Shipped a full Electron desktop app with local AI integration, solo — from "installer wouldn't even build" to Microsoft Store submission in
Context: I've been building UluP Spaces (visual project canvas) for about a month, and a few days ago decided to extend the same "canvas instead of a list" idea to the actual desktop — files and folders as nodes you can drag, connect, and organize, instead of a traditional file explorer.
What ended up in v1.0:
- Real position persistence (nodes stay exactly where you put them, forever — this was actually a real gap I found while dogfooding, not something I planned from day one)
- Real image thumbnails and text previews on the canvas, not generic icons
- Drag-and-drop import straight from File Explorer
- Cut/move/delete (Recycle Bin, never permanent) both via right-click and a guided mode
- Export any folder straight into my other product (UluP Spaces) as a shareable cloud project — every subfolder becomes a node, every file becomes a task with the real file attached
The part I didn't expect to build but ended up being the most fun: it runs a local MCP server that Claude Code can connect to. You can just talk to it — "organize this folder", "connect these two files" — and it happens live on the canvas. Since it's a local stdio server (not hosted), there's no extra API cost if you already have a Claude subscription — genuinely free.
Some of the less glamorous stuff that ate real time: getting electron-builder configured for both a direct-download NSIS installer AND a Microsoft Store APPX package from the same config, setting up auto-updates through a generic feed since GitHub Releases wasn't an option for me right now (unrelated account flag issue), and a security pass converting a few command-execution calls from string-interpolated shell commands to argument arrays after realizing the injection risk pattern.
Submitted to the Microsoft Store today, currently in review. Direct download is live now.
Genuinely curious what people think of "files as a spatial map" as a concept — does it click, or does it feel like solving a problem nobody has? Also happy to go deeper on the MCP/Electron architecture if useful to anyone building something similar.
Does the spatial map actually change how people organize files, or is it mainly a nicer way to browse them?
Honest answer: too early for me to claim real behavioral data — I don't have enough users yet to say "people organize differently" with confidence. So take this as reasoning, not proof. Where I think it's genuinely different, not just prettier: two specific affordances a normal file browser doesn't give you. First, position is remembered per-file, permanently — so spatial clustering itself becomes a real organizing signal (things I put near each other stay near each other), not just alphabetical/date sorting. Second, connections between files don't require a shared folder — I can link a note to an asset it references even if they live in completely different places on disk, which a tree view structurally can't represent at all.
Where I'd be lying if I said it changes things: for simple "open the file I already know the name of" browsing, it's probably not better, maybe even slower than search. The place I actually believe it changes behavior is project-level thinking — seeing the shape of what depends on what, at a glance, instead of holding it in your head while tabbing through nested folders.
Genuinely open to being wrong on this once there's more usage data. Appreciate the question — it's the right one to be asking before believing your own pitch too much.
That’s a very honest distinction — especially separating what you can reason about from what you can actually claim from usage. Appreciate the candid answer, Manu.
For anyone curious: ulupstudio.com/desktop — free download, Microsoft Store submission in review.