Weirling vs NewsBlur
NewsBlur is the closest analog to Weirling: it has a genuinely trainable per-user ranking (the Intelligence Trainer) and is open source. The difference is what the training is based on.
| NewsBlur | Weirling | |
|---|---|---|
| Ranking basis | Train on likes and dislikes of keywords, authors, tags (transparent, Bayes style) | Semantic embeddings plus real reading telemetry (scroll depth, active time, completion) |
| Interest model | Per-feed classifiers | Decaying interest graph with two-hop topic relevance |
| Signal source | Explicit like and dislike training | Reading behavior, explicit feedback and what you open from suggestions |
| Pricing | Free (64 sites); Premium $36/yr (1,024 sites); Archive $99/yr (4,096); Pro $29/mo (10,000) | Self-hosted: free (MIT). Hosted: Reader $6/mo or $54/yr, Pro $12/mo or $108/yr, 30-day Pro trial |
| Self-hosting | Yes (open source, Docker) | Yes (Docker Compose, MIT) |
Where Weirling differs
- NewsBlur learns from what you mark; Weirling learns from what you actually read, without extra clicks.
- Weirling builds a topic graph across sources, so related articles rank even when they share no keywords.
- Every Weirling pick comes with the reason, in plain words.
Both are open source and self-hostable. NewsBlur is more mature; Weirling's difference is ranking by reading behavior and meaning.
Prices and features last checked October 21, 2026. Sources: