← Chris Little · AI work Launch the demo →
Pressure CookerRecipes to shopping list

Real recipes for what's in your kitchen, turned into a shopping list.

Tell us what's in your kitchen and the app will find real recipes, store your favorite recipes, let you pick your meals for the week, and generate shopping lists based on the recipes you choose.

Launch the demo → See how it works
146Real recipes from TheMealDB in the starting pool, every ingredient priced against the store list
0Recipes or numbers written by a model. Code does every count and every package
$3Daily cap on AI spend. Past it, the AI steps switch off and the rest keeps working
Why it exists

What makes it different

Real recipes

Each one links back to its publisher. Nothing is made up, and instructions stay on the source page.

Counts you can trust

“Uses 7 of yours, buy 3” is computed in code, never estimated by a model.

A list a store understands

Amounts across recipes are added up and rounded to real package sizes, by aisle.

The workflow

From fridge to list

  1. STEP 1

    Add what you have

    Type what's in your kitchen.

    AI matches odd names
  2. STEP 2

    Find recipes

    Real recipes, ranked by how many of yours they use.

    AI can search further
  3. STEP 3

    Plan the week

    Save favorites and pick this week's meals.

    You choose
  4. RESULT

    Shopping list

    One list for the week, by aisle.

    Code
AI and code

Each does the part it is good at

The model does

  • Matches an item it can't find (“tenderstem”) to one on the store list, for you to confirm
  • Picks a few recipes from a shortlist code has already ranked and filtered
  • Writes a one-line reason for each pick, rejected if it contains a number
  • Chooses search terms and when to stop, in search further

Code does

  • All package math and the shopping list
  • Match counts and ranking
  • Dietary filters, which fail closed
  • Suggested chips and the “results are weak” rule
  • Checking every recipe the agent finds, and every limit it runs under
What you get

A list you can shop from

uses 3 of yours, buy 5

Produce

  • Buy 3 bags of Carrotrecipes need 32 oz · 12 oz per bag
  • Buy 1 bulb of Garlic Clovesrecipes need 4 cloves · 10 cloves per bulb

Pantry

  • Buy 1 can of Canned Tomatorecipes need 7.94 oz · 14.5 oz per can
  • Buy 1 container of Paprikarecipes need about 0.38 oz · 2.12 oz per container

Probably have, check before you go: Black Pepper, Olive Oil, Onion, Salt

Lines from the demo's sample week of two real recipes. The match badge is an illustration of how results are labeled.

Scope

In this demo, and what's next

In this demo

  • Shopping list by store section, with items to cross off
  • Use what I have: real recipes ranked by how many of your items they use
  • Search further, when the pool comes up short
  • Recipes: browse, save, add by link, pasted text or typing
  • This week: several recipes combined into one list
  • Pantry and a one-screen profile for diet, cuisines and a “never” list
In development
  • Fridge, pantry and receipt photos to add ingredients
  • Automatic weekly meal planner
  • Household accounts shared across devices
  • Instacart link
  • Learning tastes from ratings
Under the hood

The AI and coding elements

Claude Haiku 5.5Claude Sonnet 5.5Anthropic APITheMealDB APITypeScriptNetlify FunctionsNetlify Blobs

AI

Mapping
Claude Haiku 5.5, the lowest-cost model. It may only answer with a name from the store list, or none.
Picks
Claude Sonnet 5.5, choosing from a code-ranked, diet-filtered shortlist. Any id outside it is dropped.
Search further
Claude Sonnet 5.5 with four tools: search by ingredient, search by name, check a recipe, finish. Code runs every tool.
Structured output
Answers come back as JSON against a schema, then code checks them again.

Code

Engine
A rule-based ingredient parser, unit conversion, package rounding, diet table and ranking, all in TypeScript.
Server
One Netlify Function holds the API key, the spend caps and the fetch rules. It runs no schedule.
Your data
Your household lives in your browser. The server keeps the recipe pool, the day's spend and cached recipe data. No accounts, analytics or cookies.
Recipes
From TheMealDB's public API, which invites this use and asks for credit and a link back. Both are given.
Scorecard and limits

What is measured, and what is not

Package math

Checked line by line against the original version on every recipe it could price: 634 lines, all identical.

Diet filter

A 20-recipe test set: no restricted recipe gets through, and anything unmapped says “can't confirm.”

Limits

$0.10 per search further, $3 a day, 10 AI actions per device per day. Tested with the caps set to one cent.

A filter, not a guarantee. Diet settings rule recipes out in code, but anything the table can't vouch for says “can't confirm, check the label.” It makes no health claims. Always read the label when it matters.

Try it with your fridge

Starts with a sample household, so there's something to see straight away. Your changes stay in your browser.

Launch the demo →