Building a Cross-Platform App with Rust, TypeScript, and AI - Without a Framework

I built a URL note-taking app that does more than bookmark. Share a link from any Android app and it automatically extracts the article's content, classifies it as movies, books, or games, generates a social-ready clickbait title, and publishes a structured article to a Supabase-backed website, all from a native desktop and mobile app built with Tauri 2 and vanilla TypeScript.

The stack is intentionally lean: Rust handles the backend (HTML parsing, HTTP fetching, LLM calls, local JSON storage), TypeScript drives the UI with no framework, OpenRouter provides free AI inference for content analysis, and Supabase Edge Functions handle serverless publishing. The Android share intent bridges Kotlin to file to Rust to TypeScript via a lightweight file-based IPC pattern that needs no messaging library.

The biggest optimization was consolidating what was originally two separate AI calls into a single OpenRouter request that returns extracted titles, per-title descriptions, a clickbait headline, and category classification in one JSON response. Publishing runs as a background process with 3x retry and user-facing error recovery, so the user saves their note instantly and the article goes live silently. A local cache avoids reprocessing URLs that have already been seen.

The result: a 12 KB JS bundle, ~15 MB desktop installer, full Android share integration, 34 unit tests across Rust and TypeScript, and a pipeline that takes a shared URL from raw link to published article with one AI call and zero user friction.

The app is live at share.vikaph.com. The Android app itself is local-only and built for personal use, so the sharing and publishing pipeline runs entirely from my device. The published articles on the website are what drive the Threads and Instagram autopost experiment, driving traffic back to Vikaph as another distribution channel for MSME content.