FitOpe

A personal fitness, nutrition, sleep and body-transformation tracker. Training with real progression, Indian food logging in katoris and rotis, sleep and recovery, and analytics built from what you actually logged.

Next.js (App Router) + Convex + Tailwind v4.

Run it

npm install

Two processes, two terminals:

npm run dev:convex

npm run dev

dev:convex uses Convex's anonymous local backend (no account needed) and writes NEXT_PUBLIC_CONVEX_URL into .env.local on first run. To use a hosted Convex project instead, run npx convex login && npx convex dev.

Auth needs three deployment env vars. Generate them once:

npx @convex-dev/auth

or set them by hand with npx convex env set JWT_PRIVATE_KEY …, JWKS, and SITE_URL=http://localhost:3000.

The shared exercise and food libraries seed themselves on first sign-in; npm run seed forces it.

npm run check   # asserts the nutrition/strength math and the seed data

Release

Vercel builds the frontend from main; the Convex backend is deployed separately. Ship the backend first so the new frontend never calls a function production doesn't have yet:

npx convex deploy

git push origin main

What's in it

Onboarding collects only what changes the maths — age, height, weight, goal, experience, activity, schedule, equipment, diet, sleep window — and every step is skippable. It ends by showing the calorie and macro estimate it derived, labelled as an estimate, before anything is saved.

Training. A generator turns the profile into a 2–6 day split from a 105-exercise library, picking movements that match the equipment you actually have. Sessions are logged set by set with the previous session's weights prefilled, a rest timer, per-exercise history, and estimated-1RM personal records. Plans are fully editable: reorder, retitle, change sets/reps/rest, duplicate, activate, or build one from scratch. Partial and skipped sessions are first-class.

Nutrition. 265 Indian foods with real serving units — roti, katori, bowl, glass, piece — and raw vs cooked kept distinct where it matters. Custom foods, multi-ingredient recipes with automatic per-serving maths, saved meals, recents, favourites, repeat-yesterday, and a free-text quick add. Protein and fiber are the two macros given visual priority.

Body, sleep and recovery. Weight with an exponentially-weighted trend line (so a heavy dinner does not read as fat gain), tape measurements, manual sleep logging with a consistency score, and a readiness score built from sleep, soreness, energy, stress and recent training load.

Analytics. Weekly volume, strength curves, muscle-group balance, protein and fiber adherence, sleep averages, derived milestones, plus deterministic written insights that read the data and say something specific.

Reminders are in-app only: a due reminder shows on the home screen and disappears the moment the thing it is nudging you about is logged. Nothing is pushed or emailed.

Units. Body weight and measurements can be shown in kg/cm or lb/in; training loads stay in kg because that is how plates are marked.

Design language

Taken from the supplied references and applied throughout:

• Light-first, cool paper ground with white cards, generous corner radii and soft elevation. Dark is fully supported and swaps the palette rather than dimming it.
• Near-black primary actions — the reference's "Join" button — with lime reserved for the dark theme's CTA and for data series.
• Pastel media tiles. Every workout, exercise and food row carries a coloured thumbnail with a category glyph, derived deterministically from muscle group or food category. No stock photography is used, since none is licensed for this build; the tiles do the same scanning job down a list.
• Circular minute badges on every row, and a pastel hero card (chip, oversized title, one action, decorative discs) as the top of Home, Train, Nutrition and the plan detail screen.
• Plan detail follows the reference's structure: hero, goal chips, a four-cell "Session details" grid, numbered exercise rows, and a sticky primary action.
• Bottom nav is a floating white pill with a dark active item, label included.

Design notes

• Data model separates planned from performed. Program days are templates; workouts, exercises and sets are what happened.
• Meal entries snapshot their nutrients. Editing a food later never rewrites history.
• Targets are versioned by effectiveFrom, so changing a goal does not retroactively alter whether past days "hit target".
• Everything is scoped by userId and every mutation re-checks ownership.
• The insight engine (convex/analytics.ts) is deliberately deterministic. It is the seam an AI coaching layer plugs into later — the data it reads is already shaped for that.

Not a medical device

Calorie, macro and readiness figures are estimates from public formulas and composition tables. The app does not diagnose, treat or advise on health conditions.