NOTE: This is pre-ALPHA quality code! Please expect everything to be broken.
- 2013-11-18: bre: Hacking on search/address and gravatar sync
- 2013-11-12: smari: Internationalization work!
- 2013-11-18: bnvk: Tweaking the compose UI
- 2013-11-18: Run
setupandrescanto enable GPG and gravatar importers - 2013-11-08: New API endpoint /search/address for to/cc/bcc autocomplete
- 2013-11-05: CLI hackers: check out
help hacksandhacks/pycli - 2013-10-29: Tag metadata greatly was enhanced (please run
mp --setuporsetupin the Mailpile CLI to update your config). - 2013-10-28: python-gnupginterface is no longer a dependency.
- 2013-10-22: New config format live, may break many things.
Mailpile (http://www.mailpile.is/) is a free-as-in-freedom personal e-mail searching and indexing tool, largely inspired by Google's popular proprietary-but-gratis e-mail service. It wants to eventually become a fast and flexible back-end for awesome personal mail clients, including webmail.
WARNING: Mailpile is still experimental and isn't actually very useful yet. It'll tell you that you have mail matching a given search and let you sort it, browse threads and read messages... but the user interface and message composing/sending functionality is still very immature. If you just want a useful tool and aren't interested in hacking on the code, you should probably check back later or follow @MailpileTeam on Twitter and watch for updates.
Mailpile is developed on a Debian 7 system, running:
- Python 2.7
- python-imaging 1.1.7
- python-lxml 2.3.2
- python-jinja2 2.6
It might work with other versions, most of these packages can be installed
directly using apt-get install. :-)
You also need your e-mail to be in a traditional mbox formatted Unix mailbox, a Maildir or a gmvault backup repository.
On Debian, this should work:
$ sudo apt-get install python-imaging python-jinja2 python-lxml
Alternately (and on other operating systems) you can use Python's PIP tool to install the required packages:
$ pip install -r requirements.txt
Note that installing lxml may require certain C header files that are not necessarily included on your machine. For Debian-based distributions, this can be fixed by running:
$ sudo apt-get install libxml2-dev libxslt1-dev
as per this Stack Overflow answer.
For best results, the first step is to tell the program your e-mail
address and set up basic tags (New, Inbox, etc.) and filters so
Mailpile will behave like a normal mail client. Mailpile can do this
for you, but if you are importing lots of old mail, you may want to
postpone the filter definition until after the import (see below), to
start with a clean slate:
$ ./mp --setup
$ ./mp --set "profiles.0.email = [email protected]"
$ ./mp --set "profiles.0.name = Your Name"
...
If you do not have a local working mail server in /usr/sbin/sendmail,
you may also want to configure a default outgoing SMTP server:
$ ./mp --set "profiles.0.route = smtp://[email protected]:[email protected]:25"
..
Mailpile does not by default access IMAP or POP3 servers directly, it
relies on other tools (such as fetchmail) to take care of downloading
new mail.
Note: You can add multiple accounts by replacing the 0 in the profile
variable name with higher numbers.
Mailpile will create and use a folder in your home directory named
.mailpile for its indexes and settings.
A simple test run might look like so:
$ ./mp --add /var/spool/mail/YOURNAME --rescan
The program prints details of its progress as it runs. Note that just opening the mailbox may take quite a while if it is large enough (it takes about a bit over a minute to open my 500MB mailbox). Once the mailbox has been opened, my laptop (a 1.66Ghz Intel Atom with a 5400rpm HDD) the program can index roughly four messages per second, so if you are processing thousands of messages you should expect it to take a few hours.
You can repeat the add command to specify multiple mailboxes. Boxes can be in mailbox or maildir format. Boxes are not recursive, though. If you have many maildirs in a tree, you must specify each one individually.
Stopping the program with CTRL-C is (relatively) nondestructive - it will try to save its progress and re-running should continue the scan from where it left off.
Mailpile has a built-in web server and will eventually include a proper web-based interface for searching, reading and composing e-mail.
The web interface currently has just one input field, where you can
type terms to search for. If you start the line with a / character
you can use any of the normal CLI commands, including viewing tags or
reading messages (/view 1-15).
