browse_drinks
ChatGPTList drinks by filter with no search text: category, alcoholic, iba, taste, difficulty, max_minutes, base_categories, occasion, and an order. Call this when the user gives constraints rather than words — "three crowd-pleasing gin drinks under 5 minutes", "easy mocktails for brunch", "official IBA classics". At least one filter is required. If the user wants a suggestion with no constraints at all, call random_drink instead; if they are asking what the catalog contains, call catalog_overview. Differs from search_drinks in taking no query, and from what_can_i_make in knowing nothing about what the user owns.
catalog_overview
ChatGPTHow many recipes exist, split by alcoholic, non-alcoholic, official IBA, and per category. Call this to answer "what have you got?", "how big is this collection?", or "do you have mocktails?" without spending a search that would return an arbitrary ten drinks. Returns counts only, no recipes — follow up with browse_drinks or search_drinks once you know which category to look in. Also carries the responsible-drinking notice and the app store links.
drinks_with_ingredient
ChatGPTList drinks whose recipe actually contains ONE named ingredient. Call this for "what else can I do with mezcal?", "I have a bottle of Chartreuse going spare", "drinks that use egg white". Takes exactly one ingredient, in plain English ("sweet vermouth", "Campari", "limes") — it is resolved against the catalog for you, so never pass a slug. For several ingredients at once, or to know what is actually makeable, use what_can_i_make. This tool ignores everything except the one ingredient and the filters you pass. More reliable than search_drinks for ingredient questions, because it matches the structured ingredient list rather than the description text. Accepts the same filters and paging as browse_drinks.
get_drink
ChatGPTThe complete recipe for one drink: every ingredient with its measure, numbered method steps, glassware, tools, ABV, calories, serves, taste profile and food pairings. Call this as soon as the user settles on a drink and wants to make it — the other tools return summaries only and never include measures. Takes the id from any other tool result, not a name. If you only have a name, call search_drinks first. Measures come back already formatted and localized: reproduce them as given rather than converting units. To scale a recipe, multiply from these measures. An unknown id returns a "not found" result, not an error.
random_drink
ChatGPTOne or more drinks picked at random. Call this for "surprise me", "pick something for me", or when the user wants inspiration with no constraints. Default 1, maximum 5. Accepts alcoholic, so "surprise me, but nothing alcoholic" works. For any other constraint use browse_drinks, which filters properly; this tool only understands alcoholic.
search_drinks
ChatGPTFree-text search over drink names and descriptions. Call this when the user names a drink or describes one in words — "negroni", "something with elderflower", "a less sweet daiquiri". query is required and must be non-empty. Do NOT use this for a list of ingredients the user owns — that is what_can_i_make. Do NOT use it when the user gave only constraints and no words to match on ("something bitter under 5 minutes") — that is browse_drinks. For "which drinks use crème de cassis?" prefer drinks_with_ingredient, which matches the recipe's actual ingredient list rather than its prose. Returns summaries plus total, so you can say "showing 10 of 143" and page with offset. Call get_drink for measures and steps.
similar_drinks
ChatGPTDrinks close to a given one in category and taste profile. Call this for "something like that but less bitter", "what else is in that family?", or to offer alternatives after the user rejects a suggestion. Takes a drink id and returns up to 6 by default. Use search_drinks instead when the user describes what they want in words, and browse_drinks when they give explicit filters — this tool derives everything from the reference drink and takes no filters of its own.
what_can_i_make
ChatGPTWork out which cocktails a person can make from the bottles, mixers, juices and garnishes they already own. Call this whenever the user says what they have — "I have gin, Campari and sweet vermouth", "all I own is vodka and orange juice", "what can I make with these?". Use this INSTEAD OF search_drinks for any list of ingredients: search_drinks matches text in names and descriptions and cannot compute makeability, so it will miss nearly everything. This is the only tool that knows what a recipe needs versus what the user has. Pass have as plain names in the user's own words — never slugs. Returns two groups: can_make (nothing missing) and almost (within missing_max, default 1), each with a match percentage, plus matched and unmatched so you can tell the user which of their bottles were not recognized. Carry their ingredient list forward on later turns — nothing is stored between calls.
