A photo-based nutrition tracking app. Enter body stats and a goal, get daily calorie/macro/micronutrient targets, then photograph meals to log intake against them. All user data lives in the browser (IndexedDB) — there is no server-side database and no environment variables.
pnpm install
pnpm devOpen https://nutrientracker-lake.vercel.app/dashboard. No .env file is needed.
Set up your profile at /profile first (this computes your daily targets),
then log meals from / (photo, barcode scan, or manual search) and check
/dashboard for your daily/7-day view. Bring your own Anthropic API key in
/settings for photo analysis, or run Ollama locally.
The food database (public/data/foods.json and public/data/foods-search-index.json)
is generated once, offline, and committed to the repo — it is not built at
Vercel build time or fetched from a live API at runtime. This keeps the Vercel
build a no-op for nutrition data and avoids shipping any database or API key.
To regenerate it from the current USDA FoodData Central bulk CSV release:
pnpm build:fooddbThis downloads the USDA Foundation Foods + SR Legacy CSV bulk datasets (public
domain, no API key) into scripts/.cache/ (gitignored), keeps only those two
data types (~8,000 foods, excluding Branded Foods), prunes to ~28 core
nutrients, and writes:
public/data/foods.json— compact parallel-array food + nutrient datapublic/data/foods-search-index.json— a prebuilt MiniSearch index
Both generated files are committed deliberately. You normally don't need to run this script — only re-run it if you want to refresh the dataset from a newer USDA release.
Packaged/branded foods and barcode lookups are resolved at runtime directly against the Open Food Facts public API from the browser (no key, no proxy — its CORS headers allow direct browser calls).
Import this repo into Vercel and click Deploy.
No configuration is needed: no vercel.json, no environment variables, no
custom build/output settings. Vercel's standard Next.js auto-detection is
sufficient.
pnpm testAll nutrition math (targets, calibration, etc.) is written as pure, unit-tested functions, run with Vitest.