Building and Launching Invoiceur

Challenge

Independent professionals want polished invoices without adopting accounting software — and many don't want their client list living on someone else's server. Adding AI to a product that handles money introduces a specific risk: a model that invents a client, a price or a total.

Approach

• Offline-first, no account. Data lives on the customer's devices. Optional sync uses the customer's own private iCloud database.
• AI extracts; code calculates. Apple Intelligence runs on-device and only extracts what the customer described. Amounts, due dates, client matching and totals are handled by deterministic Invoiceur code.
• Ground every draft. Names and numbers the customer didn't say are flagged, ambiguous clients go to a review screen, and output is always a draft. Nothing is sent automatically.
• Claims follow evidence. Marketing is limited to features that pass verification, tracked in a claims register.

Implementation

• Invoice creation and editing with currency-correct totals, per-invoice currency and unique numbering.
• Multiple business profiles with separate numbering, currency and branding; client management; full and partial payment tracking; overdue status.
• An insights dashboard, line-item suggestions from invoice history, and three PDF layouts that print payments and balance due.
• Keyboard-first workflows on Mac and iPad.
• "Describe an invoice in your own words" drafting, processed on-device. Requires Apple Intelligence on a supported iPhone, iPad or Mac running iOS 26, iPadOS 26 or macOS 26 or later.

Outcome

Invoiceur is available on the App Store for Mac, iPhone and iPad from a single SwiftUI codebase. Its on-device model integration tests pass alongside deterministic test suites for every calculation path.

Technical perspective

The hard part was letting a language model help without letting it near the arithmetic — and keeping one SwiftData and CloudKit codebase behaving correctly across three platforms. It's the same discipline we bring to client systems: put probabilistic components behind deterministic guarantees.