The visa manager is the smallest project in this blog by line count and one of the largest by daily usefulness, a desktop cockpit that turned four browser tabs and four logins into one window and one master password. This finale is the honest accounting of what a glue project teaches, because glue turned out to have its own engineering discipline.
What actually got built
A PyQt5 application embedding four QWebEngineView browsers, the Fluent Forms entries screen, the visa status admin, the Tawk.to inbox, and the hosting webmail on its 2096 port, declared as a services map and assembled into a styled sidebar over a QStackedWidget so every tool stays alive while another is in front. One named persistent QWebEngineProfile shared by every view, ForcePersistentCookies and a real storage path, so a single sign-in covers the cockpit and survives restarts. A password manager holding credentials only as Fernet ciphertext under a never-stored master password, filling login forms through runJavaScript keyed by each tab’s service tag. And a PyInstaller onefile windowed build with a real name and icon, launching from a double-click on a machine with no Python, errors surfacing in dialogs because the console is gone.
What glue work teaches
That integration is a discipline, not a lesser form of building, the cockpit wrote almost no business logic and still demanded real engineering, profiles and sessions, closure traps in dynamically-built navigation, encryption at rest, packaging physics. That curation is a feature, the app’s value over a browser is precisely what it excludes, four tools, zero tabs of drift, always signed in, the workspace shaped to the work. That state is the feel of quality, persistent sessions and stacked live views are invisible individually and together they are the entire difference between a workspace and a launcher. And that the best projects are often subtractive, this one removed friction from every single day of an existing business, which compounds better than most additions ever do.
A few things people ask me about this
Would a browser extension have been simpler? Simpler to start, weaker where it counts, no unified window, no native vault, no curated separation from the day’s other browsing. The desktop shell is what makes it a workspace.
What breaks over time in an app like this? The services’ own login pages, autofill selectors follow their forms, so a redesign upstream means a selector update here. Glue inherits the motion of everything it holds, smaller maintenance than scraping, same principle.
What transferred from earlier projects? The windowed-app obligations from the shop software, hidden consoles demand visible errors, the key hygiene from Stripe, secrets never in source, and the services-as-data pattern from every plugin, declare, then loop.
Where this leads
The next project stayed on WordPress but moved to the front of the shop window, a reusable testimonial carousel I built once and reused across client sites, and the honest questions it raised about real Elementor widgets versus pasted HTML. That story is next.