The operating system for caregiving.

The problem, the payment model, and the plan

63 million Americans, nearly one in four adults, are family caregivers. They put in 49.5 billion unpaid hours a year, worth more than $1 trillion, and they do that work with paper notebooks, portals that don't connect, and a group text. The load lands on one person per household, and it is making those people sick: strained spousal caregivers carry a 63% higher four-year mortality risk. No software holds this job. Clinical systems treat the family as a visitor, consumer apps have no clinical pipe, and when the crisis ends, the record dies with whichever app briefly held it.

Pat Health closes that gap. It gives the family caregiver a real system of record, connected to the clinical stack over FHIR: one care recipient, one longitudinal record, one care circle with roles and permissions. The record outlives the crisis. Siblings coordinate without grinding each other down. The clinical team gets pre-visit signal without the family typing anything twice.

The timing case is the CMS GUIDE Model, the first Medicare payment model that pays for the caregiver rather than the patient. 390 participant organizations are contractually required to deliver caregiver support services through June 2032. They collect $150-$390 per member per month in the first six months. The 294 New Track orgs that started July 2025 are picking their technology stacks right now. Pat Health sells them the caregiver-facing platform plus the workflow and reporting surface those obligations demand, at a $60K-$150K annual license plus $25-$60 PMPM.

The work has started. The planning package runs five phases, roughly 475,000 words, 406 cited sources, from market research through engineering architecture, and a working codebase exists with 751 passing tests. Nothing is deployed yet. The next stage needs a completed founding team and a Seed round, and section 15 lays out both.

Unprepared and under-equipped, constantly

PAT stands for personalized at-home healthcare. It was also my grandmother's name. Pat lived with advanced Alzheimer's for years before she passed, and my other grandmother, Sandra, is living with advanced dementia today. This company exists because of the two of them, and because of what caring for them did to the people I love.

My parents and my aunt took care of Pat. I watched them struggle to understand the system they had been dropped into, struggle to keep track of medications and appointments and what changed week to week, struggle to know how to help her or whether anything they were doing was working. They felt unprepared and under-equipped, constantly. Not because they lacked devotion or ability, but because nothing taught them the disease, nothing organized the work, and everything that might have helped sat behind a portal login or a phone tree. Sandra's household is living the same thing right now.

The tools they had were the ones every American family gets: paper notebooks, sticky notes, three portals that don't talk to each other, a group text that never kept up with the person at the kitchen table. I sat at those tables and watched the medications the aide gave drift from the ones the family thought she gave. Twice.

The numbers say those two households are the norm. 63 million Americans are family caregivers, nearly one in four adults. The primary caregiver persona in our research, a 52-year-old daughter coordinating her mother's dementia care, spends 25-30 hours a week on it alongside a full-time job and pays roughly $7,200 a year out of pocket. Strained spousal caregivers carry a 63% higher four-year mortality risk than non-caregivers (Schulz and Beach, JAMA 1999). 7.4 million Americans over 65 have Alzheimer's now; 13.8 million will by 2060.

The market has left this job unstaffed. Hospitals get an EHR, agencies get scheduling software, and the person running a household's care gets a notebook. Section 07 walks the full landscape; the short version is that no product makes the family's record and the family's work its center of gravity.

A 24-month window with named accounts

Medicare started paying for the caregiver

The CMS GUIDE Model launched July 1, 2024 and runs eight years, through June 2032. It pays participant organizations $150-$390 PMPM for the first six months per beneficiary and $65-$220 after, tiered by complexity and by whether an unpaid caregiver is present, plus up to $2,500 a year in respite per beneficiary. In exchange, participants must run a 24/7 helpline, assessments, care plans, caregiver education, respite coordination, and CMS quality reporting. At 1,000 members that is $2M-$4.7M of annualized first-period revenue per org. Most participants have no purpose-built software for the caregiver side of those obligations.

The buyer list is published and finite. 96 Established Track orgs have been live since July 2024, 294 New Track orgs since July 2025, and the New Track cohort is choosing vendors in 2025-2026. That is a 24-month greenfield window with named accounts.

The clinical data finally has a legal door

FHIR R4 with US Core is mandatory for certified health IT, the CMS Patient Access API took effect January 1, 2026, and information blocking now carries penalties up to $1M per violation. A patient, or her proxy, can authorize a read of her own chart from Epic through SMART on FHIR without any EHR vendor partnership. Our MVP interop path needs no certification and no BD deal.

The graveyard taught the lesson

CareZone raised big, sold to Walmart for ~$200M, and was shut down in 2023; families lost their data. Papa raised $257M and lost dozens of contracts to trust failures. Every company in the consumer caregiving graveyard ran the same play, App Store CAC against exhausted buyers with no payer behind them. We start where reimbursement already exists and we treat the record as a covenant. Nobody in the graveyard had either.

Seven primitives, nothing else

The MVP ships seven primitives. Anything not on this list does not ship in year one.

