Skip to content

Desktop App · Windows & macOS Time Tracker

StaffVertex — Desktop Tracker

My work on the StaffVertex desktop tracker at Alphinex: a Tauri 2 and Rust app for Windows and macOS that records offline to an encrypted SQLite database, survives crashes and sleep, and detects clock tampering.

  • Running timer, task picker and weekly reports. Shown with demo data.
  • Task details beside the running timer, and the sync audit log that makes every sync and integrity decision inspectable.
  • Light and dark themes, plus settings for startup behaviour and diagnostics.
On this page

My role

I worked on the StaffVertex desktop tracker at Alphinex Solutions alongside the web platform it reports to, first as a developer and now as Team Lead. It was built with the team. The parts I can speak to directly are the signed update flow with its server-controlled silent mode, the macOS fix described below, and the server contract that keeps every installed build working.

The problem

A time tracker lives on someone else's computer all day and produces a figure that decides pay and invoices. That is unforgiving. Wi-Fi drops in cafes and on flights. Laptops sleep, crash and get force-quit. Organizations disagree about what idle time means. The person being paid owns the machine, so the system clock cannot be believed. And the app has to be light enough to leave running, on Windows and macOS, without becoming a second browser.

Key decisions and why

Tauri and Rust instead of Electron

An always-on background agent should not carry a bundled browser. Tauri 2 uses the system webview, so the installer and resident memory stay far below a comparable Electron build. The price is that the real logic lives in Rust, and the team had to be comfortable reviewing it.

A thin UI that cannot invent time

The React 19 side never touches the network or the database. It calls typed commands over Tauri IPC and listens for events the core pushes back. The UI can ask the core to record time, but it cannot write a time log itself. That keeps the trust boundary in one place.

Local first, with two separate sync queues

Everything is written to SQLite before anything else happens, so the tracker works with no network. Time logs and screenshots drain through separate sync engines with retry and backoff. I like this split because a stuck screenshot upload then cannot delay time data. If the database is locked, writes go to a pending queue that the next heartbeat flushes, so nothing is silently dropped.

Idle is a policy, and worked time is never trimmed

The organization picks one of four behaviours when someone stops typing: keep, discard, prompt or stop. An earlier version auto-stopped timers on a timeout and destroyed hours that were really worked, so that mechanism was removed entirely. The only time ever taken out is idle that someone explicitly chose to discard.

Distrust the clock, but keep the work

If the PC clock moves beyond a small tolerance mid-session, the session stops at trusted time, tracked work is kept and marked verified, and new starts stay blocked until the clock is fixed. The rule is that integrity checks limit what can be started or claimed, and they do not quietly delete real work. A sync audit screen makes each decision visible.

What I built and shipped

  • Architecture: long-running Rust services started at boot, covering the heartbeat, both sync engines, activity, screenshots, app usage, settings, the trusted clock, notifications and the updater. The core is about 23,500 lines of Rust and the UI about 7,900 lines of TypeScript.
  • Evidence: screenshots at an organization-set interval with optional blurring, each hashed with SHA-256 and uploaded straight to Cloudflare R2 through presigned URLs, so image bytes never pass the app server.
  • Security: the local database is encrypted with SQLCipher and the key sits in the OS keychain, release builds fix the server URL at compile time, payloads are HMAC-signed and the webview runs under a strict content security policy.
  • Desktop behaviour: tray residency, a global shortcut, autostart, native notifications and deep links that start a timer on a given task from the web app.
  • Updates: signed artifacts checked every four hours and on reconnect. I added a server-controlled silent mode: updates download in the background and install only when no timer is running.

The macOS 15 crash

An upstream input-listening crate crashed on every keypress on macOS 15. I did not fork it or drop it. I vendored a copy with the one fix applied and nothing else changed, so a reviewer can see the whole difference at a glance.

Results

  • The tracker is in production on Windows and macOS from a single codebase, currently at v8.6, with real people tracking against it every working day.
  • Outages, flights, sleep, crashes and forced quits no longer cost users hours, and state reconciles on reconnect.
  • Rebuilding idle as an organization policy closed a class of bugs where worked hours were trimmed.
  • Every server change since launch has been additive, and older builds in the field still track and sync.
All projects

Have a project in mind?

Whether it's an MVP, a dashboard, or a rescue mission on an existing codebase — let's talk about it.

Prefer to talk first? Book a free 30-min call (opens in a new tab)