LucasAI

Every core function is model output. A medical trained large language model turns the transcribed encounter into a structured note in real time, CodingAI derives CPT, ICD-10 and E and M suggestions from the same conversation, Case Guardian draws guideline based recommendations from it, and the behavioral templates produce PHQ and GAD assessments. The newer workflows build on the same layer: Prior AuthAI assembles supporting documentation, DME Ordering pre fills orders and drafts medical necessity narratives, and voice agents handle intake and outreach calls.

Insurance verification, packing slips and proof of delivery tracking in DME Ordering are workflow software around the models.

Clinicians and coders sign off on what reaches the chart and the claim. Coding suggestions are reviewed, edited and confirmed by the clinician or coder before a claim is submitted, with the company calling CodingAI decision support rather than autopilot; notes are formatted for review before entry into the EHR; and every DME order is reviewed and signed by the physician after Medicare contractor checks run.

How a generated PHQ or GAD score is confirmed before it is recorded is not described.

The pipeline runs in two stages: a HIPAA compliant transcription service converts the conversation to text, and privacy conscious NLP and a medical trained large language model turn it into structured notes such as SOAP notes in real time. The models are tuned by specialty for workflows, care models, reimbursement rules and prior authorization requirements, and the company is building self hosted speech to text and NLP to reduce outside dependencies.

The transcription service, the language model and any accuracy figure are not named or published.

Encounters pass through a HIPAA compliant transcription service and a medical trained large language model before the note returns, and LucasAI requires business associate agreements from vendors that handle protected health information. The company is building self hosted speech to text and NLP to reduce outside dependencies.

None of the transcription, model or hosting providers is named, and no subprocessor list is published.

Outcomes come from LucasAI's own data across more than 1,200 providers in 2026, reported as time saved, visits per provider and revenue per provider. Physician testimonials name the settings: pulmonary critical care in Michigan, internal medicine in Nevada, acute rehabilitation in Idaho, concierge medicine in Florida and hospital medicine. A hospitalist's public app store review compares it with Freed, Ambience, Microsoft DAX Copilot and Sunoh, which that physician tried first.

No study, controlled evaluation or accuracy benchmark has been published.

LucasAI processes the encounter on the spot and leaves the finished document in the patient's EMR, saying no patient data is stored permanently, with an option for additional anonymization. Data is encrypted in transit and at rest, access uses two factor authentication and role based controls, and providers must obtain patient consent before using audio transcription.

That commitment sits beside PushAI Note Forwarding, described as AI that remembers prior notes. Whether customer content trains the models is not stated.

LucasAI signs business associate agreements, offering an organizational BAA as a plan feature, and requires compliant agreements from its own vendors that may handle protected health information, with those vendors bound to protect data passed to subcontractors and to report breaches within 60 days. A HIPAA and security page sets out administrative, physical and technical safeguards, from access revocation and unique user IDs to contingency testing and six year retention of security documentation.

The agreement itself and the plans it applies to are not published.

The HIPAA and security page lists the controls LucasAI runs: encryption in transit and at rest, two factor authentication, role based access, automatic session timeout, login monitoring, malware detection, audit controls, backups and disaster recovery, a designated security officer and staff security training. A public status page reports uptime.

No SOC 2 report, HITRUST certification, ISO 27001 certificate or outside penetration test is named.

The platform is sold as documentation, coding and workflow software, and no clearance is held for any part of it. Several outputs sit close to clinical judgment: Case Guardian offers guideline based recommendations during the visit, the behavioral templates generate PHQ and GAD assessments from conversation rather than from the patient answering the instruments, and the top plan lists patient triage and an early detection assistant.

No intended use statement covers those outputs.

Coding output is framed around compliance: CMS compliant ICD-10, CPT, HCC and RAF codes, HCC and V28 analytics, an audit mitigation assistant and a Magic Addendum to support the chosen codes, sold as reducing clawbacks and denials. The product serves bilingual encounters, Spanish speaking practices and specialty templates across many settings.

No fairness statement, subgroup result or monitoring practice is published for notes, codes or generated screening scores.

Review before commit is the main safeguard. Coding suggestions are confirmed by the clinician or coder before any claim goes out, DME orders are checked against Medicare contractor requirements and signed by the physician, and notes are formatted for review before they enter the EHR. Providers carry responsibility for consent to audio transcription under the terms.

No accuracy figure, warranty, indemnity or remediation commitment is published, including for generated screening scores.

LucasAI publishes integration pages for Epic, Oracle Health (Cerner), athenahealth, eClinicalWorks and NextGen, each describing ambient capture, a finished note and suggested codes kept alongside that system's workflow through a lightweight browser based approach, and EMR integration is a feature of the higher plans. The company also says no specific connectors are needed, since the note or its sections can be copied into any EHR, and states it is not affiliated with or endorsed by those EHR vendors.

No marketplace listing, interface standard or named production integration is published.

Delivery is a web platform with iOS and mobile apps and a browser extension, sold as software as a service. The company says encounter data is processed on the spot rather than stored, with the finished note going into the practice's EMR, and it is working toward self hosted speech to text and NLP. The terms are governed by Delaware law.

Hosting provider, region and tenant isolation are not stated.

Four plans are priced on the site. Resident costs $39.99 a month for up to 50 notes, Premium $69.99 a month with unlimited notes (marked down from $149.99), Pro $299 a month and the top tier $999 a month, with custom terms through sales for groups. A free trial of 30 days or 10 notes starts online without a sales call, and an ROI calculator and a groups page support larger practices.

Higher plans add Case Guardian, Prior AuthAI, document management, medication reconciliation, coding analytics, EMR integration, an organizational BAA and, at the top, voice agents and nursing and front desk assistants.

Coverage is built around the documents each setting needs. Primary care gets HCC and V28 risk adjustment, prior authorizations and care gaps; hospital medicine gets rounding notes, admission and progress notes and one click discharge summaries across inpatient, emergency and urgent care; surgical work gets pre operative evaluations, operative and post operative notes and risk assessments; behavioral health gets PHQ and GAD assessments. Specialty pages cover cardiology, orthopedics, podiatry, pulmonology, emergency medicine and family medicine, DME Ordering serves wound care clinics and podiatry, and the top plan adds skilled nursing, physical therapy and home health.

Named users include physicians in pulmonary critical care, internal medicine, acute rehabilitation, concierge medicine and hospital medicine. Performance is not reported separately by setting.