Maybe someday you will build a fancier UI for us. :-)
If you want to run the web UI without the CLI interface, start the program like this:
$ ./mp --www
The server listens on localhost:33411 by default, meaning you cannot
access it from a different computer (for security reasons). You can change
the host and port by setting the http_host and http_port variables
(more about internal variables below).
For example if you want to run the server to be accessible
from another computer as well, you can run Mailpile
with:
$ ./mp --set sys.http_host=0.0.0.0
Setting sys.http_host to disabled disables the server.
The most important command Mailpile supports is the search command.
The second most important is probably help. :-)
All commands can be abbreviated to only their first character (the less commonly used commands use capital letters for this).
Some searching examples:
$ ./mp
mailpile> search bjarni einarsson
...
mailpile> search subject:bjarni
...
mailpile> search from:bjarni to:somebody
...
mailpile> search from:bjarni -from:pagekite
...
mailpile> search group:family -from:mom
...
mailpile> s att:pdf
...
mailpile> s has:attachment
...
mailpile> s date:2011-1-30 +date:2011-1-29
...
mailpile> s year:2011 month:12
...
mailpile> s dates:2011-12..2012-04-15
...
mailpile> s mailbox:path/fragment/or/filename
...
The default search will search in message bodies, from lines, attachment
names and subjects. Using a to/from/subject/att/... prefix will
search that part of the message only. There's no way to only search
bodies, they're too full of crap anyway.
Adding terms narrows the search, unless the extra terms are prefixed with
a +, then results are combined. Prefixing with - removes matches for
that term instead.
You can paginate through results using next and previous.
To view a message, use the view command with the number of the result
or one of the magic words all or these:
mailpile> search year:2011 month:12
...
mailpile> view 1 2 6
...
(Mailpile currently assumes you have less installed and in your path for
viewing e-mail. This is a temporary hack.)
You can also search from the command line with mp -s term,
but that will be a bit slower because the metadata index has to be
loaded into RAM on each invocation.
Here is a brief list of the special search terms:
all:mail All messages
att:<word> Search within attachment file names
dates:<B>..<E> Search dates from B to E
in:spam Same as tag:Spam
in:trash Same as tag:Trash
is:unread Same as tag:New
group:<name> Messages from people in a group
has:attachment Messages with attachments
has:pgp Messages with signed or encrypted content
togroup:<name> Messages to people in a group
The order command lets you sort results. Available sort orders
are: index, random, date, from and subject. Threading
may be disabled by prefixing the order with flat-, and the order
may be reversed by further prefixing it with rev-. Examples:
mailpile> order rev-subject # Reverse subject order
...
mailpile> order rev-flat-date # Flat reverse date order
...
mailpile> order # Default sort order
...
You can also change the default sort order by using the order
setting:
mailpile> set order = rev-flat-date # Change default order
...
mailpile> unset order # Use program defaults
...
Mailpile allows you to create tags and attach any number of tags to each message. For example:
mailpile> tag add Inbox
...
mailpile> search to:bre from:klaki
...
mailpile> tag +Inbox all
...
mailpile> inbox
...
The tag command accepts a single tag name, prefixed with a + or -
(for adding or removing the tag), followed by a description of messages.
The message description can be:
allwill affect all messagesthesewill affect currently listed messages- A list of numbers or ranges (
1 2 3 5-10 15)
All these are relative to the last search, so 1 is the first result
of the most recent search and all would be all matching messages.
Tags names are themselves recognized as specialized search commands in
the mailpile CLI.
If you want Mailpile to automatically tag (or untag) messages based on
certain search criteria, you can use the filter command instead:
mailpile> tag add Lists/Diaspora
...
mailpile> search list:diaspora
...
mailpile> filter +lists/diaspora -inbox Diaspora Mail
...
This will tag all the search results and then apply the same rules as new messages are received.
Filters are always processed in a fixed order, so even if one filter
adds a tag, a subsequent one may remove it again. This allows you to
define common patterns such as "All mail goes to the Inbox and is
tagged as new, except this mailing list and that junk mail". Run the
filter command on its own to get a brief summary of how to remove,
edit or reorder the filters.
Mailpile doesn't yet know how to read and index encrypted e-mail, but it
will in the future. In the future Mailpile may also know how to log on to
your remote IMAP and POP3 accounts and download or index remote mail.
