Clinical Inbox & Messaging AI

Keona Health is the closest thing in this index to Clearstep's opposite number, and the two make a useful pair. Both route patients using the same family of licensed telephone triage protocols. Clearstep puts a patient in front of them directly. Keona puts a nurse or a call centre agent in front of them, and only recently began letting a model take the call instead. The company qualifies for this index on the newer half of that sentence, and the older half is what it actually sells. CareDesk, previously Health Desk, is a healthcare customer relationship management and patient access platform built for contact centre work: intelligent call routing, adaptive queues, scheduling, prescription refills, referral management, health information calls and care management, unified across phone, text, email, chat and web. Underneath the triage module sits Schmitt Thompson Clinical Content, licensed rather than written, covering more than 600 adult and paediatric topics, reviewed annually by a panel including triage nurses, call centre medical directors, primary care and emergency physicians, and used in the great majority of North American medical triage call centres. Practices can add their own protocols alongside it. The artificial intelligence arrived on top of that architecture. Kara is an assistant that handles repetitive calls autonomously around the clock and transfers to staff with full context when a situation needs a person. CareDesk Flo is the phone agent, described as using intent checks and fallback logic as guardrails, with uncertain cases handed to a human and edge cases flagged for nurse review. This is a 15 year old protocol engine with a voice layer on it, which is a different proposition from a model native product and should be understood as one. The safety design is the strongest part of the record. Escalation levels are defined and triggered by risk thresholds, running to on call physician contact, home care, urgent follow up or emergency referral. Protocols adapt to age, comorbidities, medications and a practice's own standard operating procedures. Every triage decision generates structured documentation aligned to the clinical reasoning pathway, explicitly to make the decision defensible afterwards, and supervisors get visibility into protocol adherence. Telephone triage doctrine including the duty to warn is built into the workflow rather than left to the nurse. A spin out of the University of North Carolina at Chapel Hill, still headquartered there, founded around 2010 and grown deliberately small at roughly 23 employees. It began on a $450,000 grant from the National Institutes of Health and has raised about $10.1M in total, with Riverside Acceleration Capital named as an investor. Founder accounts differ across sources: the university's own 2014 announcement credits Oakkar Oakkar, now chief executive, with professor Javed Mostafa, while company and database sources name Oakkar Oakkar with Stephen Dean, now chief operating officer. Integration is claimed in real time with Epic, Cerner, Allscripts, athenahealth and Greenway, writing encounter notes back to the chart, and the platform is hosted on Microsoft Azure. One thing a reader should weigh. The safety architecture is unusually well described and the performance claims around it are not verifiable, including a claim of independent audits showing better than 95 percent alignment with human decisions that names no auditor.

Clearstep sits upstream of the clinical inbox rather than inside it. Where Affineon, Droxi and Elaborate work the messages once they have arrived, Clearstep tries to stop a share of them arriving at all, by meeting the patient at the point of first symptom and routing them to a care setting before a call or a portal message is generated. The product is a conversational triage and navigation layer. A patient describes a symptom in free text, natural language processing interprets it, the system asks the clinically relevant follow up questions, determines acuity and urgency, and routes to an endpoint, with the company claiming completion in under five minutes and roughly ten fewer questions than competing tools. Around that sit intake, insurance verification, scheduling handoff, referral routing and care navigation, so the output is a booked appointment rather than a recommendation. A separate capacity optimisation suite covers predictive demand and load balancing. Deployment channels include a website widget, a patient portal embed, SMS follow up after visits, voice, and augmentation of an existing nurse or administrative call centre. The clinical logic is licensed rather than invented, and the company names its source, which is unusual. Triage content derives from the Schmitt protocols, the telephone triage standard authored by Dr Barton Schmitt and used in most nurse call centres in the United States, and Clearstep claims to be the only chat based self triage tool built on that content. That matters more than a technical credential would, because it means the acuity determination rests on protocols a nurse reviewer already knows, and the models supply the interface and the routing rather than the clinical judgement. The customer roster is the strongest signal on this record and is disproportionate to the company's size. Named organisations include Ochsner Health, Mount Sinai, Tufts Medicine, Novant Health, HCA, CVS Health, Duly Health and Care, and the Defense Health Agency. Ochsner has described the deployment in some operational detail through its own clinical leadership. Founded in 2018 in Chicago by Adeel Malik, previously an Accenture healthcare consultant, and Bilal Naved, an MD PhD trained at Northwestern. Funding totals are reported inconsistently across sources, at roughly $7M in one database and roughly $17.4M in another, and headcount is reported at 29. One thing a reader should hold in mind when comparing grades. Several axes here sit at the floor, and in every case that reflects an absence of published disclosure rather than a finding about the product. The company's public surface is a marketing site with a privacy policy and terms of use and no security page, trust centre, certification claim, pricing or model documentation. For a vendor whose software tells patients whether to go to an emergency department, that gap is itself the most consequential fact on the record.

