unhosted web apps

freedom from web 2.0's monopoly platforms

9. Sending and receiving email from unhosted web apps

(en Français)

Unhosting email

The posterchild of hosted web apps is arguably GMail. It was one of the first apps to use AJAX to its full potential, and redefined our expectations of what can and cannot be done inside a browser. This also means it is one of the hosted web apps that will be hardest to replace.

The idea of unhosting email is simple: take, essentially, the client-side javascript app that for instance GMail puts into your browser, and make its backend swappable, so that the user can connect their own backend at runtime.

For storing email messages, both received ones and sent ones, we will use remoteStorage. For relaying the outgoing messages and receiving the incoming messages, we will use Sockethub. This kind of pattern should become familiar to you by now if you have been following the previous episodes of this handbook.

What to put on port 25

As I'm writing this, it has been three months since I started using unhosted web apps as much as possible for everything. During this time I have experimented with various incoming mail servers - most of them based on the simplesmtp nodejs package. This has gone reasonably well; I had anticipated losing a small percentage of my incoming mail during this period, and also warned people about this possibility.

Spam levels are also still manageable; I registered this domain name 6 months ago, so only about 50% of incoming messages are spam, and it's not too much work for me to just click through them.

Hosting mailservers is, however, not fun. This is why this week I decided to switch my domain to Gandi and get their standard mail hosting package. You could also of course use GMail as your incoming and outgoing mailserver, that is in itself equivalent.

Pointing the MX records of your domain to an email hosting provider resolves a significant part of the engineering problem, and means we don't have to run anything on port 25 (the standard TCP port for incoming email) ourselves. Instead, our Sockethub server only has to retrieve the email from that email hosting provider, using either POP3 or IMAP.

Sending mail out

Outsourcing the email hosting for your IndieWeb domain to a standard hosting provider also resolves another problem: it gives you a trusted outgoing relay. In the old days, you could send email messages straight from your email client to the receiving mailserver. Nowadays, that is entirely impossible.

First of all, the remote receiving mailserver will probably block your IP address purely based on the fact that it is likely listed as a "residential" IP block. Second, your ISP is likely to not even allow your computer to connect out to any remote port 25 over TCP.

But even if that weren't the case, you would have to configure all sorts of things like reverse DNS matching your email domain in order to get even a chance at delivering an email message somewhere.

Services like Mailgun and SendGrid fix this problem. They are specialized in relaying outgoing emails. Their main customers would normally be hosted web app developers who need to send out notification emails to their users, but if you ask nicely, they will probably also give you an account for your personal outgoing mail.

I have been using SendGrid for the last three months (if you received email from me you may have noticed the 'via sendgrid.me' header), but switching my incoming mail to Gandi will also give me an outgoing relay server that I can easily use.

Fitting Sockethub into the picture

Unhosted web apps can only open http connections, since a few years WebSocket connections, and since a few weeks, PeerConnections. They cannot open SMTP connections, IMAP connections, or POP3 connections.

Sockethub takes on the role of connecting a WebSocket on the side of the unhosted web app with an SMTP, IMAP, or POP3 connection on the side of the remote server, and relaying the communication in two directions. It plays the role of a hub inbetween WebSockets on the one hand, and TCP sockets on the other.

For outgoing email, the unhosted web app sends an Activity with verb 'send' to Sockethub, putting the message into the Activity's object field, the various types of recipients into the target field, and the From: address into the actor field.

Sockethub then transforms this Activity into an SMTP message (using Nodemailer), and gives it to the relay server you specified it to use. The relay server then serves mainly to comply with all the spam rule obligations, and to queue the message in case the receiving mailserver is temporarily down or unreachable.

For incoming email, you tell Sockethub to retrieve email from your mailserver, and put it into your remoteStorage account. At the time of writing, that part is not implemented yet: I am still using a separate script that handles this part.

The Sockethub setup instructions include an example page for sending email. You can also check out meute.5apps.com for inspiration.

Main features of an email message

The email system is super old, at least on the scale of internet history, and has many different partial options and features. At first, I only supported sending email messages with a text-body and a subject from my email address to the recipient's email address.

After a few complaints from several of my friends though, I realised I should really add 'In-Reply-To' and 'References' headers when replying to a thread. I hadn't at first realised that this was so vital, but if you don't do this, then other people's mail clients will start a new thread each time you reply to something. The value of these header fields should be the Message-ID of the email message you are replying to.

Adding sender and recipient names is also useful; since I'm using 'anything@michielbdejong.com', I show up in people's addressbooks as the user 'anything'. That's a bit ugly. The proper way is to specify for instance 'Michiel B. de Jong <anything@michielbdejong.com>' instead. And the same for recipient addresses.

