Introducing Elfeed, an Emacs Web Feed Reader
Unsatisfied with my the results of recent search for a new web feed reader, I created my own from scratch, called Elfeed. It’s built on top of Emacs and is available for download through MELPA. I intend it to be highly extensible, a power user’s web feed reader. It supports both Atom and RSS.
The design of Elfeed was inspired by notmuch, which is my e-mail client of choice. I’ve enjoyed the notmuch search interface and the extensibility of the whole system — a side-effect of being written in Emacs Lisp — so much that I wanted a similar interface for my web feed reader.
The search buffer
Unlike many other feed readers, Elfeed is oriented around entries —
the Atom term for articles — rather than feeds. It cares less about
where entries came from and more about listing relevant entries for
reading. This listing is the
*elfeed-search* buffer. It looks like
This buffer is not necessarily about listing unread or recent entries,
it’s a filtered view of all entries in the local Elfeed database.
Hence the “search” buffer. Entries are marked with various tags,
which play a role in view filtering — the notmuch model. By default,
all new entries are tagged
unread (customize with
elfeed-initial-tags). I’ll cover the filtering syntax shortly.
From the search buffer there are a number of ways to interact with entries. You can select an single entry with the point, or multiple entries at once with a region, and interact with them.
b: visit the selected entries in a browser
y: copy the selected entry URL to the clipboard
r: mark selected entries as read
u: mark selected entries as unread
+: add a specific tag to selected entries
-: remove a specific tag from selected entries
RET: view selected entry in a buffer
(This list can be viewed within Emacs with the standard
The last action uses the Simple HTTP Renderer (shr), now part of Emacs, to render entry content into a buffer for viewing. It will even fetch and display images in the buffer, assuming your Emacs has been built for it. (Note: the GNU-provided Windows build of Emacs doesn’t ship with the necessary libraries.) It looks a lot like reading an e-mail within Emacs,
The standard read-only keys are in action. Space and backspace are for
page up/down. The
p keys switch between the next and
previous entries from the search buffer. The idea is that you should
be able to hop into the first entry and work your way along reading
them within Emacs when possible.
Elfeed maintains a database in
~/.elfeed/ (configurable). It will
start out empty because you need to tell it what feeds you’d like to
follow. List your feeds
elfeed-feeds variable. You would do this in
.emacs or other initialization files.
(setq elfeed-feeds '("http://www.50ply.com/atom.xml" "http://possiblywrong.wordpress.com/feed/" ;; ... "http://www.devrand.org/feeds/posts/default"))
Once set, hitting
G (capitalized) in the search buffer or running
elfeed-update will tell Elfeed to fetch each of these feeds and load
in their entries. Entries will populate the search buffer as they are
discovered (assuming they pass the current filter), where they can be
immediately acted upon. Pressing
g (lower case) refreshes the search
buffer view without fetching any feeds.
Everything fetched will be added to the database for next time you run Emacs. It’s not required at all in order to use Elfeed, but I’ll discuss some of the details of the database format in another post.
The search filter
s in the search buffer will allow you to edit the search
filter in action.
There are three kinds of ways to filter on entries, in order of efficiency: by age, by tag, and by regular expression. For an entry to be shown, it must pass each of the space-delimited components of the filter.
Ages are described by plain language relative time, starting with
This component is ultimately parsed by Emacs’
function. Here are some examples.
Tag filters start with
+, entries must be tagged
with that tag. When
-, entries must not be tagged with that tag.
+unread: show only unread posts.
-junk +unread: don’t show unread “junk” entries.
Anything else is treated like a regular expression. However, the regular expression is applied only to titles and URLs for both entries and feeds. It’s not currently possible to filter on entry content, and I’ve found that I never want to do this anyway.
Putting it all together, here are some examples.
linu[xs] @1-year-old: only show entries about Linux or Linus from the last year.
-unread +youtube: only show previously-read entries tagged with
Note: the database is date-oriented, so age filtering is by far the
fastest. Including an age limit will greatly increase the performance
of the search buffer, so I recommend adding it to the default filter
Generally you don’t want to spend time tagging entries. Fortunately
this step can easily be automated using
elfeed-make-tagger. To tag
all YouTube entries with
(add-hook 'elfeed-new-entry-hook (elfeed-make-tagger :feed-url "youtube\\.com" :add '(video youtube)))
Any functions added to
elfeed-new-entry-hook are called with the new
entry as its argument. The
elfeed-make-tagger function returns a
function that applies tags to entries matching specific criteria.
This tagger tags old entries as read. It’s handy for initializing an Elfeed database on a new computer, since I’ve likely already read most of the entries being discovered.
(add-hook 'elfeed-new-entry-hook (elfeed-make-tagger :before "2 weeks ago" :remove 'unread))
Creating custom subfeeds
Tagging is also really handy for fixing some kinds of broken feeds or
otherwise filtering out unwanted content. I like to use a
to indicate uninteresting entries.
(add-hook 'elfeed-new-entry-hook (elfeed-make-tagger :feed-url "example\\.com" :entry-title '(not "something interesting") :add 'junk :remove 'unread))
There are a few feeds I’d like to follow but do not because the
entries lack dates. This makes them difficult to follow without a
shared, persistent database. I’ve contacted the authors of these feeds
to try to get them fixed but have not gotten any responses. I haven’t
quite figured out how to do it yet, but I will eventually create a
elfeed-new-entry-hook that adds reasonable dates to
In my own .emacs.d configuration I’ve added a new entry action
to Elfeed: video downloads with youtube-dl. When I hit
d on a
YouTube entry either in the entry “show” buffer or the search buffer,
Elfeed will download that video into my local drive. I consume quite a
few YouTube videos on a regular basis (I’m a “cord-never”), so this
has already saved me a lot of time.
Adding custom actions like this to Elfeed is exactly the extensibility I’m interested in supporting. I want this to be easy. After just a week of usage I’ve already customized Elfeed a lot for myself — very specific customizations which are not included with Elfeed.
Elfeed also includes a web interface! If you’ve loaded/installed
elfeed-web, start it with
elfeed-web-start and visit this URL in
your browser (check your
Elfeed exposes a RESTful JSON API, consumable by any application. The web interface builds on this using AngularJS, behaving as a single-page application. It includes a filter search box that filters out entries as you type. I think it’s pretty slick, though still a bit rough.
It still needs some work to truly be useful. I’m intending for this to become the “mobile” interface to Elfeed, for remote access on a phone or tablet. Patches welcome.
Try it out
After Google Reader closed I tried The Old Reader for awhile. When that collapsed under its own popularity I decided to go with a local client reader. Canto was crushed under the weight of all my feeds, so I ended up using Liferea for awhile. Frustrated at Liferea’s lack of extensibility and text-file configuration, I ended up writing Elfeed.
Elfeed now serving 100% of my personal web feed reader needs. I think it’s already far better than any reader I’ve used before. Another case of “I should have done this years ago,” though I think I lacked the expertise to pull it off well until fairly recently.
At the moment I believe Elfeed is already the most extensible and powerful web feed reader in the world.