Murphi.ai

Models are the engine of Murphi's core modules: ambient documentation that fills OASIS and visit notes from a home visit, item by item scoring of the 112 OASIS items with the patient's words as evidence, a quality check for inconsistencies, and coding and revenue integrity reviews across plans of care, PDGM grouping and hospice certifications. Agentic AI and RPA also move data into and out of agency EHRs.

The platform carries other value as well. Care team messaging, SMS, fax, scanning and e signature, and text to pay patient collections are workflow tools that would work without models, and they are sold as modules of their own. Murphi calls itself AI native, and the documentation and coding modules are where that holds.

Murphi sets out what its AI may and may not do. Every output, from OASIS scoring and coding to appeals and claim data, is a suggestion that users must review before any reliance or submission; the AI makes no autonomous clinical decisions; and human review of clinical decisions is mandatory and cannot be waived. Customers may consent to autonomous workflows for non clinical work, and then assume responsibility for them.

Inside the documentation module, each OASIS item shows the patient's words that support its score, and a quality check flags items that conflict with other assessments, such as ambulation against the physical therapy evaluation, before the clinician syncs to the EHR.

The approach and the platforms are named. Murphi describes ambient AI, natural language processing, RPA and agentic AI in its AI terms, explains OASIS scoring with quoted evidence and a quality check, and names OpenAI, Google's Vertex AI and Anthropic's Claude among its platforms, with AWS and Google Cloud.

Which model performs which task, which versions are in use and how updates are tested and announced are not stated, and Murphi says it owns its own models and algorithms beside those platforms. An agency can see the suppliers but not when the model behind its OASIS scores or wound descriptions changes.

The model and cloud platforms are named: OpenAI, Google's Vertex AI and Anthropic's Claude appear among Murphi's platforms, with AWS and Google Cloud for US hosting, and Twilio, Sinch, Vonage and WhatsApp for messaging. Murphi adds a contractual guarantee that customer data is not used for training.

The list is a logo strip rather than a subprocessor list, so it does not say which provider handles visit audio, transcripts, wound photos or OASIS scoring, or what retention terms apply with each. The AI terms' license over de identified data is the term an agency would weigh against that no training guarantee.

Company figures and named agencies stand behind the product without a method attached. Murphi claims 99 percent charting accuracy, 98 percent accuracy for AI drafts with human review, four to six hours saved per clinician each day, a potential 10 to 20 percent revenue increase from accurate charting, 95 percent instant patient responses to care messages and 75 percent of patients paying within 30 days of a text, and says about 200 agencies use the platform.

Agencies such as Infinity Home Health, PD Hospice, Visiting Nurses of Illinois, St. Claire Medical Center and Advanced Home Health and Hospice appear among its customers, and it won a 2026 Stevie award. How charting accuracy was measured, against which reviewer and on how many OASIS items, is not shown, and the wound feature's 98 percent match comes with no test behind it.

Two statements pull in different directions. Murphi says it never trains models on customer data and calls that guarantee contractual. Its AI terms keep a license to use de identified data, feedback, corrections and suggestions to improve its services and models, with de identified data usable without limitation, and leave any training on protected health information to the customer's business associate agreement. De identification may be done by expert determination, safe harbor or Murphi's own algorithms.

No retention period for visit recordings or transcripts is stated. Safety engineering shows in the product design, evidence quotes for each OASIS score and a quality check that flags inconsistent items, without a description of how a wrong score that passes the check is caught.

Murphi executes a business associate agreement with every customer, covering all of its services, integrations and data processing, and describes administrative, physical and technical HIPAA safeguards with workforce training and an annual risk review. Its AI terms say a signed agreement controls in any conflict over protected health information.

Agencies see its terms, including how it binds the cloud and model providers Murphi works with, only after signing. The AI terms also license de identified data broadly, which an agency would want to check against what its agreement allows.

SOC 2 Type II covering security, availability and confidentiality and ISO 27001 certification of an information security management system are both named, with a live status panel showing access reviews, workforce training and an annual risk review as current. Murphi also describes annual third party penetration tests, frequent automated vulnerability scans and a responsible disclosure route.