Luminai automates the administrative work health systems still do by hand, end to end rather than task by task. Healthcare trained models structure unstructured documents and fragmented data, workflow automation carries the resulting actions across disconnected systems, and human in the loop validation gates the steps that need a person. Named workflows span referral intake, provider inbox automation, patient registration, order and referral processing, pharmacy renewals, payor contract management, denial appeals and underpayment recovery, grouped commercially under patient access, revenue cycle and compliance. The integration approach is the unusual part and comes from the company's earlier life as a cross industry automation platform. Rather than requiring an interface build, the software operates the systems a team already uses, which means it can reach applications that expose no usable programming interface. That is how the platform gets into the corners of health system operations where the data lives in a portal nobody can integrate with, and it is also why standards based interoperability is not the story here. The security and deployment posture is the strongest element of the record and is unusual for a company at this stage. Three deployment models are offered: on premise, inside the customer's own virtual private cloud, or vendor managed. Alongside that the company describes encrypted data, isolated execution and, notably, customer owned credentials, meaning the automation acts using the customer's own system credentials rather than the vendor holding them. Alignment is claimed against United States health privacy law, service organisation controls and European data protection law. A 38 million dollar Series B closed in April 2026, led by Peak XV Partners with new investor Define Ventures and continued backing from General Catalyst and Y Combinator, bringing total capital to 60 million. The team draws from Palantir, Google, Coinbase and Brex on the technology side and from Epic and Banner Health on the operator side. Based in San Francisco. Two cautions. The company states it is trusted by large health systems and names none of them publicly, so no customer, deployment scale or outcome could be verified. And several third party directory pages carry precise sounding figures for return on investment and time to value that trace to no primary source and bear the marks of generated content; none of them informed this record.

Switchboard, MD is an Atlanta company founded in 2022 that triages, routes and resolves the high volume patient communication arriving at a healthcare organisation across the portal inbox, live calls, voicemail, fax, text, email and web forms. It began with a specific problem its founder had personally: Dr Blake Anderson, a practising internal medicine physician at Emory Healthcare and previously chief health informatics officer at the Atlanta Veterans Affairs facility, applied a computer science background to build a language model that triages patient messages inside the medical record and routes each one to the correct inbox, removing roughly 40 percent of the messages that had been reaching clinicians. Those results at Emory were published in the New England Journal of Medicine artificial intelligence journal in 2025, which is the only peer reviewed publication located for any vendor across two contact centre source lists. The platform now adds conversational voice handling, multi agent orchestration, autonomous scheduling for high volume visit types, appointment reminders, wayfinding, and referral capture that classifies an inbound referral, creates or updates the patient record, routes it to a provider and drives booking. Models summarise, classify, prioritise, flag clinical urgency and assess caller sentiment. It is deployable on premise or in the cloud, states that the customer's data remains the customer's, and describes the platform as HIPAA compliant and certified to SOC 2 Type 2. It joined the athenahealth marketplace as a solution partner in July 2026. Named customers include Kirby Medical Center, a critical access hospital in Illinois, and Georgia Vision Institute.

Art is Epic's generative AI feature inside the MyChart In Basket, drafting replies to patient medical advice messages. The name matters and is frequently miswritten: Art stands for Augmented Response Technology, not Automated Response Technology. Epic's own customer documentation and every peer reviewed study of it use Augmented, and the distinction is the product's central design claim rather than a naming detail. Art never sends anything. It presents a draft; the clinician reviews it, revises it, or discards it and writes from scratch, and authorship of what reaches the patient stays with the clinician. The draft is generated by a large language model, OpenAI's GPT 4 family delivered through Epic's partnership with Microsoft, reading the patient's latest message together with relevant chart context such as current prescriptions and recent results, plus prompts configured by the individual health system. Scale separates this record from everything else in the category. Epic reported the feature in use at roughly 150 health systems and medical groups generating about 1 million drafts per month, and named deployments include Mayo Clinic, which began implementing in early 2023 and extended access across its nursing workforce, and NYU Langone. It is also the only product in this category with independent peer reviewed evidence, none of it commissioned by Epic. That literature is unusually candid about the failure modes, including clinicians reporting that drafts sometimes contained incorrect or inappropriate content and that they too often suggested an appointment. This record is scoped to Art specifically and is not an assessment of Epic Systems or of any other Epic AI feature.

