Staring into a full fridge with no idea what to make?
It happens every evening. Plenty of ingredients, zero plan — so you order in, or something quietly goes off in the drawer. FridgeAI ends the guesswork: take one photo of what you’ve got, and it finds real meals you can cook right now, from exactly what’s already there.
📸 Your fridge
✅ Detected & confirmed
🍽️ Meals you can make
“We have all this stuff… and nothing to eat.”
A recipe app assumes you already know what you want and sends you shopping. Real life runs the other way: you have a random handful of ingredients and no idea what they add up to. That gap is where takeaway orders and wasted food live.
Decision fatigue
A full fridge, a blank mind. Endless scrolling through recipes that need six things you don’t have.
Food quietly wasted
The spinach wilts, the herbs go slimy, half an onion is forgotten — bought with good intentions, thrown out unused.
An unnecessary shop
You had enough for dinner all along — you just couldn’t see the recipe hiding inside what you owned.
“Let’s just order”
The default when thinking is too much effort — money spent on delivery while the fridge stays full.
Whether it’s too much or too little, the question is always the same: what can I actually make with this? FridgeAI answers it.
Point your camera. Get a plan you can cook.
Not a search box, not a meal-kit upsell. A vision model reads your ingredients, you make it right, and it builds meals around what you have — showing exactly what’s missing and prioritising what to use first.
Reads your ingredients
Photograph your fridge, shelf, or groceries and a vision model lists what it sees — with honest, approximate amounts, never fake certainty.
You stay in control
Rename, adjust, remove, or add anything it missed. Uncertain items show as “Possibly…” so you decide — recipes never run on raw guesses.
Cooks from what you have
Recipes are ranked by how much you already own — “Uses 8 of your 9 ingredients” — favouring meals that need zero to three extras.
Required vs. optional, split
See what a dish truly needs versus the nice-to-haves, so a short shopping list never masquerades as a big one.
Hands-free Cook Mode
One big step at a time with built-in timers — designed for a phone propped on the counter and one messy hand.
An assistant that knows the dish
“Can I swap the eggs?” “Make it spicier.” “Cook for four.” It already knows your recipe and ingredients — no repeating yourself.
Photo → Detect → Confirm → Choose → Cook.
Snap a photo
Upload, drag in, or take a picture of your fridge or ingredients on the counter.
Check the read
FridgeAI lists what it spotted; you fix anything wrong and add what it missed.
Set a mood
Optional chips — vegetarian, under 30 min, high protein, use leftovers — no long form.
Pick a meal
Browse ranked recipes with time, calories, protein, and exactly what you’re missing.
Start cooking
Step through Cook Mode with timers, and ask the assistant anything along the way.
It nudges you toward what to use first.
When perishable ingredients show up — leafy greens, mushrooms, fresh herbs, dairy — FridgeAI floats a gentle “use these first” note and leans recipes toward them. Carefully worded, always: it can’t judge freshness from a photo, so it never claims food is safe or spoiled — it points you to the use-by date, storage, smell and texture, and asks you to check before cooking.
A modular AI pipeline, not a single prompt.
Every model call is a discrete, schema-validated stage, decoupled from the UI and from any one provider. If the model is unavailable, each stage falls back to a deterministic engine — so the app always works.
Vision + recipes + chat run through a keyless application over a thin gateway to a fast vision-class model; nothing model-specific leaks into the UI. Ingredient⇄recipe matching is computed locally, so the numbers you see (“uses 8 of 9”) are honest and explainable — never a meaningless AI confidence score.
An honest feature matrix.
Every capability the product set out to deliver, and where it really stands today. If everything read “done,” nobody should believe it — so this marks built, partial, and roadmap plainly.
| Capability | What it does | Status |
|---|---|---|
| Photo detection | Vision model lists visible ingredients with approximate amounts and confidence. | Built |
| Confirm & edit | Rename, adjust quantity, remove, add missed items; uncertain items flagged “Possibly…”. | Built |
| Preferences | Diet, time, cuisine, difficulty, goal — as compact optional chips. | Built |
| Ranked recipes | 3–6 meals ordered by how much of each you already have. | Built |
| Required vs optional | Splits must-have ingredients from nice-to-haves; prioritises 0–3 extras. | Built |
| Cook Mode | One step at a time, large type, per-step timers, one-handed on mobile. | Built |
| Ask FridgeAI | Grounded assistant: swaps, scaling, spice, appliance, dietary tweaks — knows your context. | Built |
| Use-these-first | Careful perishable nudges to reduce waste; never asserts food safety. | Built |
| Offline demo | Deterministic mock so the full flow works with no key and no network. | Built |
| Saved pantry | Your ingredients persist across a refresh (in-session today). | Partial |
| Barcode / receipt scan | Add items by scanning packaging or a shopping receipt. | Roadmap |
| Shopping list export | Send the short “missing” list to your phone or a grocery app. | Roadmap |
| Accounts & history | Save favourite recipes and a running pantry across devices. | Roadmap |
The roadmap.
🔜 Next
- Persistent pantry that remembers across visits
- One-tap shopping list for the missing items
- Serving-size scaling baked into every recipe
🧭 Exploring
- Barcode & receipt capture to add items fast
- Multi-photo scans for a whole fridge + pantry
- Real recipe photography instead of placeholders
🌱 Later
- Accounts, saved favourites, weekly planning
- Nutrition targets and dietary profiles
- Household sharing of one pantry
Try it in the cloud, or keep it on your own hardware.
The same app, one code change apart — flipped with environment variables alone.
Nothing to install
The live demo runs 24×7 on a small box, pointed at a fast vision model through a thin keyless gateway. Walk the whole flow from any browser — even offline it falls back to a built-in sample kitchen.
OPENROUTER_API_KEY lives only in the gateway; the app itself stays keyless.
Your photos stay yours
Run FridgeAI on your own machine and route detection to a local or private model. Images never leave your hardware; the recipe logic and matching are plain code.
One command: npm run dev — the whole experience on localhost:3120.
Access is gated: enter your email, receive a one-time PIN, and the live app opens. Close the tab and the session ends. The demo dataset is a synthetic sample kitchen — no real user data is ever on the public demo.
Stop staring. Start cooking.
Photograph what you’ve got and see dinner appear — free, in your browser, in about a minute.