# Magpie A little less to keep in your head. A native iOS assistant that turns conversations, screenshots, and shared links into calendars, expenses, notes, lists, and saved places. ## Capture first, organize as needed Everyday information arrives in the wrong format: an appointment in a message, a purchase on a receipt, a place in a shared link, or an idea in a passing thought. Magpie is my native iOS project for turning those fragments into records that are easier to use later. The entry point is conversational. The destination is a calendar event, an expense, a note, a saved place, or another structured item. The screenshots on this page come from the running app in an isolated iOS simulator with sample data. ## A shared interaction across modules The home screen and module views use a common conversation engine, input bar, and outcome presentation. Text, dictation, and shared material flow through the same interaction rather than requiring a different AI feature for each screen. The expense view combines records with category and weekly summaries. The calendar supports browsing dates and inspecting the day's events. Notes keep quick capture lightweight. MapKit supports the places experience, and the share extension brings content in from other apps. ## An assistant that asks when it is unsure The product rule is to work from known facts. If a date, amount, or intention is ambiguous, the interaction should make that uncertainty visible and ask for clarification. A useful assistant needs an understandable result that the person can inspect and correct. That principle also affects ordinary interactions. Deletions have a recovery path, feedback is consistent across modules, and generated outcomes are presented as concrete records rather than a wall of chat text. ## Native storage, a separate AI service SwiftUI renders the interface and SwiftData persists the app's records locally. The cloud AI path is served by a TypeScript Cloudflare Worker, which proxies model requests, supports streaming responses, and records usage in D1. This separates the mobile interaction and local data model from provider-specific transport. Magpie itself is an iOS application; the server component handles AI requests. A website can present the app's walkthroughs and distribution links while its backend is operated independently. ## What the walkthrough shows The short recording below moves from the sample calendar into the expense dashboard and switches the category chart from percentages to amounts. It demonstrates navigation and data presentation without requiring a live model call. The recording is silent; all of its information is visible in the interface. ## Technology reference - **Native application:** Swift, SwiftUI, SwiftData - **Platform integration:** Share extension, Speech input, MapKit - **AI backend:** Cloudflare Workers, TypeScript, Streaming responses, D1 usage records These notes describe the implementation reviewed for this portfolio. Availability of a public demo is separate from the project description.