Safa Health is a venture of Conefia LLC.
From episodic care to continuous care
Safa Health is the clinician-governed AI layer for everything that happens between visits. It answers in the clinic’s name, inside the clinic’s own protocol, and hands anything that needs a licence back to the clinic’s staff.
Patent pending · Clinic validation LOI signed · Built on SAFE-CARE, our open evaluation method
Care comes in visits. Patients live between them
A visit is eighteen minutes. The month around it is forty-three thousand. Patients arrive with questions before it and symptoms after it, and the clinic has no way to be there for either. Messages to clinics have grown 153% since 2020 while visits grew 17%, and most of that work is unpaid and lands on one or two nurses.
So the patient asks a chatbot at 11pm, and the clinic never sees the exchange.
The conversation is already happening. The clinic isn’t in it
The patient wants it
- 32% of US adults already ask an AI chatbot about their health, 65% of them because they want an answer now.
- 41% of those users are preparing for a visit they haven’t had yet.
- 38% never follow up with a clinician about what they were told.
What she gets
- A confident answer with no protocol behind it.
- 51.6% of true emergencies were undertriaged in a clinician-scored evaluation of ChatGPT Health.
- No record, and no duty to escalate.
The clinic needs it
- 50.3% of GLP-1 patients on the obesity indication stop within twelve months.
- 71% of that churn is already on the books by month three, during titration.
- The clinic finds out at a no-show.
One layer. Two customers. One promise
She asks at 11pm.
Safa answers inside the protocol her clinic approved, in her clinic’s name, in her own context.
Anything that needs a licence goes to a human.
Staff see the conversation, the reason and the urgency, not a raw inbox.
Every exchange becomes governed clinic data.
Versioned protocols, timestamped records, owned by the clinic.
Four gates. Uncertainty always resolves to a human
Refusal spine
Hard-blocks dosing, diagnosis and interactions
Triage
Emergency, urgent or routine, with confidence
Disambiguation
Bounded follow-ups, written by clinicians
Fail-safe
Any residual uncertainty escalates to staff
We start where the gap is measurable in ninety days
In a 20,217-patient obesity cohort, 35.8% had stopped their GLP-1 by month three. Most of that is tolerability, not price, and tolerability responds to contact. In a randomised semaglutide pilot in type 2 diabetes, a flexible 16-week titration with dose delays cut withdrawal from gastrointestinal side effects from 19% to 2%, with patients reaching the same final dose. The intervention is attention during titration. No clinic has a channel for it.
Sources: Do et al., JAMA Network Open 2024. Diabetes Care 2025, N=104, randomised, p=0.005.
Filed, built, signed, and honest about the rest
Patent pending
U.S. provisional 64/116,715, filed July 2026. Six control families.
Evaluation instrument built
290 triage cases, 30.7% adversarial, plus 30 multi-turn scenarios, loaded into blinded reviewer and adjudication workbooks.
SAFE-CARE Bench, open
Our evaluation method, published as a specification.
github.com/Conefia/SAFE-CARE-Bench →On the public record with FDA
A comment filed to docket FDA-2026-N-7874 on qualifying the benchmarks used to judge generative AI in medical devices.
Design partner signed
A licensed Texas clinic, no-fee validation, gated on prerequisites.
This is a public evaluation protocol, not a completed result. Safa claims no completed clinical validation, and the paid pilot offer is prospective.
She stops talking before she stops injecting
The patient most likely to quit is the one who goes quiet, and a product that only listens cannot see her leave. Next, Safa reads three signals against the same protocol: what she says, how she engages, and, with her consent, what her CGM or wearable records. What the clinic receives is not a data feed. It is a short ranked list of who needs a call today, with the reason attached.
Wear patterns already predict glycaemic outcomes in published work. Whether they predict discontinuation is exactly what our pilots are built to measure.
Clinics and med spas
Keep the patient you started, and give your nurse a shorter list instead of a fuller inbox. GLP-1 and metabolic programmes first.
Patients
A real answer at 11pm, bounded by your own clinic’s protocol, and a human when it matters. Free to you.
Partners and researchers
Clinician reviewers, health systems and AI safety researchers working on evaluation for agents that talk to patients.
Built by a clinical AI researcher who ships
Safa Health is a venture of Conefia LLC, an AI product company in Morrisville, North Carolina. Yassen Eltayeb is the founder, and the author of SAFE-CARE and SAFE-CARE Bench, the safety framework Safa runs on. The build team is Conefia’s full-stack AI, application and backend engineers, who have shipped clinical AI to production, with a metabolic-health physician advisor.
A physician safety lead in obesity medicine, a nurse informaticist for escalation, and 12 licensed raters for our 290 safety cases.
Get in touch →Care comes in visits. Patients live between them
Tell us which side you are on and we will take it from there.
Or email dev@conefia.com directly.