Cory DoctorowToday's threads (a thread)
Inside: The web is bearable with RSS; and more!
Archived at: https://pluralistic.net/2026/03/07/reader-mode/
1/
Fazal Majid@pluralistic I wrote and have been been using my own feed reader, Temboz, for over 24 years now. The two essential features of any good reader in my book are:
1) filtering. Don't want to hear about the inane Kardashians or the insane Trump any more? Double-click on their occurrence, click "thumbs-down", adjust the filtering setting and boom, mental health restored.
2) giving you control over the order articles are shown, as opposed to the Meta or Google algorithm that does not have your best interests at heart.
viq@fazalmajid
Oh! https://github.com/fazalmajid/temboz
I'll need to look at it, it may be closer to what I've been looking for recently 🤔
Oh! https://github.com/fazalmajid/temboz
I'll need to look at it, it may be closer to what I've been looking for recently 🤔
@viq I am actually in the middle of rewriting it in Rust, and switched my own usage yesterday, so if I were you I'd hold off for a couple of weeks until it stabilizes:
@fazalmajid
Is either of them worth playing around with at this point? Or, will playing with temboz now give me a decent idea what rtemboz is going to be doing and what it will feel like?
Is either of them worth playing around with at this point? Or, will playing with temboz now give me a decent idea what rtemboz is going to be doing and what it will feel like?
@viq the UI is identical, but rTemboz has much better performance and won't hang because it holds database locks while slow feeds are still returning. I've been told setting up Temboz is not for the faint of heart, and I am working on fixing that as well in rTemboz.
@fazalmajid I'm not faint of heart :D At this point I don't care terribly much about performance, I want to see what I can make it do, and how well that cooperates with how I look at things. Which it sounds I can do today with Temboz 🙇
@viq OK, use rTemboz, follow the Docker instructions at https://github.com/fazalmajid/rTemboz?tab=readme-ov-file#running
(you don't need to build from source if you are OK using my docker images on Docker Hub).
I am dogfooding rTemboz at the moment, I am intensely aware of the regressions and highly motivated to fix them as this is the single app I spend the most time in...
@viq there's a partial list on the README.md. What it doesn't list is the bugs already fixed...
@fazalmajid
Ah, the TODO list? I see you've checked some boxes since yesterday :)
Ah, the TODO list? I see you've checked some boxes since yesterday :)
@viq yeah, not being able to add filters was a big drag. I have 22K and counting. You select the stuff you never want to see again, on the page, click "red down arrow" and up pops a dialog as to how you want the rule to be set (title only or title and content, for this feed only or for all).
@fazalmajid
Do you maybe have what I would call "saved searches"? "When I click here, apply this specific set of filters to show me only matching entries [potentially: from this (sub)set of feeds]“ - say, one named "cooking" that has everything about cooking and nutrition and shop discounts (that you were able to figure out terms for)
Also, are filters retroactive, or apply only to new items?
Do you maybe have what I would call "saved searches"? "When I click here, apply this specific set of filters to show me only matching entries [potentially: from this (sub)set of feeds]“ - say, one named "cooking" that has everything about cooking and nutrition and shop discounts (that you were able to figure out terms for)
Also, are filters retroactive, or apply only to new items?
@viq Not really. The Python version has the ability to add a SQL where clause, but that's fairly dangerous and I am not sure I will carry it forward. The filters are not retroactive, usually I will just use the full-text search and quickly mark them down.
@viq Perhaps a better approach would be to tag feeds and allow filtering by feed tag. That actually sounds useful and I might get on it in a couple of months after I've brought it back to parity.
@fazalmajid
Feed tag would be "science news", whereas my current thinking is that I would like to look at "astronomy news", "renewables & power storage improvements" and "developments regarding this particular health thing" separately, in their own groups, where they can come from multiple feeds, some of them having stuff for all of those categories, and some only for some.
Feed tag would be "science news", whereas my current thinking is that I would like to look at "astronomy news", "renewables & power storage improvements" and "developments regarding this particular health thing" separately, in their own groups, where they can come from multiple feeds, some of them having stuff for all of those categories, and some only for some.
