FitForge

A bilingual (Arabic/English) fitness & health tracking web app — workouts with voice-guided coaching, nutrition logging, progress analytics, and a subscription system that unlocks AI-powered features.

Real accounts, sessions, and per-user data are backed by a Postgres database (Neon) through Vercel serverless functions under /api.

Stack

• React 19 + TypeScript + Vite
• Tailwind CSS v4 (CSS-variable based theming)
• React Router
• Zustand (client-side UI state)
• react-i18next (Arabic/English, RTL/LTR)
• Recharts (progress charts)
• Web Speech API (voice-guided workout cues)
• vite-plugin-pwa (installable, offline-capable)
• Vercel Functions + Postgres (Neon) — auth, sessions, per-user data

Features implemented so far

• Onboarding flow (language, gender, goal)
• Real accounts: registration/login backed by Postgres, bcrypt-hashed passwords, httpOnly session cookies
• Dashboard with daily summary and quick actions
• Workout library with categories, detail screens, and an interactive player (timers, sets/reps, rest periods, voice cues in the user's language with a male/female voice preference)
• Nutrition logging with a calorie/macro calculator (Mifflin-St Jeor + TDEE)
• Progress tracking (weight chart, weekly activity chart, personal records)
• Subscription plans (Free / Premium / Pro) that gate AI-powered features behind a PremiumGate component
• Five selectable workout accent themes plus light/dark/system appearance
• Full Arabic/English localization with automatic RTL/LTR switching

Getting started

npm install
npm run dev

The /api serverless functions require a DATABASE_URL (or POSTGRES_URL) environment variable pointing at a Postgres database — they won't run without it. On Vercel, connect a Postgres integration (e.g. Neon) under the project's Storage tab; locally, set it in a .env file and use vercel dev to run the functions alongside Vite.

AI and device configuration

• Set GEMINI_API_KEY on the server for coaching, training/nutrition plans, and AI reminders. GEMINI_MODEL can override the default gemini-3.8-flash with a model available to your Google project. GOOGLE_API_KEY is also accepted as a compatibility fallback. Both chat and plans share the same client, a 50-second provider timeout, and explicit configuration/service errors. Never expose these variables through a VITE_ prefix.
• The AI functions have a 60-second Vercel budget. Plain npm run dev serves the frontend only; use vercel dev with your database and server environment for live authenticated integration testing.
• Google Fit requires GOOGLE_CLIENT_ID and GOOGLE_CLIENT_SECRET, and the OAuth callback /api/fit?action=callback on the deployed origin. Existing connections must be disconnected/reconnected to grant the new heart-rate and location permissions. The Google project/API and the connected device must actually supply the requested data. Missing readings remain unavailable, not synthetic zeroes; optional sensor failures do not disable steps/calories.
• Google's granular consent lets a user grant only some of the requested Fit scopes, and some accounts have no default data source for a type — so every metric is fetched and failed independently (steps fall back to the merged estimated_steps source) and one broken reading never blanks the others. Settings explains when a reconnect with all permissions is needed. A revoked refresh token (invalid_grant) clears the stored connection and asks the user to reconnect instead of showing a connected-but-dead state, and the connect/disconnect buttons show a loader for the whole round-trip.

Activity and appearance

• Daily totals use the local calendar date. Workout streaks count unique completed-workout days and remain valid through today if yesterday was active.
• Google Fit total energy includes resting metabolism. Home separately labels that total and estimates active energy by subtracting elapsed resting energy (Mifflin–St Jeor BMR); it does not add overlapping local workout estimates.
• Guided sessions save actual unpaused elapsed time and a duration-scaled calorie estimate, not the full catalog calories for a skipped session. Failed saves expose retry rather than reporting success. Historical saved estimates are not rewritten because actual historical durations cannot be reconstructed.
• Heart rate is the daily average, not a live medical measurement. Distance and heart rate require real synced readings. Progress has no demo/random data.
• Light, midnight dark, and device-system appearance are available independently of the accent theme. The system setting follows live OS changes and persists.
• Training and nutrition share the AI plan/PDF component. The PDF button opens a print-friendly document; select Save as PDF in the browser print dialog. Plans are fetched in the selected app language.
• Subscription selection is still a demo: the one-month trial label is shown, but no payment, automatic billing, or trial-expiration engine is implemented.

Verification

npm test
npm run lint
npm run build
npx tsc -p tsconfig.api.json

Unit tests cover local-day streak boundaries, active-calorie estimates, sensor aggregation/unavailable readings, shared Gemini configuration and errors, and Arabic/English key parity, in addition to existing calculation/voice/pose tests.