1. The longitudinal record. Event-sourced, append-only, exportable from day one in JSON, PDF, FHIR bundle, and printable-binder formats.
2. The inner-circle care team. The care circle as a first-class object with roles and permissions, task-first instead of chat-first.
3. The dementia clinical model. GDS/FAST staging, CDR-SB, IADL-then-ADL tracking, DICE behavior log, gait speed.
4. The FHIR read pipeline. Patient-mediated SMART on FHIR read from Epic MyChart first, translated into views a family can read.
5. The GUIDE workflow surface. Assessments, care plans, ZBI-12 caregiver strain screening, respite coordination, CMS quality reporting.
6. Clinical time tracking. CCM/PCM/99483 billing capture with an OIG-defensible audit trail, so the platform pays for itself inside the org.
7. The 2am ER binder. The whole clinical picture as a phone-shaped handoff, shareable by QR or SMS, offline-capable, with per-share expiration.

The AI stance

Three AI features ship in the MVP, and only three: voice-to-text capture, a structuring worker, and the pre-visit summary. The operating philosophy is written down. AI is proactive in surfacing, never proactive in acting. Draft everything, send nothing. Every output is labeled, provenance-attributed, and reversible. Every feature ships with a filed FDA CDS carve-out memo before production, and no clinical-facing AI feature ships without Clinical Advisory Board approval. No risk scores to families. No diagnostic AI, ever. We conform to HTI-1 DSI transparency voluntarily because it reads as a buyer signal in every GUIDE sales conversation.

The org pays, the family never does

B2B2C. The GUIDE participant pays; families never do in year one. Pricing has two parts, an annual organization license plus a per-member-per-month fee that scales with their CMS reimbursement.

Direct-to-caregiver is part of the plan, not a rejection. We market to caregivers on the product's own promise, an easier way to track and care for a loved one at home, and that demand works twice: caregivers pull their care organizations toward Pat, and families who start inside a GUIDE contract keep the record on the continuity plan. What we refuse is the graveyard's version of D2C, buying App Store installs from exhausted buyers with no payer behind the product. Reimbursed channels fund the company; direct caregiver demand compounds it.

We never charge for data export, interoperability, safety features, accessibility, or HIPAA-compliant messaging.

TAM, SAM, and a bottoms-up SOM

The spend is enormous and most of it runs on no software. Home health spending hit $169.4B in 2024, growing 10.2% a year. Dementia care alone costs $409B in 2026 before counting unpaid hours. The software layer on top of that spend, sized three ways, with the derivation shown for each:

The demographic floor keeps rising. 10,000 Americans turn 65 every day, and the 85+ population quadruples between 2000 and 2040. Rock Health flagged caregiver solutions as an emerging investable category in a $14.2B digital health funding year.

Nobody owns both layers

Every competitor sits on one side of a divide. Consumer products treat the family well and the chart not at all. Enterprise products live in the chart and treat the family as a visitor. A newer field of dementia-care and GUIDE-focused vendors is filling in around the model, which validates the wedge; none of them makes the family's longitudinal record the product. The open quadrant, family-focused clinical intelligence, is where Pat Health sits.

Direct: the dementia-care and GUIDE field

This is the field diligence asks about first, so it comes first. These companies prove the GUIDE wedge is real; each one is a care service or a navigation layer, where Pat Health is the record-and-workflow platform any of the 390 can run.

Two structural facts about this field. Every service-led GUIDE company still needs the workflow, reporting, and family-record layer we build, so several of these names are as much channel as competitor. And because the GUIDE payment ends in June 2032, a services business built on it faces a cliff; a system of record that families and orgs run on does not.

Consumer: shallow or dead

Enterprise: the family is invisible

What Pat Health is not

Written to hold up in a sales call, a board meeting, and a hiring loop. Not a standalone medication reminder app. Not chat-first. Not an outpatient scribe. Not home-health agency SaaS (we integrate with AlayaCare, HHAeXchange, WellSky, and Sandata rather than replacing them). Not a paid-acquisition consumer app; we market directly to caregivers, and reimbursed channels carry the early revenue. Not a wearable hardware company. Never a diagnostic AI product, never ad-supported, never a de-identified data marketplace.

Founder-led, named-account, proof before scale

Year one is a direct, named-account motion. The buyer list is the published GUIDE participant roster, 390 organizations, and the priority slice is the 294 New Track orgs choosing their technology stacks in 2025-2026. Sales stays founder-led through the first six deals or twelve months, whichever comes first; the open-questions log sets that boundary so founder-led selling ends on a date instead of a habit.

Land

One to two lighthouse design partnerships open the market. Each lighthouse gets the founder on-site and a contractual commitment that the first care recipient is live within 60 days of signature, because the risk register names deployment stall as a top-five company risk. The sales conversation runs on three proof points: the pre-visit summary the clinical team gets without asking the family to type, the CCM/PCM/99483 time-tracking module that pays for the platform inside the org's existing billing, and the D30 kill switch, which tells a buyer we will not sell them churnware.

