Textual Chef · for developers
A global recipe dataset, as JSON.
4,500+ complete recipes across more than a hundred cuisines, from Filipino and West African to Peruvian and Turkish, each structured the same way: ingredients with three batch sizes, the method at each size, nutrition, allergens and diets.
Coverage
The thirty largest cuisines in the catalog today. The long tail runs to over a hundred labels; the free sample takes one recipe from nearly every one of them.
| Cuisine | Recipes |
|---|---|
| American | 604 |
| Italian | 175 |
| Indian | 145 |
| Vietnamese | 136 |
| Indonesian | 135 |
| Mexican | 133 |
| Filipino | 132 |
| Spanish | 131 |
| Japanese | 129 |
| Korean | 128 |
| Chinese | 128 |
| French | 126 |
| Cajun | 124 |
| Middle Eastern | 122 |
| Turkish | 120 |
| Peruvian | 120 |
| British | 117 |
| Cuban | 116 |
| Thai | 113 |
| Russian | 111 |
| Italian-American | 108 |
| German | 108 |
| West African | 106 |
| Brazilian | 105 |
| Greek | 105 |
| Caribbean | 88 |
| North African | 87 |
| Southern American | 87 |
| Mediterranean | 69 |
| Moroccan | 36 |
Counts as of this page's last build. The catalog grows weekly.
Three ways to take it
- The free sample, now: 100 full records at
/api/v0/index.json, one file per recipe. - The index, now:
/recipes.jsonlists every recipe with its page and markdown links. Citation and summarization are welcome; link the source page. - The full dataset, in beta: every record, as an API or a one-time export. Pick "A one-time dataset export" in the form below if that is what you need. Commercial terms are set with beta testers, at a lifetime discount.
Uniform, on purpose
Scraped recipe data arrives in a hundred shapes: "1 (14 oz) can", "salt to taste", steps that refer to "the sauce" nobody defined. Every record here has the same shape and passes the same checks before it publishes: ingredients as structured lines with three batch sizes, the method keyed to those lines, labels from closed vocabularies. You write one parser.
Where the recipes come from
The recipes are written with the help of language models and published only after automated schema and culinary checks pass. Every ingredient carries half, standard and double amounts, and the method refers to ingredients by id rather than retyping them, which is how the amounts get written into each step. Nutrition is estimated, not measured in a lab, and every record says so with "estimated": true. Treat it as a planning figure, not a label.
Join the beta
The keyed API opens to a short list first. Beta testers get a lifetime discount on the launch price, whatever it turns out to be, and a say in the record format while it is still v0. The price is not set yet, so there is nothing to pay and nothing to cancel.
We store your email and your two answers, and nothing about your browser or IP address. Reply to the confirmation and we delete your address.