Health is the governed clinical AI surface your practice already wished lived next to the EHR. It reads patient charts stored in Core, screens each proposed medication against that patient's charted allergies, and routes recommendations through an approve step gated to the physician and practice-admin roles in Core. Prescriptions seal to an append-only ledger, and Core exposes a role-gated export of those entries. Drug-interaction and dosage-range checking need a commercial clinical database that is not licensed yet, charts are developer-seeded rather than synced from an EHR, and no BAA is executed — so this is not a surface for real patient data today.
Same team. Governed. More capable.
Health augments the multidisciplinary clinical staff that runs an independent physician practice. It does not replace your EHR, your prescribing authority, your malpractice carrier, or your clinical judgment. Physicians still decide what gets prescribed. Practice administrators still own staffing and procurement. Patients still have a licensed practitioner making the clinical call.
What Health replaces in practice is the glue tax: prior auth phone calls that eat thirteen hours a week, inbox triage that runs fifty-two minutes a day per physician, alert-fatigue popups that physicians dismiss out of habit, and a paper trail for AI-assisted decisions that does not exist anywhere when payer or CMS asks for it after the fact.
What you should expect: faster diagnostic capture from patient charts once an EHR read path is wired, less administrative tax per encounter, and a defensible record when someone asks who decided what, on what evidence, under what authority. What you should not expect: autonomous diagnosis without physician sign-off, a magic FHIR-universal connector that works on every EHR, or a HIPAA certification that the regulatory framework does not actually issue. Today the diagnostic scan is a deterministic chart-rules stub rather than a clinical model, the architecture is designed for HIPAA-aligned deployment under your BAA chain, and no BAA is executed yet. Whether recommending a specific drug and dose brings this surface under FDA device jurisdiction is an open question for counsel, not one this page answers.
Independent practices are getting squeezed in a measurable way. Sinsky and colleagues found in Annals of Internal Medicine (2016) that for every hour of direct face time with patients, physicians spend nearly two additional hours on EHR and desk work during the clinic day, plus another one to two hours of work after hours. That published time burden has not shrunk. It has compounded as portal message volume and prior authorization complexity have grown.
The economic squeeze runs in parallel. The U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics for May 2024 reports a national median annual wage of roughly $235,930 for Family Medicine Physicians (SOC 29-1215). Medicare reimbursement growth is roughly one to two percent annually, while operating costs at independent practices have grown closer to eleven percent year over year. The math is not survivable indefinitely.
Two structural changes coming in 2026 and 2027 force the buying conversation. The CMS Interoperability and Prior Authorization rule mandates FHIR-based prior authorization APIs across Medicare Advantage, Medicaid, and ACA plans. The 2025 HIPAA Security Rule NPRM proposes that encryption become mandatory rather than addressable, with multi-factor authentication required and AI systems explicitly addressed. Practices either adopt compliant AI infrastructure or absorb reimbursement friction and regulatory exposure.
Hiring another medical assistant or prior auth specialist does not solve this. It lengthens the cost curve and still leaves you without a sealed record when the auditor asks for one. Health is the alternative when you want software-enforced safety gates, human authority preserved in the UI, and ledger-backed evidence on every AI-assisted clinical step.
Primary sources. Sinsky C, Colligan L, Li L, et al. "Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties." Annals of Internal Medicine. 2016; 165(11):753-760: acpjournals.org/doi/10.7326/M16-0961. U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2024, Family Medicine Physicians (SOC 29-1215): bls.gov/oes/current/oes291215.htm.
Specific capacity numbers depend on your specialty mix, your payer mix, how aggressively you route prior auth and inbox triage through agent assistance, and how strict your safety gates are. The magnitude is what we model together once we see a week of real chart traffic and audit needs. Directionally, practices should see relief first on prior authorization and inbox triage, then on diagnostic capture, then on outcome tracking as calibration data accumulates.
Illustrative only. Uses U.S. compensation benchmarks for clinical roles and common hiring friction assumptions. Your geography, specialty mix, and total rewards package will move these bands.