Expand

Inside a signed contract, growth is measured in circles, with a target of 60% of care circles holding two or more active members. Across the market, the channel ladder from section 05 takes over: agencies and memory care in year two, MA plans and health systems in year three, with GUIDE logos and published retention numbers as the reference base.

Alongside the named-account motion, we market directly to caregivers on the promise itself: an easier way to track and care for your loved one at home. That demand pulls participating orgs toward Pat, seeds circles that outlast any single contract, and feeds the continuity plan, without the paid-install spend that killed the consumer graveyard.

Ranges we will defend, not points we invented

Every figure below is a documented range from the planning package's roadmap gates and pricing architecture. Nothing is a point estimate, because the honest flag in section 05 applies to all of it: pricing is validated against live buyers in Q3-Q4 2026, and the model firms up behind it.

What the Seed buys

Four covenants, written into every term sheet

These four are corporate covenants. They carry weight on every decision and get written into every financing conversation. They exist because the FTC has already fined GoodRx, BetterHelp, and Cerebral for the ad-tech patterns we refuse, and because CareZone shut down in 2023 and its families lost their records with it.

Growth has a kill switch, on purpose

The wedges stack. GUIDE contracts are the commercial wedge, the pre-visit summary is the clinical wedge, and voice-first capture is the design and workforce wedge. Each one makes the next easier to land.

When something breaks, it fails closed

Every observation, capture, clinical read, and family action is a timestamped event on an append-only log in Postgres. Every screen (pre-visit summary, shift handoff, quality report) is a view over that log. Web is Next.js, mobile is React Native offline-first, the API is TypeScript on AWS. EHR data arrives because the patient or her proxy authorizes a read of her own chart via SMART on FHIR, so no vendor certification is required. Permissions are care-circle-native. Every query is scoped to one care recipient, roles get scoped bundles, and row-level security in the database backstops application bugs. When something breaks, the system fails closed.

HIPAA-first from day one means no PHI in logs, error traces, analytics, URLs, or push payloads. No third-party analytics touch PHI. CI runs on synthetic fixtures only. Every PHI-touching vendor signs a BAA; Anthropic and Deepgram both run with BAAs and zero retention.

Compliance is priced into the model, including insurance premiums that scale from roughly $120-200K a year at Seed to $850K-1.6M at Series B. A PHI leak is the one failure this company does not recover from. The architecture is shaped around that sentence.

Built, tested, unreachable

This section is the unvarnished version, straight from the build-status doc: a large, real, well-tested application exists on a laptop and in CI, and a person outside this team cannot reach any of it.

Two housekeeping facts diligence would find anyway. The codebase diverged from the architecture plan in places (Hono+tRPC instead of Fastify, a custom permission engine instead of OpenFGA, no pgvector yet); each divergence is documented and needs a deliberate keep-or-migrate call. And the planning package still says implementation has not started. That was true when it was written and is now false in our favor. Reconciling the two documents is queued because the package is the artifact investors read.

One cultural note matters more than either. We spent sprint 15 removing demo surfaces that faked data, including hardcoded charts, and the standing rule since is that no surface is ever described as shipped when it is a stub. That rule predates any outside money, and this document was written under it.

What can kill it

The register tracks 72 risks. Six of the top ten stay high even after mitigation. Here are the ones that matter most, in the register's own framing.

What it takes from here

Dakota Milner is the founder. The planning package, the codebase, and this document are the work to date. The next stage needs the founding team the package specifies: a technical co-founder or founding CTO, Head of Product, Head of Design, Medical Director, Head of Compliance, and the first engineers, with a five-person Clinical Advisory Board seated by Q2 of year one so AI features have a release path.

The first 90 days of engineering

• Authentication and identity, then first deployment to AWS. The plan of record is WorkOS for enterprise SSO plus Clerk for consumers, with MFA and passkeys mandatory.
• The SMART on FHIR read path against Epic MyChart. The MVP cannot ship without this one integration.
• Voice capture end to end with Deepgram under BAA. Two-tap voice entry is the product's primary input.
• The keep-or-migrate calls on the documented architecture divergences, plus SOC 2 groundwork so the Series A gate never becomes the blocker.

The next two quarters of proof

• Five priced GUIDE conversations and two LOIs (Q3-Q4 2026), so pricing stops being an assumption.
• The first two named pilot participants chosen from the 390. This currently blocks everything downstream.
• A Seed close targeted for end of Q3 or early Q4 2026, on financial-only terms, funding an 18-24 month runway.

The case this document makes is checkable at every step: the problem is measured, the payment model is law through 2032, the buyer list is published, the pricing has named validation deadlines, and the codebase has a test suite anyone can run. The argument trail behind every load-bearing decision is preserved in the planning package for anyone who wants to challenge one.