ADR 0003: VAPID self-hosted today, DeviceToken model agnostic to
platform so iOS/Android (Expo/APNs/FCM) plug in as new senders later.
Backend (apps/api/src/modules/notifications)
- sender.ts: NotificationSender interface, WebPushSender via VAPID
- notifications.routes.ts: GET /me/notifications/config, POST/DELETE
/me/device-tokens, PATCH /me/notifications/settings, POST /me/notifications/test
- scheduler.ts: node-cron Mon 09:00 Europe/Rome for weekly weigh-in
reminder; auto-cleanup of expired tokens on 404/410
- env: VAPID_PUBLIC_KEY/PRIVATE_KEY/SUBJECT (all optional → push gracefully off)
Frontend
- public/sw.js minimal (push + notificationclick)
- lib/notifications/push-client.ts: subscribe / unsubscribe / getCurrentSubscription
- profile/notifications-{actions,panel}.tsx: editorial panel with toggles,
device list, "send test", per-device removal
- pushReady requires both permission AND active subscription (covers the
case where the user revoked the SW but kept the browser permission)
Schema
- DeviceToken { userId, platform, endpoint, p256dh, auth, token, userAgent,
createdAt, lastSeenAt } with unique(userId, endpoint)
- ExtendedPrismaClient type exported from @ketopath/db
- NotificationSettings zod schema in @ketopath/shared
Tooling
- lint-staged: split .js out of eslint glob so service worker is only
formatted (it lives outside the TS project)
i18n
- Notifications namespace (it) with typed error keys
Smoke tested: POST /me/device-tokens 201, POST /me/notifications/test 200,
real push delivered to a macOS Chrome device.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Picks application-level field encryption via prisma-field-encryption
(AES-256-GCM, master key from KMS) as the primary defence, combined with
volume encryption on the production Postgres host as defence in depth.
Documents:
- which fields are sensitive now (Profile.weight*, Profile.targetDate) and
which arrive in V1 (WeightEntry, FastEvent, ProgressPhoto)
- which fields stay in clear text and why (email/login, age/gender for
aggregate analytics, height/activity for plan generation)
- alternatives rejected: pgcrypto (key in queries), volume-only (no app-level
protection), client-side E2E (kills BMR/TDEE server-side calculation)
- consequences and the implementation roadmap for a follow-up PR
Status: proposed — must be implemented before opening V1 to public users.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reasons (full ADR in docs/decisions/0001-auth-provider.md):
- KetoPath targets EU users with Art. 9 GDPR data; Clerk on US infra
raises Schrems II concerns
- Free, self-hostable, stays in our Postgres
- TypeScript-first, fits Next.js + Fastify + Prisma stack
CLAUDE.md updated:
- Tech stack: Better Auth on EU Postgres
- "Cosa NON fare mai": no manual password hashing
- References: Better Auth docs link
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
