appears in the chart's trends beside labs.

Specialty labs and multi-omics results are described as auto-ingested and trended, with no laboratory named, and migration from a previous EHR is offered as an AI-driven service. Because this is the system of record, ask which labs interface directly, whether e-prescribing and exchange connections exist, and in what format a practice can export its full record on exit.

A single hosted service with the country stated and nothing more. Both privacy policies say the services are hosted in the United States, that data may be moved to other countries for storage and processing, and that the services are intended for users in the United States. No cloud provider, region, tenant isolation or customer hosted option is published. The patient app runs on iPhone and Android. Ask for the hosting provider and region, which processing leaves the United States, and what the app keeps on the device.

Tiers and seat prices are published, and starting still goes through a form. The Core plan for practices above 50 patients is priced per seat: $525 for a prescribing clinician, $300 non-prescribing and $50 a month for unlimited admin seats, with patient app and platform features included; implementation is $995 per practice, and data migration is custom priced. Practices under 50 patients apply for special pricing, and multi-site or 1,000+ patient practices get custom pricing. The billing period for clinician seats is not stated, and every Get started link opens an intake form. Ask for the seat period, contract term and migration estimate.

Coverage is stated clearly with nothing validating it. The product is built for functional, integrative and longevity medicine, concierge and cash-pay practices, and women's health and hormone clinics, each described on the home page, and the company says 75 clinics are live. No evaluation of the AI output in any of these settings is published, and the specialty knowledge base is described without its sources. Ask which specialties the knowledge base covers and how its protocols were reviewed.