The model gateway healthcare builders actually need.

One OpenAI-compatible endpoint. A physician-curated catalog of frontier and healthcare-tuned models, ranked by evidence on real clinical tasks — not marketing. And the part everyone else skips: a compliance envelope built toward one BAA that covers your whole model stack.

Request early access Become a design partner

Not for PHI — yet. That's the point of the name. PHIGate is in its pre-launch build phase. Nothing you send through any PHIGate surface today may contain protected health information, and our launch tier will enforce that with fail-closed screening. The BAA-grade tier — the gate PHI must pass through — is what we are building toward, and it opens only when the agreements and controls that make it legal exist. We will never imply otherwise.

Why this exists

Every healthcare app team hits the same wall: which model can we legally and defensibly use for this clinical task, and how do we get the paperwork without four enterprise sales cycles? Generic model routers don't answer either question. As of July 15, 2026, a search for the literal term “medical” across the largest public model router's catalog returned zero results in every modality — and we found no curated healthcare catalog anywhere that pairs model-specific compliance metadata with physician-reviewed, task-level evaluations. That's the product.

Evidence, not vibes

Which model drafts a patient message at a 6th-grade reading level without inventing facts? Which one interprets a lab panel most faithfully? We test, a physician reviews against a published rubric, and the results are versioned and dated.

Compliance metadata per model

Retention policy, training-use posture, region, and BAA availability at the configuration level — with sources and observed dates. Availability listed here never means coverage for your deployment; we say so on every entry.

One integration

OpenAI-compatible. Change the base URL and your SDK, framework, and agents keep working — with per-key policy, spend limits, and honest per-request cost accounting.

Directory, evaluations, and early-access waitlist are being staged now. Every claim on this page carries its date; anything we can't evidence, we don't say.

Who it's for

US digital-health teams building patient- or clinician-facing AI features — intake, messaging, summarization, coding assist, navigation — who are close enough to real data to be worried about it. If that's you, the design-partner seat is deliberately small, paid, and comes with a direct line to the physician doing the evals.