RSS feed builder
Build one RSS feed out of many, and filter it
Several feeds in, one feed out, with rules deciding what survives the merge. What those tools do, what a filter can actually see, and the five things worth checking before you point your reader at one.
What a feed builder does
A feed builder takes feeds in and puts one feed out. What comes back is an ordinary address, so anything that reads RSS can read it without knowing where it came from. All of the work happens between the two ends, and it is three jobs rather than one.
- It merges
- Several addresses in, one address out. Every tool in this category does this, and it is the part that is genuinely easy.
- It filters
- Rules that decide which items survive the merge. A word, a publisher, a tag, something about the item itself. What those rules are allowed to express is most of the difference between one of these tools and another.
- It republishes
- The result is written back out as RSS, Atom or JSON Feed at an address you paste into whatever you already read in. Nothing in the output tells a client it came from a builder rather than from a publisher.
None of the three asks you to change readers, which is usually why somebody is looking for one of these instead of for a new app.
Merging is the easy half
Point five feeds at one output and you get one list in publication order. That holds until the sources publish at different rates, which they always do.
A newspaper puts out forty items a day and a weblog puts out one a week. Merged by time alone, the weblog appears once and is then four hundred items down, and you have rebuilt the thing you left a platform to get away from: the loudest source wins on volume rather than on being worth reading. Most builders stop here, because from the outside the output looks correct. Every item is present, in order.
The fix is that the output has to take turns. Round-robin between sources, a weight per source, or a rule that no single publisher takes more than a share of the result. A tool that does not say which of those it does is doing none of them, and the way to find out is to merge one high-volume source with one quiet one and look at what comes back.
What a filter can actually see
A rule can only test what the publisher put in the file, and feeds differ more than most people expect. This is the single most useful thing to know before writing one, because a rule that matches nothing looks identical to a rule that is not running.
- The headline and the summary
- Always present, and what a keyword rule usually matches. It is also why the same rule behaves differently on two feeds: one carries the whole article and the other carries three sentences, so a word that appears in paragraph nine is there to match in one case and absent in the other.
- The publisher, and the tags they set
- A merged feed loses track of who published what unless the tool deliberately keeps it. Category tags are set by each publisher to no shared standard. Both are useful and neither is consistent across sites.
- The address the item points at
- The most reliable field there is, because a site's sections almost always live under their own path. Matching part of a path is how you drop every sponsored post or every liveblog from one publisher without touching the rest of them.
- When it was published
- Dependable enough to filter on, and the field that lets a rule say today rather than recently.
- How long it is
- Only when the feed carries the article rather than an excerpt. On an excerpt feed a length rule matches nothing in either direction, and it will not tell you that is what happened.
What no rule can see is the article behind the link, unless something went and fetched it. A tool offering to judge an item on its full text is either extracting the page or guessing from the summary, and which one it is worth asking.
Five things worth checking before you rely on one
These are the ones that decide whether a built feed is still working in six months. None of them are usually on a features page.
- Is the output address secret, or guessable?
- A built feed is a description of what you read, which is not something most people mean to publish. An address carrying a long random component is a different proposition from one carrying your username, and the difference only shows up after you have pasted it somewhere.
- Does it keep a copy?
- Merging does not require storage, but most hosted tools store the merged items anyway so the output can be paged backwards. That is a copy of your reading sitting on somebody's server, and it is worth knowing whether it exists and how long it lasts.
- Do the item ids stay stable?
- Every item carries an id, and a client uses it to decide what it has already seen. A tool that regenerates ids on each poll shows you the same articles as new, over and over. It is the most common way one of these fails and it is invisible until the day it starts.
- What happens when a source dies?
- Publishers move feeds without telling anyone. A builder that quietly drops a source that stopped answering produces output that still looks correct, which is the one failure you cannot notice on your own. Something has to say so, in the output or on a screen you will actually open.
- Does it republish somebody's full text?
- Taking a full-text feed and serving it from an address of your own is a redistribution question rather than a technical one, and it gets sharper the more public that address is.
