NewsBlur

Self-hosted personal news reader, honestly reviewed. No marketing fluff, just what you get when you run your own instance.

TL;DR

• What it is: Open-source (MIT) personal RSS reader with machine-learning intelligence filters, social sharing, and native mobile apps — think Feedly, but the source code lives on your server [1].
• Who it’s for: Founders, journalists, researchers, and heavy news readers who want control over their feed data, a permanent story archive, and inbox-zero-style reading habits without paying SaaS rates forever.
• Cost savings: Feedly Pro runs $8/mo ($96/yr). Inoreader Pro runs $7.99/mo. NewsBlur Premium is $36/yr — less than five months of either competitor. Self-hosted is the software cost itself: $0 [1].
• Key strength: Intelligence training that learns your preferences per-feed — hide stories you don’t care about, surface stories you do — combined with a permanent story archive and one of the rare MIT-licensed self-hostable readers that hasn’t been abandoned [1][README].
• Key weakness: The UI is dated by modern standards. The free tier is capped at 64 sites. Self-hosting requires Docker, PostgreSQL, MongoDB, Redis, and Elasticsearch — a genuinely complex stack for a tool in this category [README].

---

What is NewsBlur

NewsBlur is a personal RSS reader built by Samuel Clay, a solo developer who has been running the service since before Google Reader shut down in 2013. The tagline — “A new sound of an old instrument” — is a fair description. It’s an RSS reader, which is a technology that most of the internet considers solved-and-abandoned. What NewsBlur adds is a trained intelligence layer: classifiers that learn what kinds of authors, tags, titles, and story content you want to see, and hide or highlight accordingly. You don’t just subscribe to feeds; you train them [README][1].

The project sits at 7,378 GitHub stars. It’s MIT-licensed, meaning you can self-host, fork, or embed it without a commercial agreement. The tech stack is notably heavy: Django/Python 3.7+ on the backend, Backbone.js on the frontend, PostgreSQL for relational data, MongoDB for stories and read states, Redis for caching, Elasticsearch for search, and Celery for background feed fetching. That stack is not accidental — it’s what you need to handle real-time RSS at scale across thousands of concurrent readers [README].

The social angle is genuinely unusual for an RSS reader. Every user gets a public “blurblog” — a curated feed of stories they share with commentary. You can follow other users’ blurblogs, effectively turning RSS into a social layer built on top of the open web rather than a closed platform algorithm [1].

---

Why people choose it over Feedly, Inoreader, and the alternatives

The “Google Reader refugee” angle explains a lot of why NewsBlur has the audience it does. When Google shut down Reader in 2013, every RSS reader picked up refugees — but NewsBlur was already self-hostable, already MIT-licensed, and already had the intelligence training concept that the others were just starting to build.

The decision criteria that come up repeatedly across the community:

Intelligence training. This is the feature no other self-hostable RSS reader does comparably well. You can train by author name, tag, title keywords, or full text content. Classifiers are per-feed, not global — so you can hide all sports coverage from a general news feed while keeping it in a dedicated sports feed. On the Archive and Pro tiers, you can apply training globally across all feeds, or train on full story text, not just headlines [1]. Feedly has a similar feature called “Mute Filters,” but it’s gated behind their Pro plan at $8/mo and is less granular.

Permanent archive. The Archive tier ($99/yr) keeps every story you’ve ever subscribed to, searchable forever. Feedly doesn’t offer this. Inoreader’s article archiving is limited. If you treat your news reader as a research archive — journalists, analysts, researchers — this is the tier you want and it’s the feature that has no good free alternative [1].

Self-hosting. The meaningful distinction over Feedly and Inoreader is that you can run your own instance with full data control. No vendor can raise your bill, deprecate features, or sell your reading patterns. The story data (what you read, when, for how long) stays in your database [README].

MIT license. Feedly, Inoreader, and Readwise Reader are all closed-source SaaS. If the company folds or raises prices, you have no recourse. With NewsBlur, the codebase is yours [README].

What drives people away: the UI hasn’t been redesigned in years. The interface density and visual style is closer to 2014 than 2026. Feedly and Inoreader both have significantly cleaner modern interfaces. If aesthetic polish matters to you, you’ll notice it within five minutes.