@viq Either you build your tag hierarchies, which is a drag, , or you have a LLM classifier do this according to a taxonomy you choose, or you implement some form of vector search with embeddings. I don't want rTemboz to require a GPU to run. The data model already supports human entered tags vs AI, but the infrastructure to actually do it isn't in place yet. And some people have deep objections to LLMs.
Assuming you don't, what would be your preference for integrating AI assistance: self-hosted, API, CPU only, etc?
@fazalmajid
Personally I'd rather not get LLMs involved in this. Thus my idea of "saved search" (if searching for articles is possible), and/or rules for applying tags to articles or otherwise support for "virtual folders" with "some" mechanism for making articles appear in them.
I wonder how well Bayesian analysis would work for such?
Oh, also, does (r)Temboz support fetching full articles if feed has only snippets?
@viq yes, it has full-text search using SQLite's fts5, but not semantic search. It does not fetch full feeds, I've been on the receiving end of a DMCA take-down for just publishing a feed of the articles I personally find interesting.
You don't need a LLM to classify, an embedding like SBERT, GTE or Microsoft E5 combined with vector search would suffice, unless you are philosophically opposed to any kind of neural net tech.
Have you tried it yet? Any first impressions you'd like to share?
@fazalmajid
Huh. Damn, that's trigger-happy :(
LLMs to my knowledge require models to be of any use, and currently AFAIK all of those are of dubious origin. The current situation and how LLMs under a marketing term are shoved into everything, and the wide lens view of "why" is what soured that particular technology for me. AFAIK neural nets are just underlying tech, shared with other uses, and do not require ingesting all of everything without compensation to do useful things.
@fazalmajid
Quick try, "oh, no folders", and not sure how to work with filters, or whether it's even possible right now.
Does "this story did interest me" (or not) result in anything other than SNR score for the feed?
Quick try, "oh, no folders", and not sure how to work with filters, or whether it's even possible right now.
Does "this story did interest me" (or not) result in anything other than SNR score for the feed?
@viq That's right, the approach is inspired by Google News and its "river of news" rather than like an email reader. There was an article recently about how that design creates "phantom obligation" and a form of FOMO:
https://www.terrygodier.com/phantom-obligation
Articles marked "Up" are kept indefinitely, filtered or "Down" articles have their contents purged after 2 weeks to free up disk space. But otherwise, yes, the point of flagging an article is to bump up the feed SNR as a measure of quality, and to keep them for future reference in lieu of a bookrmarking service like the late del.icio.us.
@fazalmajid I think I will need to play with Temboz for at least a bit, rTemboz doesn't currently offer anything beyond "here's a list of articles", which isn't for me enough to get a proper feel for intended functionality.
Out of curiosity, approximately how many feeds (or maybe "articles per day" is better metric?) do you have?
Out of curiosity, approximately how many feeds (or maybe "articles per day" is better metric?) do you have?
@viq 761 feeds, around 1000+ articles/day
@fazalmajid
Hm, any hints which python version to aim for with Temboz? Wrapping with uv, with `uv add temboz sanity` I keep getting complaints about missing `flaskext` even though `flask-sqlalchemy` is installed, so far with python 3.14 & 3.12.
Or is that a case of "yeah, I don't want to deal with that anymore, just wait for rTemboz"?
Hm, any hints which python version to aim for with Temboz? Wrapping with uv, with `uv add temboz sanity` I keep getting complaints about missing `flaskext` even though `flask-sqlalchemy` is installed, so far with python 3.14 & 3.12.
Or is that a case of "yeah, I don't want to deal with that anymore, just wait for rTemboz"?
@viq not sure why you have flaskext and SQLalchemy, I despise ORMs and would never use one.
@fazalmajid pulled in by `sanity`, since `UV run temboz` was complaining about not finding it
@viq after I fix the "word" filter rules, I am going to do autodiscovery. If you go to a page and it doesn't have a link to the RSS feed, Temboz can usually find it through autodiscovery or using some heuristics, it works 90% of the time. rTemboz doesn't have that yet.