What you get
Thousands of recipes in one shape. Ingredients parsed into quantities, units and canonical names. Steps numbered in order. Times in seconds. Servings, nutrition, dietary flags and media in fixed fields. Cuisines and courses from closed lists, so a filter means the same thing on every recipe.
It is a read API with semantic search, meant for apps, agents and anything else that needs recipes it can actually compute with. On top of single recipes it can scale servings, convert between metric and imperial, match recipes to what is in a pantry, suggest substitutions, and merge several recipes into one shopping list.
How a recipe gets in
A recipe starts as the structured data a recipe page already publishes for search engines. We read it, normalize it into one model, and then have a language model split every ingredient line into a quantity, a unit, a canonical name and a preparation, in the recipe’s own language and in English.
After that, each ingredient is checked against the EU’s 14 regulated allergens and the US definitions of wheat and tree nuts, and the recipe is tagged with the diets it suits. Last, it is embedded for search, which is why a query for “creamy pasta” also finds carbonara.
Where the recipes come from
Public recipe pages. Every recipe links back to its source and credits its author. Publishers who would rather not be included can write to chris@pullpanda.io.
Start building
The free plan comes with 50 requests a day and 300 a week, and the paid plans include more. The quickstart gets you a response in a couple of minutes, and you can run every endpoint straight from the documentation.