Salary anchors. U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2024. Family Medicine Physicians (SOC 29-1215): bls.gov/oes/current/oes291215.htm. Registered Nurses (SOC 29-1141): bls.gov/oes/current/oes291141.htm. Medical Assistants (SOC 31-9092): bls.gov/oes/current/oes319092.htm. Applied here as one of each role with benefits load approximated inside the range rather than as a precise payroll quote.
There is also a recovered-time dimension that does not appear in the hiring tables. Sinsky's 2016 study and subsequent AMA practice surveys document roughly thirteen hours per week per physician on prior authorization, and roughly fifty-two minutes per day per physician on inbox triage. At independent-practice physician compensation rates, that recovered time is real money and real burnout exposure, not a marketing number.
The cost of augmentation is a fraction of the cost of hiring for the same coverage. The exact delta depends on your specialty mix, your payer mix, your existing EHR vendor, and whether you channel-roll Health under an operator group or acquire it outright. We want to model this with your real inputs, not a generic family-practice calculator.
This is not headcount reduction. It is physicians who spend more of the work week on patients and less on the inbox.
You ramp on your own clinical calendar.
Your data does not train any model
Project0 does not use your patient data, clinical decisions, charting, billing records, or any operational data to train, fine-tune, or improve any model. This is contractual.
Your data stays in your control
Patient stores live in your HIPAA-aligned cloud project with the BAA chain you sign. You hold admin access to those stores. You set retention, residency, and access controls in line with your compliance posture.
Deletion is real
Operational deletes follow your retention policy. Append-only audit semantics mean historic decisions are tamper-evident by design, which is different from silent erase. Patient-record deletion follows your patient-rights process and your records-retention obligations.
Sub-processor transparency
Sub-processor list (cloud hosting, model provider under BAA, observability under BAA or with PHI scrubbed) is documented and extended under your entity at acquisition or engagement.
Access your systems without authorization
Core's role gate scopes clinical reads and mutations to physician, nurse and practice-admin memberships with per-patient scoping. Disclosed gap: the Health BFF still attaches the server credential to unauthenticated requests and per-patient scope is bypassed for API-key actors, so identity enforcement is not yet end-to-end. Break-glass entries are justified and recorded; logging routine PHI views from the console, and mounting the break-glass UI, are remaining work.
Make decisions without physician approval
Physician-in-the-loop is the default and, today, the only authority mode in practice. The system recommends, the physician decides, the system records. Agent-autonomous mode exists in the authority model but is not gated by live calibration data, so nothing runs under it.
Lock you in
Acquisition path is designed so the BFF surface is portable and the Core dependency is concentrated in a thin server client and BFF proxy. FHIR ingest stays on your EHR vendor, not on a proprietary clinical lock-in.
Compete with your practice
We sell governed clinical infrastructure to run independent practices, not a competing care brand that would sit on top of your patient relationships.
Expect a 30 to 60 day pilot on one practice (often one specialty, one EHR, one site) with weekly checkpoints on clinical safety, audit completeness, and physician adoption. A full engagement typically runs about eighteen months under a Founding Partner style agreement so we can sequence BAA chain confirmation, FHIR integration, calibration accumulation, and any Phase 1.5 protocol work with your roadmap. Founding Partner pricing is conversation-led because practice size, specialty mix, and payer mix swing cost more than seat count does.
Integration load on your side is intentionally BFF-first plus FHIR-first: plan for a small number of focused engineering hours per week from your IT or EHR integration lead during the pilot to validate auth, FHIR scopes, practice id wiring, and access logging, then taper as runbooks stabilize. Clinical onboarding is faster than new hires because the dashboard already speaks clinical language and the audit trail is the same one your compliance officer would have asked for anyway.
Strategic buyers, corp dev, and family offices interested in acquiring Health as a going concern: see project0.io/acquire.
Need a custom build on the substrate? Hire the team →
Health is built on Project0 infrastructure.