This means for sensitive messages, the search index becomes a potential
security risk, as does the configuration file. More broadly, easy access
to all your communications can be a privacy risk in and of itself:
consider the search naked att:jpg as an example. It is almost certainly
worth taking steps to protect your Mailpile.
The simplest and most effective strategy, is to store your .mailpile
folder on an encrypted volume.
Alternately, if you have a GPG key and run Mailpile in an environment where gpg-agent is available for key management, you can tell Mailpile to encrypt its config and data using your key, like so:
$ ./mp --set "prefs.gpg_recipient = [email protected]"
Note that this only encrypts the main index and config file, and only
works if gpg is in your path. The search terms themselves are not
encrypted, which means the contents of individual messages could at
least in part be derived from the index. This problem can be mitigated,
at the cost of some performance, by telling Mailpile to use a one-way
hash to obfuscate the search terms:
$ ./mp --set "prefs.obfuscate_index = Some RaNdoM LongISH SECRET"
Note that if you change this setting, whatever has already been indexed will "disappear" and become unfindable. So do this first if you do it at all!
Mailpile's python code lives in mailpile/.
Mailpile's default HTML templates and Javascript lives in static/default/
Miscellaneous documentation is in doc/.
Test data lives in testing/.
There are a bunch of variables that can be tweaked. For a complete list:
mailpile> help variables
...
To set a variable to some value either run Mailpile with:
$ ./mp --set section.variable=value
Or alternatively run ./mp and issue:
mailpile> set section.variable=value
after which you need to restart the program for it to take effect
(Ctrl+D and ./mp). You can print the value of a variable using:
mailpile> print variable
We are slowly migrating the code to use the doctest module for
internal unit tests.
Black-box regression tests can be invoked by running
scripts/mailpile-test.py. For experimenting and testing, the blackbox
test script can be run in an interactive mode:
$ ./scripts/mailpile-test.py -i
JSON and XML versions exist for most web-based commands and requests and most Mailpile functionality is (or will be) accessible over an HTTP REST-style API.
Please see doc/URLS.md for details.
The Makefile includes a recipe for setting up a virtualenv for use
with Mailpile:
$ make virtualenv
$ source mp-virtualenv/bin/activate
$ mailpile
This allows easy, sandboxed usage.
You can build a docker image:
$ docker build -t mailpile scripts/docker/
and run it:
$ docker run -i -t mailpile
or enter the container's bash prompt directly:
$ docker run -i -t mailpile bash
Searching is all about disk seeks.
Mailpile tries to keep seeks to a minimum: any single-keyword search can be answered by opening and parsing one relatively small file, which should take on the order of 200-400ms, depending on your filesystem and hard drive. Repeated searches or searches for closely related keywords will be up to 10x faster, due to help from the OS cache.
This includes the time it takes to render the list of results.
This level of performance is possible, because all the metadata about the messages themselves is kept in RAM. This may seem extravagant, but on modern computers you can actually handle massive amounts of e-mail this way.
Mailpile stores in RAM about 180 bytes of metadata per message (actual size depends largely on the size of various headers), but Python overhead brings that to about 250B. This means handling a million messages should consume about 250MB of RAM - not too bad if you consider how much memory your browser (or desktop e-mail client) eats up. Also, who has a million e-mails? :-)
(Caveat: Really common terms will take longer due to the size of the result set - but searching for really common terms won't give good results anyway.)
Bjarni R. Einarsson (http://bre.klaki.net/) created this! If you think it's neat, you should also check out PageKite: https://pagekite.net/
The GMail guys get mad props for creating the best webmail service out there. Wishing the Free Software world had something like it is what inspired me to start working on this.
Contributors:
- Bjarni R. Einasson (http://bre.klaki.net/)
- Smari McCarthy (http://www.smarimccarthy.com/)
- Brennan Novak (https://brennannovak.com/)
- Lots more, run
git log |grep Author |sort |uniq -cfor a list!
This program is free software: you can redistribute it and/or modify it under
the terms of either the GNU Affero General Public License as published by the
Free Software Foundation or the Apache License 2.0 as published by the Apache
Software Foundation. See the file COPYING.md for details.