Cross-platform without Electron

Electron is the default answer to "one codebase, every desktop." It is also a Chromium runtime shipped with every app. For PairSync — a tool that should feel native, launch fast, and sip memory while it waits in the background — that tradeoff never sat right. So I moved to a native core with thin platform clients, and this is what it cost.

A core that knows nothing about UI#

The sync engine, transport, and storage live in one portable core with no view layer at all. It exposes a small command surface and emits events; every platform binds to that same contract:

type Command =
  | { type: "pair"; code: string }
  | { type: "send"; path: string }
  | { type: "revoke"; deviceId: string };

The rule is strict: no UI concept leaks into the core, and no platform reaches past the command surface. That boundary is the whole design.

Thin clients, real controls#

Each client is small and native — the file pickers, notifications, and menu-bar presence people expect, wired to core commands. There is no shared widget layer pretending every OS is the same.

What it traded away#

• More surface to build — a native shell per platform instead of one webview.
• Less code reuse in the UI — deliberately. The reuse lives in the core.
• Fewer "just npm install it" escape hatches — you own more of the stack.

In return: cold starts measured in tens of milliseconds, idle memory a fraction of a webview app, and an interface that feels like it belongs on the machine. For a background utility, that was the right bet — even though it was the harder one.