---

Features: what it actually does

Reading:

• Real-time RSS with push updates [1]
• Original site view — reads content in context, not stripped [1]
• Full-text extraction for truncated RSS feeds [1]
• Four story layouts per feed: Grid, List, Split, Magazine [1]
• Dark mode across web, iOS, and Android [1]
• River of News — read entire folders as one scrollable stream (Premium+) [1]
• Offline reading on mobile [1]

Intelligence and filtering:

• Train by author, tag, title keywords, full text [1]
• Natural language classifiers (highlights green, hides orange-red) [1]
• Regex training (Premium Pro only) [1]
• Apply training globally or by folder (Archive+) [1]
• Story Clustering — groups duplicate stories across feeds (Archive+) [1]

AI features:

• Ask AI questions about any story [1]
• Daily Briefing — customizable daily summary of top stories (Archive+) [1]
• MCP server and CLI integration for connecting AI agents (Archive+) [1]

Organization:

• Full-text search across all subscriptions (Premium+) [1]
• Saved Searches with their own feed views [1]
• Story Tagging with searchable tags [1]
• Story Archive — every story kept forever (Archive+) [1]
• Track Changes — see how a story was edited post-publication [1]

Social:

• Blurblog — public or friends-only shared story feed [1]
• Follow other users’ blurblogs [1]
• Reading stats [1]

Content types beyond standard RSS:

• Email newsletters (receive via a unique address, read in the feed) [1]
• YouTube channels [1]
• Web Feeds — scrapes any website for content, even without RSS (Archive+) [1]

Third-party integration:

• Supports Reeder, ReadKit, Unread, and others via NewsBlur API [1]
• IFTTT and Zapier hooks [1]

Self-host:

• Docker + Docker Compose install [README]
• Makefile with common operations (make, make log, make shell) [README]
• Full database access: PostgreSQL, MongoDB, Redis [README]

---

Pricing: SaaS vs self-hosted math

NewsBlur tiers:

• Free: up to 64 sites, real-time RSS, basic training [1]
• Premium $36/yr: up to 1,024 sites, search, text view, River of News, custom RSS feeds, 5× faster fetching [1]
• Premium Archive $99/yr: up to 4,096 sites, full archive, Ask AI, Daily Briefing, Web Feeds, story clustering, MCP server, global training [1]
• Premium Pro $29/mo ($348/yr): up to 10,000 sites, 5–15 minute fetching for all feeds, regex training, priority support [1]

Self-hosted:

• License: $0 (MIT) [README]
• Infrastructure: $8–20/mo depending on your VPS and whether you split services across containers
• Note: self-hosting gives you full feature access — there’s no “community edition” capped feature list, unlike some tools in this category

Competitor comparison:

Service	Free	Entry Paid	Power Tier
NewsBlur	64 sites	$36/yr	$99/yr archive
Feedly	100 sources	$8/mo ($96/yr)	$18/mo ($216/yr)
Inoreader	150 sources	$7.99/mo ($95.88/yr)	$14.99/mo ($179.88/yr)
Readwise Reader	No RSS	$7.99/mo ($95.88/yr)	Same

Self-hosting math for a typical power user:

Say you’re following 300–400 feeds with full-text search and a story archive. That’s comfortably in the Archive tier if you’re using the SaaS version at $99/yr. On Feedly’s Pro+ that’s $18/mo ($216/yr). Self-hosted NewsBlur on a $10/mo VPS (with enough RAM for the full stack): $120/yr. Over three years, self-hosted saves you ~$288 versus Feedly Pro+ and requires an upfront time investment of a few hours of setup.

The stronger case for self-hosting NewsBlur isn’t the monthly bill savings — it’s data permanence. The Archive tier’s “every story kept forever” promise is only as durable as the company that runs the SaaS. Self-hosted means the archive is in your MongoDB instance, and it’s yours as long as you maintain the server.

---

Deployment reality check

The self-hosted install path is git clone, cd NewsBlur, make. That’s genuinely three commands. What makes it harder in practice is the stack underneath [README].

