NewsBlur — Self-Hosted RSS Reader with Intelligence Training

Introduction

NewsBlur is a personal news reader that combines RSS feed aggregation with machine-learning-style intelligence filters. You train it by marking stories, tags, authors, and publishers as liked or disliked, and it adjusts what it shows you.

What NewsBlur Does

• Aggregates RSS and Atom feeds into a unified reading interface
• Trains on your preferences to classify stories as green (like), yellow (neutral), or red (dislike)
• Displays the original site in a reading pane alongside extracted text for flexible reading
• Supports story sharing with built-in social features like blurblogs and comments
• Provides full API access for building custom clients or integrations

Architecture Overview

NewsBlur is a Django application backed by PostgreSQL (users, feeds), MongoDB (stories, social data), Redis (sessions, caching), and Elasticsearch (search). A Celery task queue handles feed fetching in the background. The frontend is a custom JavaScript single-page application. All components are containerized in the official Docker Compose setup.

Self-Hosting & Configuration

• Docker Compose deployment includes all services: Django app, PostgreSQL, MongoDB, Redis, Elasticsearch, Celery workers, and Nginx
• Environment variables configure database connections, secret keys, and email settings
• Feed fetch frequency is configurable per feed or globally via Celery beat schedules
• OPML import supports migrating from other RSS readers
• Backups involve dumping PostgreSQL and MongoDB databases

Key Features

• Intelligence trainer: thumbs-up or thumbs-down on authors, tags, titles, and publishers to filter your feed
• Original site view renders the publisher's page inside NewsBlur for full visual fidelity
• Blurblog: share stories publicly with comments, creating a curated linkblog
• Full-text search across all stored stories via Elasticsearch
• OPML import/export for easy migration between RSS readers

Comparison with Similar Tools

• FreshRSS — PHP-based RSS reader that is lighter to self-host; NewsBlur adds intelligence training and social features
• Miniflux — minimalist Go-based reader focused on speed; NewsBlur has a richer UI and story classification
• Tiny Tiny RSS — long-standing self-hosted reader; NewsBlur offers a more modern interface and built-in intelligence
• Feedly — popular hosted reader with AI features; NewsBlur can be fully self-hosted with similar training capabilities
• Inoreader — feature-rich hosted reader; NewsBlur's open-source codebase gives full data ownership

FAQ

Q: What are the system requirements for self-hosting? A: NewsBlur uses several services (PostgreSQL, MongoDB, Redis, Elasticsearch). A machine with at least 4 GB of RAM is recommended for a personal instance with a few hundred feeds.

Q: Can I import my feeds from another reader? A: Yes. NewsBlur supports OPML import, the standard format used by all major RSS readers for feed lists.

Q: Is there a mobile app? A: Official iOS and Android apps are available. The codebase for both is open source alongside the server.

Q: How does the intelligence trainer work? A: Click the thumbs-up or thumbs-down icons next to an author, tag, title keyword, or publisher. NewsBlur then highlights or hides future stories matching those criteria.

Sources

• https://github.com/samuelclay/NewsBlur
• https://newsblur.com