RSSMonster is an open-source, self-hosted RSS reader built for people who follow a lot of feeds and want better ways to deal with information overload.
It started years ago as a Google Reader-inspired application, but has gradually evolved into a modern three-pane reader with Smart Folders, advanced search, duplicate detection, related-article discovery, event clustering, recommendations, and source-quality signals.
The core philosophy is still RSS-first: feeds remain chronological and fully under the user’s control. The semantic features sit on top to help organize and discover content rather than replacing the feed with an opaque algorithm.
RSSMonster includes a local semantic pipeline using small ONNX models:
- Qwen3-Embedding-0.6B for article embeddings and similarity
- Qwen3.5-0.8B for summaries, inferred tags, and semantic labels
- ModernBERT for quality, promotional-content, and sentiment classification
These can run locally on CPU using Transformers.js. Article embeddings, classification, Smart Folder suggestions, tagging, event/topic labels, and interest modeling therefore do not need to send reading data to an external AI provider.
RSSMonster also exposes authenticated, user-scoped MCP tools for searching articles, retrieving content, resolving feeds and categories, inspecting tags, and refreshing feeds. An optional assistant can use those tools to answer questions such as “show me the most important unread stories this week” or “compare how different sources covered this event.” The agent does not receive unrestricted database access.
Other features include:
- SQLite and MySQL support
- Docker/self-hosted deployment
- OPML import/export
- Fever and Google Reader-compatible APIs
- full-content extraction
- tags and bookmarks
- Smart Folders with composable query expressions
- related articles and duplicate/revision detection
- personalized Recommended ranking
- non-personalized Top Stories ranking
- responsive/PWA interface
- dark mode
A key design goal is keeping the intelligent features explainable. RSSMonster exposes the signals contributing to recommendations and allows weak matches to return no recommendation at all rather than filling the interface with marginal results.
The project currently has around 480 GitHub stars and is under active development.
GitHub:
https://github.com/pietheinstrengholt/rssmonster
Comments and Reviews
Uses NodeJS so not sure how hard it is to setup on shared hosting. Don't have time to research that right now :p