InsureWise
An educational, decision-support web app for U.S. car and home insurance.
InsureWise is not an insurer, agent, broker or quote engine. See
CLAUDE.md for the non-negotiable product rules and
docs/ for requirements, architecture, calculations and data model.
Status: all 8 MVP roadmap groups are implemented — car/home profiles,
scenario and comparison tools, review findings, guest persistence,
authentication, a grounded AI assistant, and the reference-content
infrastructure. The app is functionally complete; what's still open is
content and deployment infrastructure, not features:
- No real Supabase or Anthropic project is connected (
.env.example has
placeholders only) — auth, persistence-sync and the AI assistant degrade
gracefully to a "not configured" state without them.
docs/RESEARCH-SOURCES.md has no fact
verified to V2 yet, so the coverage catalog, glossary, rating factors and
risk-checklist content all ship empty on purpose — the accessors and UI
are built and tested, waiting on real sourced content.- Supabase migrations/RLS SQL are deferred to the documented Phase 7.
- Playwright (e2e) and axe (accessibility) are not wired up yet.
Stack
Next.js (App Router) · TypeScript · Tailwind CSS · shadcn/ui · Vitest ·
Supabase (Postgres, Auth, RLS) · Anthropic API (behind an adapter)
Getting started
npm install
npm run dev
Open http://localhost:3000.
Commands
| Command | Description |
|---|
npm run dev | Start the Next.js dev server |
npm run build | Production build |
npm run start | Serve the production build |
npm run lint | ESLint, including the module-boundary rules |
npm run type-check | TypeScript, no emit |
npm test | Run unit/property tests once (Vitest) |
npm run test:watch | Run tests in watch mode |
A seed of the AI evaluation harness runs as part of npm test
(tests/ai-evals/, scripted against a fake model provider — no live
Anthropic account needed). End-to-end (Playwright) and accessibility (axe)
checks are not wired up yet.
What's implemented
- Car & Home profiles (
/car, /home) — vehicle/driver/property
intake, coverage breakdown, line-level premium composition, home coverage
ratios, all with provenance labels (Entered/Calculated/Illustrative).
- Scenarios (
/scenarios) — deductible trade-off, break-even, and
N-year cost views.
- Compare (
/compare) — baseline vs. up to three policies, row-level
differences, comparability warnings, no ranking.
- Review (
/review) — rule-driven review findings, "questions to ask a
professional," and an AI assistant grounded in that computed data.
- Guest persistence — IndexedDB-backed, same Zod schemas as the server
model; works fully without signing in.
- Authentication (
/sign-in) — Supabase Auth adapter (magic link,
Google) with guest-to-account migration, gracefully inert without a live
project.
- AI Assistant (
/api/v1/assistant) — fact-packet grounding, read-only
lookup tools, output validation (schema/citation/numeric-grounding/
language-policy), rate limiting; gracefully inert without a live
Anthropic account.
- Reference layer (
src/reference) — sourced citation registry,
jurisdiction support, and a V2-only verification gate for coverage
definitions, glossary terms, rating factors and checklist items.
Repository layout
See docs/ARCHITECTURE.md §11 for the authoritative layout and §4 for the
module boundaries (enforced by the ESLint config). In short:
src/schemas — Zod schemas and derived types (the dependency root)src/engine — pure calculation and rules engine, no I/Osrc/reference — typed accessors over verified reference data (ships
empty content arrays until docs/RESEARCH-SOURCES.md reaches V2)src/providers — PolicyProvider implementations (manual entry, sample)src/ai — fact-packet builder, prompts, model adapter, output validatorsrc/lib — Supabase/auth clients, guest storage, rate limitingsrc/features — feature UI and hookssrc/components — design-system primitives (shadcn/ui lives in components/ui)docs/ — product and engineering documentationreference-data/ — the offline verified-fact pipeline (workbook → exports → scripts)supabase/ — migrations, seed data, RLS policies (empty — deferred to Phase 7)tests/ — unit, property, ai-evals (e2e and a11y not set up yet)
Documentation map
