Spoon & Ledger(Demo)

AI tools

Spoon & Ledger uses an AI model for one job: reading pictures you take, so you type less. There are five places it does that. Each one fills a form or a list for you to check, and nothing is saved until you have.

Scan a shelf

Film or photograph a fridge, freezer or cupboard and get a list of what is on it.

Where
Pantry › Scan a shelf
Sends
Up to 10 stills, the place’s name, and your ingredient names so it can use them.
You check
The list, item by item. It marks what you have or are low on and leaves amounts alone.

A recipe from a photo

A cookbook page, a magazine cutting or a handwritten card, read into a recipe.

Where
Recipes › Import › From a photo
Sends
Up to 4 pictures of the pages, and nothing else.
You check
The draft against the page, before it is saved.

A paper receipt

Photograph a till receipt and it is read line by line, like a pasted one.

Where
Shop › Receipts › Photograph a receipt
Sends
Up to 4 pictures of the receipt, and nothing else.
You check
Each line and its match, in the usual receipt review, before it reaches the pantry and price book.

A nutrition label

Photograph a label and its numbers fill the food’s nutrition form. Useful when a barcode isn’t found.

Where
An ingredient’s nutrition form, or New food in the + Log form
Sends
Up to 3 pictures of the label and pack, and nothing else.
You check
The figures against the pack. It copies them and never estimates.

A meal from a photo

Photograph a plate and get the foods on it as rows with amounts, not one calorie number.

Where
+ Log › Photo
Sends
Up to 3 pictures of the plate, and your food names so it can match them.
You check
Every row: change a food, an amount or a unit, remove or add, then log.

How it works

Pictures go through Vercel’s AI Gateway to Google’s Gemini (google/gemini-2.5-flash), or to openai/gpt-5.2 when the first is down or refuses. A copy’s owner can choose another model. A read costs about half a cent. The pictures are not kept, and a video is cut into stills on your device rather than uploaded.

The model reads; it doesn’t decide. It copies what is printed and says what it could not read rather than guess. What it names is then matched to your own ingredients and foods by the same code that handles a pasted recipe or receipt.

On a copy with no password, such as the public demo, reading pictures is off unless the owner turns it on, because every read is paid for.

What isn’t AI

Importing a recipe from a link, matching ingredients, suggestions, filling the week, Cook tonight, costs and adaptive targets are all ordinary rules and arithmetic over your own data. They work the same with the AI tools switched off, and they can say why they chose what they did.

Exactly what each tool sends is also on the privacy page.