Open pipes.
Signed answers.
The technology layer under every site in the network.
Every healthcare AI company needs the same plumbing and the same missing piece. The plumbing — FHIR, HIPAA tooling, identity, PROMs — is open source under Apache 2.0. The missing piece is a licensed physician who will put their name on what your model produced. That is the part that cannot be automated, and the part we are built around.
- Open code
- Apache 2.0
- Public data API
- No key, no signup
- The signature
- A physician signs or declines
curl 'https://solvinghealth.com/api/v1/clearances/search?q=knee&limit=3'- One real request to the keyless public data API. Nothing is stored.
SolvingHealth in 15 seconds
- Rebuilding the same pipes again? Every health AI team needs FHIR, identity and PROMs first.
- One layer under every site. One assistant, one data API, client files with nothing to install.
- Every answer names its source. FDA, NPI and CMS records, with the source URL and fetch time.
- Make a real call. No key. Knee clearances or Boulder hospitals: solvinghealth.com/#try
- SolvingHealth. The technology layer under the network. solvinghealth.com
Make a real call. No key, no signup.
Pick a public dataset. The request below runs live against the same keyless API every site in the network uses, and every answer names the exact source it came from. Public federal data only — no patient information goes in or comes out.
curl https://solvinghealth.com/api/v1/…- Pick a dataset to run it.
Raw JSON
One routing layer. Every physician in the loop.
A patient asks. A model answers. Somewhere between those two events, a licensed human has to own the decision — otherwise nothing downstream is billable, defensible, or fair to the person receiving it.
Consented data
Records, PROMs, device signal, and what the person tells you themselves — over open connectors, never used for training.
Any model
Yours, ours, or whatever your team already uses. We are not in the model business and have no interest in being.
Draft until a physician signs it
Nothing clinically meaningful passes unsigned. A named, specialty-matched physician signs or declines.
A determination with a receipt anyone can check
NPI-bound, timestamped, and hash-anchored — checkable by an auditor without trusting anyone in the chain.
One layer underneath. Every site in the network runs on it.
SolvingHealth is the technology layer the rest of the network is built on. Each site owns its audience and its voice; underneath, they share the same parts: one design system, one assistant, one public data API, one gate that checks the work, one place a physician signs, and one ledger that proves it later.
SurgeonValue
The surgical practice's AI front office, and the clearest picture of what the layer does. Agents draft the prior authorisation, the coding, the PROM record and the remote-monitoring documentation. A deterministic check grades the codes. The surgeon's name stays on the output, and the practice bills under its own NPI.
Visit surgeonvalue.com →- Prior authorisationDraft
- Coding, graded by a deterministic checkDraft
- PROM recordDraft
- Remote-monitoring documentationDraft
ClinicalSwipe
Where a physician signs. Review AI-generated clinical output in your own specialty, on your own licence. Declining counts as much as signing.
Altru
The clinical side of the network: the entity behind the family-facing work, where a physician reviews a case before anything is signed.
co-op.care
Worker-owned companion care. Where a family's care already qualifies, a physician-signed letter of medical necessity documents it — and gives the honest no when it does not.
And the rest of the network: JointClass and JointCoach and SurgicalTrack for the patient side of the joint episode, MSKvalue for the hospital, HarnessHealth for the gate, Hashcare for proof, and the condition-entry sites where people arrive already asking the first question. See every site →
Connect to the network. Open.
Open-source pipes. Every healthcare AI company needs them, none of them differentiate you, and gating them would only slow the thing we actually want — more output arriving at the gate.
FHIR & EHR connectors →
Read and write the formats the industry already speaks: a FHIR R4 client and an Omaha-to-FHIR mapping. Production EHR integration is designed, not built; the one loop running today is a sandbox.
HIPAA tooling →
Transport, audit logging, encryption and PHI-detection helpers, built to be inspected.
PROM collection →
Structure and scoring for KOOS JR, HOOS JR, PROMIS-10, ODI, NDI and QuickDASH. The questionnaires belong to their owners; you load each one's licensed wording.
MCP server →
Tools for agents over the public data API: NPI records, 510(k) clearances and recalls, drug labels, trials and CMS hospitals, each answer carrying its source URL.
What a physician owns, and why it is the only honest line.
Routing and the specialty gate →
A case reaches a physician licensed in the patient's state and matched to the specialty — and is held, not reassigned, when no match exists.
Versioned criteria →
Published, versioned and hash-anchored — and no condition is offered until the medical director has signed its criteria. Adding a condition is a signature, not a product decision.
Attestation and receipts →
Every determination bound to an NPI and anchored as a PHI-free fingerprint on a public ledger.
A no counts as much as a yes →
A determination is never tied to a percentage of an approved sale, a claim, or a saving. That single decision is what makes every determination in the network defensible.
Mathematical verification
Change one character. Watch the proof fail.
This is a sample determination record, not a real attestation — no case, patient or physician behind it. The fingerprint on the right was taken when it was “signed”. Edit anything in the record and your browser recomputes the SHA-256 on the spot.
computing…
computing…
In the network, that fingerprint — never the record itself, and never patient data — is what goes on the public ledger, so an auditor can check a decision without trusting anyone in the chain. How verification works →
The data ventures — provider graph, device clearance and recalls, drug labels, trials — run on the same keyless gateway and the same rule: the pipes are open, the judgement is signed.
Interpretation got cheap. Accountability did not.
When the cost of producing a clinical judgement collapses, the industry built on that cost collapses with it — the intermediaries, the queues, the second opinions nobody asked for. What survives is a body in a home, signal, hands nearby, a licensed human who owns the decision, and proof anyone can check. SolvingHealth is the routing layer for exactly that: it keeps the cheap part open, and puts a physician's name on the part that has to be owned.
Talk to us.
We work with health systems, payers, device companies, and AI developers who want physician-governed AI infrastructure they don't have to build from scratch.