What you actually need:

• A Linux VPS with at least 4GB RAM — the combination of Django, MongoDB, PostgreSQL, and Redis is memory-hungry at idle. 2GB is technically possible but will be tight [README]
• Docker and Docker Compose installed
• A domain name, reverse proxy (nginx or Caddy), and TLS certificate if you want HTTPS (the default is a self-signed cert the browser will complain about)
• Storage: MongoDB stores all stories and read states — plan for this to grow. 100+ feeds with full archive means GB, not MB, over time

What you actually get after make:

• Web interface at localhost
• All services containerized (web, celery workers, mongodb, postgres, redis, node services)
• Makefile shortcuts for logs, shell access, and database connections [README]

What can go sideways:

The Elasticsearch dependency is optional but required for full-text search. It’s the most memory-hungry service in the stack and commonly the reason a self-hosted instance runs sluggishly on a $5 VPS [README].

Feed fetching at scale is a real infrastructure problem. The SaaS handles this across a distributed system. Self-hosted, you’re running Celery workers that fetch every feed on a schedule. If you have 500 feeds and want 15-minute updates on all of them, you need enough Celery worker capacity to actually finish the queue. On a small VPS, some feeds will consistently be slower than the nominal update frequency suggests.

The bootstrap_popular_feeds command pre-seeds the database with curated RSS fixtures. This is useful for testing and demo purposes, but be aware it doesn’t fetch content — it just creates records [README].

Realistic time estimate: A technical user who has deployed Docker apps before: 1–2 hours to a working instance. A non-technical founder following a guide: half a day, with a real chance of needing to ask for help on the MongoDB + Elasticsearch configuration. If you’ve never run a multi-service Docker stack, this is a harder first project than most tools in this category.

---

Pros and cons

Pros

• MIT licensed with no feature tiers in self-hosted. Unlike some open-source tools that gate “enterprise” features behind a commercial license, self-hosting NewsBlur gives you the full feature set. The paid SaaS tiers fund the service, but the code is fully open [README].
• Intelligence training is genuinely useful. Not a gimmick. Classifiers that train per-feed on author, tag, title, and full text are meaningfully different from the blunt “mute keyword X” filters most readers offer [1].
• Permanent story archive on Archive+ tier. For researchers and journalists tracking a beat over years, this has no good alternative among RSS readers [1].
• Social layer that doesn’t require a platform. Blurblogs turn your curated reading into a public feed others can follow via RSS. It’s opt-in social without the algorithm [1].
• Email newsletters as feeds. Receive newsletters to a unique NewsBlur address and they appear in your feed reader alongside your RSS subscriptions [1].
• Actively maintained solo project. Samuel Clay has been building and running this for 15+ years. The GitHub repo shows consistent commits. This isn’t abandonware [README].
• Native apps, not wrappers. iOS, macOS, and Android apps described as first-class, not a WebView wrapper. Available on F-Droid for degoogled Android installs [README].
• Third-party reader support. Works with Reeder, ReadKit, Unread, and others via the NewsBlur API. You’re not locked into the native interface [1].

Cons

• Stack complexity. PostgreSQL + MongoDB + Redis + Elasticsearch + Celery + Node.js services is an unusually heavy dependency tree for a news reader. Compare to FreshRSS, which runs on PHP + SQLite. Self-hosting NewsBlur is closer in complexity to running a small SaaS than deploying a simple web app [README].
• UI is dated. The web interface hasn’t had a design refresh in years. It works, it’s functional, but you’ll notice the gap immediately if you’re coming from Feedly or Inoreader.
• Free tier is stingy. 64 sites is below what most active news readers need. Feedly’s free tier gives 100 sources, Inoreader gives 150. The jump to $36/yr is reasonable, but the free tier is a weak entry point [1].
• Solo maintainer risk. NewsBlur is largely one person’s project. This has been its strength (consistent vision, no VC pivot pressure) and its risk (single point of failure for the SaaS, bus factor of one for core development). The MIT license mitigates this for self-hosters, but it’s worth knowing [README].
• MCP and AI features are Archive-tier locked at $99/yr on SaaS. Self-hosted, you get them, but you’re also managing the Elasticsearch dependency they require [1].
• Limited documentation on self-hosting edge cases. The README is clear on the happy path. Configuration options for production hardening (email setup, S3-backed storage for media, scaling Celery workers) require reading the code or asking the community.

---

Who should use this / who shouldn’t

Use NewsBlur if:

• You’re a journalist, analyst, or researcher who treats their RSS reader as a primary work tool and needs a permanent, searchable story archive.
• You’re paying $8–18/mo for Feedly or Inoreader and the intelligence training in those tools feels insufficient — NewsBlur’s classifier system is more granular.
• You want a genuinely MIT-licensed self-hostable reader where the full feature set is available without a commercial license.
• You’re comfortable with Docker and multi-service deployments, or you’ll pay someone to set it up once.
• You want to escape the social media timeline but still stay informed — blurblogs give you a human-curated social layer built on the open web.

