← All workCase study No. 16Native macOS App16 / 17

AI Meter.

What your AI coding tools really use and cost

A local-first macOS app that shows what AI coding tools use and cost, without reading prompts or code.

Client: Concept Project
Industry: Developer Tools
Engagement: Build
Relationship: An independent studio exploration — not a client commission

AI Meter macOS app overview with usage and cost cards, in demo mode

IThe ground

The brief.

Developers who work across several AI coding CLIs — Kiro CLI, Claude Code, Codex CLI and OpenCode — rarely know what they are really spending. Paid plans, local estimates and API-equivalent prices blur into one misleading number, and most tools that could help would need access to prompts and code.

IIThe plan

Our response.

We built AI Meter as a native, local-first macOS app in Swift 6 and SwiftUI. It reads only structured usage metadata from each tool's local files, keeps actual paid, estimated, API-equivalent and allocated subscription cost apart, and adds forecasts, budgets, anomaly alerts and value insights. A menu bar meter and WidgetKit widgets keep usage in view, data exports as formula-safe CSV or encrypted backups, and a deterministic demo mode lets every screen be explored without connecting a provider.

macOS App Development
Data Visualisation
Privacy Engineering
Widgets & Menu Bar

IIIThe buildStrategy → craft → engineering → care

Brick by brick.

Course 01
Privacy Boundary First
A metadata allowlist and hashed project IDs, so prompts and code are never stored, with read-only, per-provider folder access.
Course 02
Honest Cost Semantics
Separate engines for pricing, cost, forecasts, budgets and anomalies, so every figure says exactly what it is.
Course 03
Native Everywhere
SwiftUI, SwiftData and Swift Charts, a menu bar extra, Today and Month widgets, App Intents and Spotlight.
Course 04
Tested and Portable
139 tests with per-provider fixtures, CSV and JSONL export, and AES-GCM encrypted backups with a device-only Keychain key.