K Health closed its app. The hospitals kept calling.

A company spends years putting a doctor in your pocket. Then it closes the app. This would usually be the point where the story gets short. At K Health, it is the point where the story gets interesting: the consumer platform ended on December 31, 2025, while care continued through health-system partners. The company had found another place to put its technology - inside institutions patients already knew.

There is a tempting version of this story in which a machine simply learns enough medicine to become a doctor. The more useful version concerns a different problem: an answer that arrives without the next appointment, the previous prescription or the person responsible for following up. Healthcare has plenty of information. Making it travel with the patient is another occupation altogether.

The app ended. The care did not.

K Health began in 2016 with Allon Bloch, Ran Shaul, Israel Roth and Adam Singolda. Bloch’s published origin story starts with his father, who suffered a stroke in 2010. Afterward, the family gained access to a home monitoring device. Bloch wondered why better information had become available only after the catastrophe. His account is a founder’s explanation of motivation, rather than evidence that his eventual product could have prevented that event.

The original proposition was direct: ask a system about your symptoms, learn about comparable cases and connect with a clinician. The US consumer launch followed in 2018. It made a complicated service look like a familiar act - opening an app. For patients who found conventional care expensive or inconvenient, that was an appealing invitation.

The current invitation comes through a partner. K Health’s own transition FAQ directs former users toward health systems and explains how to retrieve their records. It is a consequential distinction for anyone arriving via an old review or a remembered subscription price. Today’s company serves institutional customers; patients encounter its work inside those customers’ care services.

The first network was too small

An early episode supplies the sort of detail that a corporate timeline tends to lose. Designer Anders Sandell describes a New York physician-network pilot launched in summer 2018. The team had conducted research with physicians and built an appointment flow. The pilot nevertheless failed, in his account. App users already had doctors and were reluctant to switch to a small network.

“The size of the network really matters and ours was too small.”Anders Sandell, early K Health designer

That is an uncomfortable lesson for a software company: a usable interface can lead to a destination people do not want. The network was part of the product. A patient was being asked to make a change in trust and access, even if the button merely said to book an appointment. Sandell’s retrospective makes that mismatch unusually plain.

It would be neat to say this failure caused the later hospital strategy. The public account does not establish that. But the later arrangement addresses the same friction. In October 2023, Cedars-Sinai launched Connect using K Health technology, with urgent access, primary-care appointments and visit information integrated into its existing record. The destination now belonged to a system with an established clinical network.

A question with a place to go

The present product covers several stages of a visit. PatientGPT holds a patient-facing conversation informed by the medical record. Dynamic Intake gathers context before the consultation. KMED puts intake and record information together for the provider, with diagnostic recommendations. Perfect Note prepares a draft clinical note. These are company-described capabilities; their names describe different jobs rather than four interchangeable chatbots.

The distinction matters when comparing alternatives. A symptom-assessment tool addresses the initial uncertainty. A virtual clinic addresses access to a provider. K Health’s proposition brings those functions into a health system’s record and care network. For a hospital customer, the potential attraction is a connected service rather than another destination to which it must send patients.

PatientGPT makes that ambition visible. Hartford HealthCare announced the tool in March 2026. In August, Atlantic Health announced a MyChart-integrated version alongside virtual primary care, starting with a gradual beta rollout. Patients could ask an everyday health question, bring relevant record context into the conversation and proceed toward a clinician when appropriate. The important word in that sequence is “proceed.”

“Trust is the whole game in clinical AI.”Ran Shaul, co-founder and chief product officer, August 2026

461 visits, and a less convenient headline

A study published in the Annals of Internal Medicine in April 2025 examined 461 eligible adult visits at Cedars-Sinai Connect. It compared initial AI recommendations with final physician recommendations for selected common urgent complaints. Expert adjudication rated the recommendations equally in approximately 68% of cases, favored the AI in 21% and favored physicians in 11%.

The comparison has boundaries. It was retrospective, at a single center, and included cases with sufficiently confident AI recommendations. Adjudicators knew where recommendations came from. Physicians had access to the AI suggestions, but whether they actually viewed them was uncertain. K Health funded the research. Those details belong beside the result, because they determine what the result means.

The study suggests a role for decision support. It does not establish that an autonomous AI doctor improves patient outcomes. There is also a revealing practical detail in Cedars-Sinai’s account: physicians had to scroll down to see the suggestions during the study period. Even an informed recommendation must survive the interface. A result in a paper and a prompt at the right moment are different achievements.

Capital has a price, too

This has required substantial funding. K Health announced a $132 million Series E in January 2021. A July 2024 equity round brought another $50 million, led by Claure Group with participation from Pablo Legorreta and existing investors. The reported valuation was about $900 million, below the $1.4 billion reported in 2021. Investor enthusiasm did not exempt the company from a change in price.

The August 19, 2026 securities filing makes another distinction necessary. It reports $58,949,942 sold toward a $68,949,885 offering, with the first sale on June 5. The larger number is the target, rather than the amount sold at filing. Treating them as interchangeable would manufacture nearly $10 million with a sentence.

For patients, the relevant cost is different again. Partner care follows the partner’s billing arrangements. Cedars-Sinai Connect says costs vary and that it collects insurance copayments or coinsurance. Historical K Health membership offers should therefore stay in the historical column. A current user needs the eligibility, coverage and charges for the particular service they are entering.

The handoff is the thing to copy

The portable lesson is to examine where a useful answer leads. Before building a new network, establish whether users want to join it. Before advertising a clinical score, specify the task and population behind it. Before adding decision support, check whether the clinician can see it at the moment a decision is made. Each sounds mundane; each changes the product.

The arrangement has conditions. It needs usable records, patient permission, clinical capacity and routes to the appropriate next service. A virtual consultation cannot supply every physical examination or procedure. A record-aware conversation also depends on the record it can access. Local eligibility and partner coverage remain part of access, however elegantly the chat window opens.

K Health’s interesting bet is that established healthcare institutions can supply what a small standalone network struggled to provide: the place where the next step happens. The app that once carried the company’s name has gone. The work now is to make the patient’s information arrive with the patient - and make sure somebody is there to receive both.