Recipe and shopping list app for Android, iOS and smartwatches, with AI recipe import: you paste a link and the app fills in the recipe. Published on Google Play and the App Store. Formerly called BlueList.
This repository is a product showcase: the source code is private because the app is commercial (subscription). Here I explain what it does and how it is built.
Your recipe collection and your shopping list in the same app.
- Recipes with photo, time, servings, ingredients and steps; they can be split into several parts (dough, filling, sauce) and searched by name or ingredient. Favorites, notes, star ratings and filters to find them again.
- AI import: you paste the link of a recipe website or a YouTube video (or share it from the browser, Instagram, TikTok or YouTube) and the app fills in title, photo, servings, ingredients and steps. The AI runs in the cloud behind the app's own servers and only receives the page text, never the user's name or email.
- More with AI: translate a recipe into your language without touching the original, read a handwritten shopping list from a photo and estimate the nutrition per serving.
- Cooking mode: one step per page, full screen, large text, timers and the screen always on. Recipes can also be printed with a cookbook-style layout and a QR code to the original.
- Shopping list one tap away from the recipe, a catalog of more than 340 products, items grouped by aisle and logos for more than 300 supermarkets.
- Cookbooks and a weekly meal plan on a calendar that turns the week's recipes into a shopping list with the amounts added up.
- Loyalty cards scanned, shown full screen and at maximum brightness.
- Watch apps (Wear OS and watchOS): lists and cards from your wrist.
- Backup: export your recipes to a file with their photos and bring in the ones you have in Paprika or Mela.
- Premium (subscription): cloud sync across devices, more AI imports per month and sharing recipes, cookbooks and lists by link, with the lists updating in real time.
19 languages and no ads. A free account is required and it can be deleted with all its data from the app.
| Layer | Technology |
|---|---|
| Android | Kotlin, Jetpack Compose, Room, WorkManager (background import queue), Glance widget |
| iOS | Native, standalone SwiftUI (shares no code with Android), SwiftData, Share extension, widgets and App Intents |
| Watches | Wear OS and watchOS as companion apps |
| AI | A cloud LLM, called only from Cloud Functions. The key never ships in the app and every call needs App Check and a signed-in user; the model, quotas and spending caps live in Firestore, so they change without an app update |
| Backend | Firebase: Auth, Firestore, Storage, Cloud Functions (AI, verification of Google and Apple purchases, shared-list notifications, transactional email, scheduled cleanup, account deletion), Cloud Messaging, App Check, Hosting |
| Web | Landing page on Cloudflare Pages; shared-link pages (/join, /recipe, /cookbook) and legal pages on Firebase Hosting |
- From on-device AI to the cloud, on purpose. The first versions imported recipes with AI inside the phone (Gemini Nano on Android, Apple Intelligence on iOS): free per user and private, but only on recent phones and with too many mistakes. In October 2026 I benchmarked both and moved all the AI to the cloud: it works on every phone and it is far more accurate (on 52 photos of handwritten lists it read 91.5% of the items, against 67–73% on the device). The trade-off is a cost per request, so every account has a monthly allowance and the server has a daily spending cap. All the AI goes through a single gateway per platform, so changing the model or the provider does not touch any screen.
- Structured data first. When a recipe page already carries the recipe as schema.org data, the app takes it from there and the AI only tidies it up; the phone checks the result against the original and keeps the original if a quantity changed.
- Two native apps with parity. Android and iOS share no UI or domain code, but they do share the data format: the logic that exists on both sides is written twice, as twins, and checked against the same test cases, so that both parse a recipe or sort a shopping list exactly the same way.
- Tested security rules. The Firestore rules have a suite of more than 500 cases that runs against the emulator before deploying, so that sharing a list never opens up more than it should.
- Purchases verified on the server. Google and Apple are validated in Cloud Functions, not on the client, and Premium is only turned on when the server confirms it.
Two stores, subscriptions in both, store listings in 19 languages, a privacy policy and the maintenance that comes after: new versions, real bugs and real users.
Pedro Antonio Flores Casquet · github.com/PedroFlores199




