Projects Case study
FileComb: one codebase, three platforms, AI that never leaves the machine.
How a web app for working with documents became a native desktop app shipping on Intel Macs, Apple Silicon, and Windows — with a local AI stack where file contents never leave the user's machine.
The product
FileComb is a desktop app for working with documents, with private AI-powered search across them. The defining constraint is privacy: the AI — including vector search across the user's own files — runs entirely on-device. File contents never leave the machine, which isn't a marketing line; it's the architecture.
One codebase, native on three platforms
FileComb started as a web app, then became a native desktop app built with Tauri: a React and TypeScript front end on a Rust core. That choice paid for itself at distribution time — the same codebase ships on Intel Macs, Apple Silicon, and Windows, without maintaining three products or paying Electron's memory tax. The web skills carry over; the native behaviors don't get faked.
The AI stays on the machine
The AI features run locally: on-device inference plus vector search (Qdrant) across the user's files, so semantic search works without uploading anything to anyone. Building AI this way means designing for constrained hardware instead of assuming a data center — smaller models, smarter retrieval, and guardrails in the prompt architecture.
Shipping is part of the product
A desktop app isn't done when it runs; it's done when it installs. FileComb ships with macOS code signing and notarization under a Developer ID — so Gatekeeper doesn't scare users away — and Windows installers. "Simple things, finished" includes the boring last mile.