The reports and certificate are not offered for download, and no audit dates or auditors are given, so currency can only be confirmed by asking. Zero trust controls, end to end encryption and geo redundant hosting with regional failover are described in detail.

Murphi states plainly that its platform is not a regulated medical device and gives its reasoning. Its AI terms say the AI provides informational and workflow support, does not diagnose, treat, cure or prevent disease, and is designed to meet the four Cures Act criteria for clinical decision support that sits outside FDA review: it does not process medical images or signals, it displays information to a professional, the professional can review the basis independently, and the professional is not meant to rely on it primarily. Clinical decisions always require human review.

Most of the product fits that position. OASIS and HOPE items come scored with the patient's own words or the record they came from, and clinicians review and sign before anything reaches the EHR. One marketed feature does not fit. A clinician can upload a wound photo and Murphi describes the wound clinically, including size, tissue and drainage, with an example reading a stage 3 pressure injury at a 98 percent match. Analyzing a photo is what the first of Murphi's four criteria says the product does not do, and the terms do not reconcile the two. The OASIS and HOPE work also sits under federal assessment rules that drive home health and hospice payment and survey findings.

Murphi describes responsible AI practices covering fairness, transparency and accountability, with regular bias and compliance reviews, alongside no training on customer data, mandatory human review and an information security program certified to ISO 27001. What those bias reviews test, and what they have found, is not shared, and no evaluation of OASIS scoring, coding or wound description performance is offered.

The AI terms concede that ambient output depends on audio quality, accent, language and speaker clarity. Home health patients include people with hearing loss, cognitive impairment and limited English, and OASIS items carry payment and quality consequences. No results show whether scoring accuracy differs for those patients or by discipline, so an agency cannot tell where the suggested scores most need checking.

Evidence quotes for each OASIS score, the quality check and audit trails give an agency the means to catch and trace an error. Murphi's AI terms are explicit about everything else: outputs may be inaccurate or incomplete, they are suggestions that are not guaranteed, and owning the output does not increase Murphi's obligations.

Where a customer consents to autonomous non clinical workflows, the customer assumes all responsibility. No error rate with its method, warranty or remediation commitment stands behind OASIS scores or codes that affect payment, despite the 99 percent accuracy claim.

Murphi reads from and writes back to the major home health and hospice EHRs: WellSky, Axxess, HCHB, MatrixCare, KanTime and Netsmart among them. Six connection paths are described, agentic AI, RPA, FHIR R4, HL7 v2, a direct REST API and adapter platforms, with referral and history documents retrieved and OASIS, plans of care and notes written back where the agency and the EHR permit.

Murphi is also embedded under other names inside home health EHRs. With about 200 agencies live, write back runs in production, though no agency is tied to a named EHR and no marketplace listing is cited.

Data stays in the United States on AWS and Google Cloud, Murphi states, under a zero trust design in which every request is authenticated, authorized and encrypted and access follows least privilege. The platform runs as Murphi's own app, as MurphiConnect, or embedded inside home health EHRs.

How agencies' data is separated from one another is not described, and whether model processing through the AI platforms it names also stays in US regions is not stated. Those are the open pieces for an agency that must answer where its patients' recordings go.

No price is listed, and agencies start with a demo. The selling model is visible from the product structure: ambient documentation, documentation and revenue integrity, care team engagement and patient payments are separate modules, and EHR companies can embed Murphi under their own brand through an API, single sign on or an iFrame, with Murphi saying they carry no token or infrastructure cost.

Coding, billing and consulting firms and accreditation bodies are listed as buyers alongside agencies. Whether an agency pays per clinician, per visit, per OASIS or per agency, and how the embedded version is priced for an EHR's own customers, is not stated.

Home health and hospice are the core setting, and the product is built around their paperwork: OASIS, HOPE, plans of care, PDGM, hospice certifications and discipline specific notes for nursing, physical, occupational and speech therapy, aides, chaplains and social workers. About 200 agencies use it, and many are named.

MurphiConnect stretches the same platform to behavioral health, public health and corrections, primary care, specialties and health systems, and no deployment in those settings is named alongside the home health product. Validation sits in home health and hospice; the wider settings rest on the sister product's pitch.