Fridge

AI-assisted home fridge inventory with use-by suggestions and hazard highlighting (see Description.md and docs/inventory-and-spoilage-alerts.md).

App (web/)

Stack: Next.js 15, React 19, Prisma + SQLite, and the OpenAI API for structured hazard + shelf-life inference (default model gpt-5.4-mini), with local rules if the API is not configured.

Run on your Mac

1. Install Node.js (LTS, includes npm).
2. In a terminal:

cd web
cp .env.example .env

3. Edit web/.env and set your OPENAI_API_KEY (from OpenAI API keys).

4. Then:

npm install
npx prisma db push
npm run dev

5. Open http://localhost:3000.

Without OPENAI_API_KEY, the app still runs; use-by and hazard hints use local rules and tables only.

Environment (typical setup)

Variable	Required	Notes
OPENAI_API_KEY	Yes (for AI)	Your sk-… key.
OPENAI_MODEL	No	Defaults to gpt-5.4-mini.
DATABASE_URL	Yes	Default in .env.example: file:./dev.db (SQLite under web/prisma/).

Optional: local LLM (Ollama / LM Studio)

If you set OPENAI_BASE_URL (e.g. http://127.0.0.1:11434/v1 for Ollama), the app talks to that server instead. Use OPENAI_MODEL to match the local model name (e.g. llama3.2). If you also keep an sk-… key, a failed local call can retry on OpenAI using OPENAI_FALLBACK_MODEL (default gpt-5.4-mini). See web/.env.example history or ask in the repo if you need this.

Repo layout

Fridge/
├── Description.md          # Master feature list
├── docs/                   # Feature specs
├── README.md
└── web/                    # Next.js application
    ├── prisma/
    │   └── schema.prisma
    ├── src/
    │   ├── app/            # Routes & API
    │   ├── components/
    │   └── lib/            # Inference, rules, Prisma client
    ├── package.json
    └── .env.example

API (local)

• GET /api/health — Liveness + LLM config (llmConfigured, provider, model).
• GET /api/items?sort=useBy|addedAt — Active inventory.
• POST /api/items — Create item (runs inference; optional useBy override).
• POST /api/infer — Preview suggestion without saving.
• PATCH /api/items/[id] — Update item.
• DELETE /api/items/[id] — Remove item.
• POST /api/items/[id]/consume — Mark consumed.

Caveats

• OpenAI usage is billed; calls need internet.
• Accuracy: Models can be wrong about food safety. The UI includes disclaimers; curated rules bias toward caution.

If anything fails (Prisma, port 3000 in use), check the terminal from npm run dev or npx prisma db push.