Flux

Flux started as TLDR Checker, a scraper to read TLDR digests without opening my inbox. The second version became a source-agnostic reader: any feed, any newsletter, with articles cleanly separated and content faithful to the original.

Extraction uses recipes generated by an LLM for each source, but the model never rewrites the text: it finds where the content is, and the content is copied as is. That rule keeps hallucinations out and makes sure nothing gets lost.

• A cron captures the feeds and stores the original right away, then a Cloudflare Queues consumer does the extraction: if something fails, the raw data is already there.
• A PWA with offline reading, saved items, collections, search and push notifications with a twice-daily digest.
• Worker in Hono, D1 with Drizzle, better-auth; React 19 frontend with Tailwind and shadcn/ui.

It has been in production since June 6, 2026. The first version, a Node scraper with PocketBase and React, is kept in a branch.