Elaborate sells inbox infrastructure to health systems and physician groups, positioned as curing the in basket, though results delivery is where the company began. It was founded in 2021 in New York by Nicole Bocskocsky, previously at Oscar and Parsley Health, and raised a 10 million dollar seed round in January 2023 led by Tusk Venture Partners. The platform covers four surfaces: classification tagging and summaries of inbound patient messages for triage, protocol based result summaries sent to the patient at direct release, one click refills with summarised charts, and an analytics module that measures inbox performance over time. The analytics product has no equivalent elsewhere in this category, where vendors generally sell the automation without selling the measurement of it. Two things separate this record from its peers. The company states its mechanism plainly rather than claiming agentic autonomy, describing rules based protocols across more than thirty ambulatory specialties powered by patient clinical data. And its named customers are substantially larger than anything else in the category, including Cedars-Sinai, Community Health Network and Austin Regional Clinic, each carrying a published outcome figure. Elaborate is distributed through both the Epic Showroom and the athenahealth Marketplace, and runs a public trust centre. The company's origin is regulatory rather than technological. The 21st Century Cures Act requires immediate electronic release of test results to patients, and Elaborate was built for the consequence: patients receiving out of range values with no context, and clinicians absorbing the messages that follow.

Droxi Digital Health sells an AI agent that works the clinical inbox for primary care, running natively inside the electronic health record rather than as a separate application. The company is headquartered in Tel Aviv, was founded by Gadi Shenhar and Raveh Ben Simon, and names Drive Capital, Verissimo and AirAngels as backers. The platform ships two products. Droxi Unbox handles the inbox itself across three item types. It auto closes normal laboratory results, including results slightly outside reference ranges where it judges them clinically insignificant, and sends the patient a configurable message. It consolidates refill requests arriving by fax, portal message or e prescription onto a single screen with medication history, last refill date, recent labs and diagnoses attached, flags duplicates, and allows approval or denial of multiple medications at once. And it summarises consult notes and clinical documents, linking every summary point back to the source text in the original. Droxi Unlock is a separate product that runs a practice's Advanced Primary Care Management programme, identifying eligible Medicare patients and handling enrolment and documentation. Droxi publishes three operating figures for Unbox: 38 percent of routine work handled automatically, a 51.9 percent reduction in after hours clinical inbox work, and 5.1 or more hours returned per provider per week. The first of those is unusual and worth noting, because it states the scope of the product's autonomy rather than only its benefit. Named customer organisations include Blue Ridge Health, Valley Health, Salience Health, Primary Care Partners of South Bend, Aspire Rural Health System, Northern Counties Health Care, Stedman-Wade Health Services, Rainbow Pediatric and Interior Health in British Columbia.

Affineon Health sells an AI agent that works the clinical inbox inside the electronic health record, concentrated on two workflows: lab result disposition (Labs Core) and prescription renewals (Rx Renewals). The company was co-founded with AI Fund, Andrew Ng's venture studio, is led by co-founder and CEO David Norris, and raised a 5 million dollar round announced in February 2025. The product is unusually explicit about what it does without a clinician present. Labs Core automatically handles normal, insignificantly abnormal and CC'd results around the clock, and Affineon's own product page describes this as automating the inbox items a provider does not want to see. Results that do require clinical review are summarised and trended rather than auto handled, and the protocols deciding which is which are configurable by each practice, so the organisation rather than the vendor sets the threshold. Rx Renewals tells staff whether a request meets the prescriber's protocol or needs physician review. Affineon also runs Affineon Embed, a white label programme placing its inbox agent inside other companies' products. Commure, indexed separately here, has joined that programme, so inbox capability sold under another brand may be Affineon underneath. Integration is named for athenahealth only, where the product is listed and rated on the marketplace. Pricing is published in full as a per provider per month ladder.

MarianaAI describes itself as a research lab building healthcare AI models, delivered through CARE, which stands for Comprehensive Automation and Revenue Enhancement. Ambient documentation is one module among more than twenty, alongside autonomous medical coding, pre visit charting, order management, CDI compliance, inbox triage with auto drafted patient responses, insurance coverage checks and a MarianaGPT assistant, integrated with a claimed 50 or more EHRs and reachable in the record, in browsers and on mobile. Buyers should read the commercial framing carefully, because the vendor states it plainly: CARE is marketed on simultaneously optimising RAF scores and ICD-10 and CPT capture to lift value based and fee for service revenue by 15 to 20 percent, on capturing up to 100 percent of codes, and on doing the coding with, in its own words, absolutely no human in the loop. That combination of revenue optimisation, risk adjustment coding and advertised full autonomy is the most exposed configuration assessed in this category, and it is why several axes here grade low despite genuine technical breadth.