moni - privacy-first expense tracking for iOS

Most expense trackers ask for your bank login or sit behind a cloud account. Both are a poor fit for everyday manual entry: they are slow to open, slow to log, and they inherit the privacy overhead of a server you didn't ask for. The friction is so high that most people simply stop tracking.

Most expense trackers ask for your bank login. moni asks for two seconds.

The design bets everything on one gesture. You hold a center button, drag upward through a tactile amount ladder, then drop onto a category:

• →Hold - the center action wakes instantly; the screen dims and blurs behind it.
• →Scrub - a non-linear amount ladder (10, 20, 50, 100, 200, 500, 1000...5000) climbs with pow(p, 1.35) easing under your thumb, with 0.08s-throttled haptics.
• →Drop - releasing opens a radial picker whose AnnularSector shape highlights the nearest category based on distance (hypot <= 96pt).
• →Done - the expense is saved. You never fill five fields.

Manual entry is fast in the app, but the fastest input is the one you never type. moni exposes system integrations so expenses can be captured hands-free:

• →Siri - say "Add debit in moni 250" or "Add credit in moni 1000" and it logs without opening the app (openAppWhenRun=false).
• →Back Tap - assign "Quick Expense" to a double-tap on the back of the iPhone and enter an amount in a system snippet, no app launch needed.
• →SMS Import - a Shortcuts Message Automation hands a bank SMS to moni, which parses amount, payee, type, and category locally, then saves to SwiftData.

SwiftUI

100% declarative - matchedGeometryEffect tabs, contentTransition(.numericText()) for live numbers.

SwiftData

Four @Model types behind a ModelContainer, queried with @Query and #Predicate.

AppIntents

Six intents with a shared AppShortcutsProvider; ShowsSnippetView and ProvidesDialog for hands-free flow.

Money as Int

Paise stored as Int via Decimal * 100, avoiding Double precision drift entirely.

NSRegularExpression

Four amount patterns plus keyword dictionaries for type, payee, and category inference.

No dependencies

Swift Testing for the domain calculator; zero third-party packages.

The app follows a feature-sliced layout with a deliberate Domain/Persistence separation. The content view holds live @Query arrays and exposes computed values like activeAccounts and totals; the Domain layer is stateless enums of pure functions (balance math, SMS parsing, formatting); and FinanceStore is the single boundary that AppIntents use to read and write. Keeping the parser and calculator pure made them trivially testable without any UI on the stack.

• →The amount ladder feels right or it feels dead. Tuning the pow exponent so small amounts scrub slowly and large ones fly was the difference between a demo gesture and a daily-use one.
• →Radial picker hit detection. The AnnularSector highlight needed distance-based logic (hypot <= 96pt) so the finger reliably lands on the intended category at release.
• →Money as safe math. Using Double for currency introduces drift; switching to Decimal -> Int paise for storage, then formatting back for display, removed a whole class of bugs.
• →SMS parsing without a server. Building a regex rule engine (amount, debited/credited, payee extraction with stop-word trimming, category inference) that handles HDFC, ICICI, SBI, and Axis formats is pure pattern-matching resilience.
• →No backend - by architecture. The install is the app. There is no server to trust, which means the reliability story is entirely about correct on-device persistence.

A production-style iOS app with a testable domain core, five intents wired into the system, and a capture flow measured in seconds rather than fields. The full source, an open MIT license, and the live landing site are linked below.