Nobody does this on day one.
Most people become caregivers before they feel ready. A diagnosis, a fall, a phone call, and suddenly you are responsible for someone else's health with no roadmap. When my grandmother moved into palliative care and we cared for her at home, the tools that existed assumed we had already figured out the basics. We hadn't. Nobody does on day one.
Meet people where they actually are.
Every other tool in this space is built for ongoing caregiving coordination. They assume you have already gotten organized. Hearth assumes you haven't, and meets you exactly there. You answer a few questions and walk away with something useful. No account required.
The decisions that shaped it.
Stateless by design
The decision to store nothing was deliberate, not a constraint. Someone in crisis isn't ready to trust a platform with their family's health information, and removing that ask removed the barrier between them and something useful.
Four care stages, one tool
Active care, palliative care, hospice, and bereavement are meaningfully different emotional and practical contexts. The questionnaire, output, timeline, and emotional acknowledgment all adapt based on which stage the user selects. This required significant system prompt engineering to handle those differences without building four separate products.
Short path, optional depth
Every question has a required core and optional expansion. The shortest path produces a useful brief. More detail produces a richer one. Nothing is ever forced. Caregivers in crisis have limited cognitive bandwidth and the product had to respect that before anything else.
Accessibility as the core constraint
The target user is often older, stressed, and on a phone: minimum 18px body text, 48px touch targets, no time limits, visible labels on every field, one primary action per screen, WCAG 2.2 AA compliance, and printable output built into the spec from day one.
The emotional support page is a first-class output
The "For the caregiver" section is not an afterthought. It is grounded in published research on anticipatory grief and caregiver burnout. The system prompt was built from two weeks of reading primary research. The page includes explicit language that it is not a clinical assessment, because that distinction matters to the person reading it.
Three frameworks, evaluated before a line of code was written.
HIPAA applicability was assessed and documented before building. Users are guided not to enter anything personally identifying because the tool was designed so it would never need it.
The FTC Health Breach Notification Rule applies to vendors of personal health records. Because Hearth stores nothing, there is nothing to breach.
Hearth falls clearly within FDA enforcement discretion for software that helps patients and caregivers organize health information. The AI formats and organizes; it never interprets, diagnoses, or makes clinical recommendations, and three disclosure surfaces on every output page make that clear.
The Anthropic API key lives exclusively in a Vercel environment variable, never exposed to the client, with all Claude calls routed through a serverless function acting as a stateless proxy. User responses are never logged anywhere.
When caregiving changes, the brief should too.
A lightweight follow-up brief for when the care stage changes. Right now Hearth is a one-time tool. The natural next version recognizes that caregiving is not static and offers a check-in that asks what has changed, without storing anything from the previous session.
Three things that shaped every decision.
Start with who is being left out. The person who became a caregiver yesterday is not served by any existing tool. That is the person Hearth is for.
The right metric here is not retention. It is whether the person who used it felt less overwhelmed walking away than they did walking in.
Compliance is a design input, not a legal checkbox. Every privacy decision was made before the first line of code. That is the only way it works.