Readfine

Readfine is a self-hosted web RSS reader inspired by Tiny Tiny RSS. Besides standard RSS/Atom feeds it scrapes sites that have no feed (CSS selectors), pulls clean readable text out of an article, and saves any link you paste, even one that is in none of your feeds.

It puts the articles you care about first. List a few topics and every new article gets a relevance score, with no AI and no API key involved: the title and opening are matched against your terms, and the app suggests terms to add or drop based on what you actually read. A saved search becomes a sidebar entry that reads like a feed, and a new account starts with one, "Top picks", the best of the last week. When several of your feeds cover the same event, they collapse into one row that says how many sources filed it, so a big story stops flooding the list. A filter engine handles the rest (conditions to actions: label, mark read, star, archive; regex, AND/OR, score thresholds, feed and folder scoping).

Reading is a 2- or 3-panel adaptive layout, plus a real mobile layout and an installable app (PWA), not a squeezed-down desktop. Keyboard shortcuts follow Inoreader and Miniflux, and the Tiny Tiny RSS keys work too. Folders, labels, full-text search (Chinese, Japanese and Korean included), dark mode, reading typography and list density, YouTube and Vimeo playing inside the reader, and OPML import and export with Tiny Tiny RSS compatibility. A new account can start from a set of well-known feeds, import an OPML file, or add feeds by hand.

Everything above works with no AI at all. On top of it sits an optional, opt-in, bring-your-own-key layer: summaries, AI relevance scoring that goes by meaning rather than words, chat over an article, an interest profile, and "Catch me up" digests with scheduled briefings, all with customizable prompts. Anthropic, OpenAI and Google Gemini are supported, as is any OpenAI-compatible endpoint, so scoring and summaries can run on a local Ollama, LM Studio or vLLM and never leave your machine. Nothing is sent to a model until you add an API key of your own.

Stack: Python/FastAPI, PostgreSQL, HTMX and Tailwind, deployed with Docker Compose and licensed AGPL-3.0. A hosted instance is available for people who would rather not run it themselves.