The latest feature I added to my email sending app is support for multiple To: addresses as well as multiple Cc: addresses. It seem SendGrid does not implement Bcc:, and I also never use this feature, so I left that out, at least for outgoing emails.

For incoming emails, I used a separate app at the time of writing (later, I switched to using Meute). This is a bit of a nuisance because I have to cut-and-paste each message I reply to, but as a work-in-progress, I manage with it. I store incoming emails on my remoteStorage in the 'messages' modules, and the way I keep the synchronization overload under control is by having two layers of folders: one layer for the Megasecond, and one layer for the Kilosecond.

This means that at one click of a button I can retrieve all email from the current Kilosecond (about 17 minutes), which is rarely more than two or three messages.

On top of that, each Megasecond (about 11 days), my email is archived to a different folder. The messages are saved in filenames with millisecond precision, so if you send me an email right now, then it will be stored on my remoteStorage as something like: messages/1360/680/123456.json. Whenever I receive an html-only email, or an email with attachments, I have to go into my VPS using the webshell access, to get to that additional content.

I'm telling you all this to make clear that, although we have this working, and I have been using this for months as my email client, it is far from "workable" for end-users, and there is still quite a long way to go before we can say we "solved" the challenge of writing an unhosted email client.

Spam filtering

If you host your own mailserver, then it is advisable to run some sort of spam filter software on there. Strictly speaking we could argue that this is an application-level task that should be handled by the unhosted web app that acts as the email client, but since spam filtering is such an integral part of the email platform nowadays, I think it's defendable to make it part of the server.

Anyway, if you outsource your mail hosting to a hosting provider, they wil take care of the spam filtering for you.

Pretty Good Privacy

Last week, Eben Moglen and Bdale Garbee gave a keynote at FOSDEM about FreedomBox 1.0. As we mentioned last week, the general concept of using a FreedomBox, that's to say a plugserver or other piece of hardware that is always on, running free software in your home, is a central piece of the puzzle we are trying to solve with unhosted web apps.

In the keynote, Bdale made it clear that PGP is a central part of how FreedomBox will provide freedom to its users. And I feel this makes total sense. PGP is a tested system for end-to-end encryption of email messages that people send each other. By itself, the email platform does not support much in the way of encryption and protection against eavesdropping.

Also, the move of more and more users away from desktop-based email clients like Thunderbird and Outlook, and onto webmail services like GMail, makes end-to-end encryption impossible.

But with sockethub, provided you run your sockethub server inside your home, or at least inside your trusted local network, we have an opportunity to implement end-to-end encryption in a way that is entirely transparent to the unhosted web apps that send and receive the emails.

So far, I have added PGP clearsigning to Sockethub (this has now been removed again), using the 'gpg' nodejs package. This is still an early proof-of-concept, and especially generating and storing the keys still needs a lot of work, but I think we should include this as a basic part of our platform.

When not all recipients of an email message have a PGP key, the message needs to be sent in cleartext, and PGP can only be used to sign the message, so that recipients can check from whose computer (or in this case, from whose sockethub server) it was sent.

When the sender knows the PGP keys of all recipients, the message can be encrypted before it is signed and sent, to protect it from third parties reading the content of the messages. They will still of course be able to detect the fact that a message was sent at a certain time, and roughly what length it had, but it's still a vital improvement over the current situation where commercial companies and random nation state governments have, by default, full access to spy at will on pretty much everything we do online.

Inbox search

A feature which many current GMail users will have grown fond of, is inbox search. For now, I am using grep to search through my emails on the filesystem of my server, and although it's not as good as GMail search, this still actually works pretty well on the whole.

In the future we may want to generate search indexes based on keywords or at least based on sender/recipient pairs, to make search easier.

And as more and more people will (one day!) start joining the No Cookie Crew, using their own home grown unhosted email apps, and submitting improvements back and forth to these apps, we will probably also see support for sorting by topic using tags and/or mail folders, as well as hopefully nice integrations with unhosted addressbook apps.

I guess the main message here is, it works, but there is still quite a long way to go before most of you who read this will be ready to start using your own unhosted email app.

But email and PGP are an essential part of decentralized online freedom, and it is important that a few of us live through these growing pains so that, as Eben said in the FOSDEM interview, "by the time you know you need a FreedomBox, we will have it ready for you." The No Cookie Crew is looking for brave pioneer members, to make this true also for unhosted web apps. :)

comments welcome!

Next: Linking things together on the world wide web