Skip it (choose FreshRSS instead) if:

• You want a self-hosted RSS reader with minimal infrastructure overhead. FreshRSS runs on PHP + SQLite or MariaDB and is significantly easier to deploy and maintain.
• You don’t need intelligence training, archiving, or native mobile apps — just a reliable feed sync with a clean interface.

Skip it (choose Miniflux instead) if:

• You want a minimalist, fast, Go-based reader that runs on a single binary with PostgreSQL. No MongoDB, no Celery, no complexity.
• You primarily use third-party clients (Reeder, Unread) and want a rock-solid sync backend.

Skip it (stay on Feedly or Inoreader) if:

• You’re not willing to manage a multi-container Docker deployment. The SaaS version of NewsBlur at $36–99/yr competes directly with Feedly/Inoreader pricing but loses on UI polish.
• Your team needs collaborative features — shared folders, multi-user boards. NewsBlur’s social features are individual-first, not team-first.

Skip it entirely if:

• You have fewer than 50 feeds. At that scale, the free tier of any reader is adequate and self-hosting isn’t worth the overhead.

---

Alternatives worth considering

• FreshRSS — Open source, PHP-based, far simpler to self-host. Lacks intelligence training and permanent archive. The right choice if you want self-hosted RSS without operational complexity.
• Miniflux — Go binary + PostgreSQL, extremely lightweight, strong API. Used as a backend for Reeder and other clients. No native mobile app of its own.
• Feedly — The most polished commercial RSS reader. $8/mo Pro tier. Closed source, no self-hosting, strong AI-powered filtering (“Mute Filters,” “Priority Reads”).
• Inoreader — Close Feedly competitor with slightly better power-user features (rules, filters). $7.99/mo. Also closed source.
• Readwise Reader — Not a traditional RSS reader — it’s a read-it-later service (like Pocket) with RSS integrated. $7.99/mo. Strong for document management and annotation; weak for high-volume feed reading.
• Feedbin — Clean, minimal, $5/mo SaaS. Open source (available for self-hosting). Good third-party client support. Lacks intelligence training.
• tt-rss (Tiny Tiny RSS) — The veteran self-hosted option. PHP-based, feature-rich, ugly. The developer has a famously hostile support community. Not recommended unless you’re already invested.

For someone escaping a SaaS RSS reader bill, the realistic self-hosting shortlist is NewsBlur vs FreshRSS vs Miniflux. NewsBlur wins on features (training, archive, native apps). FreshRSS wins on deployment simplicity. Miniflux wins on performance and API quality.

---

Bottom line

NewsBlur is the most fully-featured self-hosted RSS reader that still has an active maintainer, native mobile apps, and genuine intelligence training. For the target user — a journalist, researcher, or news-heavy founder who treats their feed reader seriously — the $99/yr Archive SaaS tier or a self-hosted instance offers capabilities that Feedly and Inoreader charge twice as much for, with the added option of running your own permanent archive in your own database.

The honest caveats: the UI is showing its age, the self-hosted stack is genuinely complex compared to simpler alternatives like FreshRSS or Miniflux, and the project’s long-term resilience depends largely on one maintainer’s continued interest. The MIT license mitigates that last risk for self-hosters. If you’re committed enough to RSS reading to run your own infrastructure, NewsBlur is the most capable option in the space. If you want simplicity, look at Miniflux or FreshRSS instead.

---

Sources

1. NewsBlur — Official Website, Features & Pricing — newsblur.com. https://newsblur.com/site/
2. NewsBlur — Site Feed Page — newsblur.com. https://www.newsblur.com/site/274075/economic-policy-institute-feed
3. NewsBlur — Social/Blurblog Page — newsblur.com. https://www.newsblur.com/social/51320/donkeyrock
4. NewsBlur — Site Feed Page — newsblur.com. https://www.newsblur.com/site/7736103/espacorecomecars-favorite-links-from-diigo

Primary sources:

• GitHub repository and README: https://github.com/samuelclay/newsblur (7,378 stars, MIT license)
• Official website: https://www.newsblur.com
• Pricing page: https://www.newsblur.com (pricing section)