Where Taper fits, and where it does not
Taper is a reader first. It blends the sources of one edition, applies that edition's filters, fills the number of items you set, and ends. An edition can also be served at its own secret address as Atom and JSON Feed, which is what makes it a feed builder for somebody who would rather keep the app they already read in.
Three things it will not do, and they rule it out for some of what people want one of these for:
- It will not stop your reader counting
- The address serves the edition, at the size you set, every time it is polled. Your client will still keep every item it has ever seen and put a number beside the folder, because that is what clients do, and nothing served from this end can prevent it. The filtering travels. The ending does not, and any tool claiming otherwise about your existing app is describing something it does not control.
- It does not page backwards
- A poll gets the current edition and nothing older. There is no way to walk back through what the address served last week, and that is deliberate rather than missing: Taper keeps no copy of an article you did not save, so there is nothing behind the address to page through.
- It does not rewrite items
- No full-text extraction, no rewritten headlines, and nothing generated from a page that publishes no feed. Sources in, filtered edition out.
What it does instead, against the four sections above: sources are never counted on either tier, so merging two hundred of them is a free-tier operation. Three tiers you assign by hand decide what rises, so the newspaper cannot crowd out the weblog and a source you never want to miss is never counted against the size at all. And the filter grammar takes the fields that are actually there:
Boolean logic, so (rust or zig) and not hiring keeps posts about either language and drops the job ads. Matching scoped to the headline, the publisher, a tag or the address, so not url:/sponsored/ takes one section of a site out of one edition. Tests on the item itself, so words:>1200 holds an edition to what somebody sat down to write. Filter sets saved once and used across editions, and a word muted until a date you pick.
Serving an edition at an address is part of Pro, because it is the filtering delivered somewhere else and filtering is what the paid tier is. Getting your sources and editions back out as OPML is free, has no plan attached to it, and is not going to acquire one.
The questions people actually ask
Four of these are about the category rather than about this product, and the fourth is the one a page selling you something would leave out.
Can I combine several RSS feeds into one feed?
Yes, and it is the oldest trick in the category. What separates the tools is not whether they can merge but what the merged output looks like when the sources publish at very different rates, and whether you can filter what comes through. A merge with no ordering rule and no filtering is a longer list, which is rarely what somebody wanted when they went looking.
Can I filter an RSS feed by keyword?
Yes, and it is worth asking what the rule is allowed to look at. A plain word list matches the headline and the summary and nothing else. A grammar with boolean logic in it lets you say two things at once, scope a match to the publisher or the address, and test the item rather than only its words, which is the difference between dropping every article that mentions a subject and dropping one publisher's coverage of it.
Do I need a new reader to use one?
No, and that is usually the point. A builder gives you back an ordinary feed address, so it goes into the app you already have alongside your other subscriptions. Your reader neither knows nor cares that something happened to the feed before it arrived.
Will a built feed stop my reader piling things up?
No, and be suspicious of anything that says it will. Your client polls the address on its own schedule, keeps every item it has ever seen, and puts a figure beside the folder, and no output served from the other end can change any of that. What a builder can genuinely do is make sure that what arrives is worth arriving. Reading something that ends is a property of the app you read in, not of the feed you point it at.
Is this the same thing as an RSS reader?
No. A reader is where you read; a builder is what the feeds go through on the way. Plenty of people use both, and some readers have a modest amount of building in them already, usually as a keyword blocklist. The reason to reach for a separate one is almost always that the filtering in the reader ran out.
What does building a feed cost in Taper?
Merging is free and always will be: sources carry no number on either tier, and neither do the tiers that decide what rises. The expressive filtering, and serving the result at an address your own reader can subscribe to, are what Pro is. Nothing behind the paid tier ever removes an article from your view, which is one of three constraints published so that breaking one is visible.
Decide how much news is enough.
Bring the sources you already read. Set a size. Reach the end of it, and find nothing underneath.
Free. No card, and bring every source you follow.