I discovered Urbit a decade ago when came across its now forgotten bizarre blogpost where it was first presented. It was a weird yet intriguing read. It was years later that searched for what it had become and found that it was a functioning project. More years later, it still is weird and intriguing.
It is the technical part than I find most interesting. Therefore this is the one that will mainly be covered in this post.
Urbit (also see white paper) envisions a (Martian) planet-wide computing infrastructure, designed from first principles for robustness and interoperability. Itself forms the entire application stack above the network layer. This stack can run on anything; atop another operating system, hosted as an overlay, or, in theory even, alone.
The core components in this infrastructure are frozen (see refinement post). That means it is impossible to upgrade them any further. This was done by utilizing Kelvin versioning, a recursive scheme where each component sits atop components closer to absolute zero. It draws elements from Graham’s hundred year language and parallels with the towards limit scheme used by TeX (tends to $π$) and MetaFont (tends to $e$).
The network it builds is URI-like namespace globally distributed via content-centric networking, forming a decentralized digital identity system and a single broadcast network.
In Urbit all data and code are distributed via the global namespace. For interoperability a standardized basis is required. The basis is Nock, a language to define higher-level languages. It is comparable to lambda calculus but meant to be foundational system software. From the white paper
Nock is a little like Lisp, if Lisp were only a compiler target and not a high-level language.
Basically Nock acts like a functional assembly language. Above it is Hoon, a self-hosting language that compiles in Nock and used to implement the rest of the stack. Hoon is a typed functional language but it wasn’t meant to be abstractional like Lisp and Haskell.
Expressions in Hoon are formed with runes, digraphs (two character sequence) of ASCII special characters, rather keywords. This makes Hoon code look similar to J, an APL descendant.
Overall, Urbit describes a simple and clean architecture that can potentially replace many complex system software. Though reading the white paper and through the documentation gives off academic vibes, it actually is not. It is a software engineering project and started off as Yavin’s personal project in early 2000s, years before it became public.
To give an idea, here’s how Urbit has been explained over time in various places1.
[A] static functional namespace (source)
An operating function (source)
[A] personal server stack built from scratch (source)
[A] new programming and execution environment designed from scratch (source)
A solid-state interpreter (source)
[A] new clean-slate system software stack (source)
[A] virtual city of general-purpose personal servers (source)
A clean-slate decentralized software stack (source)
One way to think about Urbit is as a 100-year computer (source)
Your last computer (source)
[A] simpler computer, a quieter computer, a more private computer (source)
[A] clean-slate OS and network for the 21st century (source)
Combining, per the documentation
Urbit is a clean-slate software stack designed to implement an encrypted P2P network of general-purpose personal servers.
After development started, it picked some custom terminology which made reading the documentation difficult to parse2 and understand what each term means and every component does. Adding on that, the site up to the year before wasn’t much help. Fortunately it was redesigned last year making things more friendly, though it’s still complicated to fully understand.
From before, the stack is comprised of the operating system, the virtual machine when it’s run atop another, the identity layer, and two languages (a low- and a high-level one).
Nowadays, the Urbit operating system and kernel is called Arvo. Its state is a pure function, meaning is determined by, the event-log. For this reason, it is referred to as purely functional. The event-log is append-only. Arvo is primarily made to run hosted therefore it doesn’t do what an real operating system is expected to do (memory management, I/O, drivers).
It’s composed of modules called vanes. Those are
|Armes||peer-to-peer networking protocol|
|Clay||filesystem and revision-control system|
|Gail||userspace supervisor and sandbox|
|Jael||identity-related information tracker|
Clay is a versioning filesystem, but also handles file-change events and maps them between Arvo and the underlying system. An interesting property is it being type aware.
On Unix systems, Arvo runs atop the virtual machine Vere, basically a Nock runtime environment, which is implemented in C. It forms the intermediate layer between Arvo and the underlying operating system. In theory, a baremetal virtual machine could be made essentially making Arvo run native.
Finally, the identity layer is Azimunth. It is built as a suite of smart contrasts on the Ethereum blockchain. Urbit’s constitution wasn’t originally on Ethereum, but the move happened for security and other reasons.
Those other reasons basically boil down to convenience. Ethereum is not hard coupled to Urbit and another method of consensus could be used instead. Azimunth isn’t built strictly for Avo and can be used as a generalized system for other projects.
There’re limited Azimunth identities which gives them value (scarce resource). Their number is big enough though for that not to pose a problem. Scarcity is used alongside reputation to fight malicious actors. There’s an hierarchy named after astronomy terms3 (galaxy, star, planet), called points. In addition to them, there’re Urbit identities not registered on Azimunth, moons and comets. Moons are meant for devices owned by a planet owner, and comment’s are disposable identities. Compared to existing (common) layers (IPs, domain names, user identities), it combines them all into one.
I’ve few objections to the identity layer. Ethereum just doesn’t feel right when you’ve designed everything else independently. Also the bad reputation for cryptocurrencies doesn’t help the project either. Talking about reputation, it means that philosophical disagreements may lead to fracture of the network, something often seen in (society itself and) reputation-based online communities. From same communities it has also been seen that people refrain from talking and behaving freely when reputation is at stake. Finally, as mentioned, scarcity means that identities have value meaning joining Azimunth by design won’t be free.
Urbit according to, its creator, Yarvin, is a combination of well-known ideas that have never been together before. In this section I’ll compare a few projects that I’ve found to bear similarities.
There’re some parallels between Urbit and Ethereum. One is both want to create a decentralized computing platform. The difference is that Urbit performs work on one’s own computer, whereas Ethereum is used to perform work on someone else’s computer.
Also, there’s a similarity in blockchain and Arvo event-log being append-only. Interestingly Urbit address space is a special case of a consensus ledger, but it lacks a consensus engine which is why Ethereum ended up being coupled to Urbit.
Emacs is a great operating system, lacking only a decent editor.
Emacs is basically Lisp code running atop a Lisp interpreter implemented in C. The stack design is similar. Arvo is (compiled to) Nock code running atop a Nock interpreter implemented in C. Also, Hoon and Nock draw elements from Lisp. Albeit those languages meant to be comparable to C and assembly respectively. Similarly applications can be built atop them.
Emacs isn’t operating system but a single-application platform. That is due to Emacs lacking parallel threads, and only got sequential threads few years ago.
Emacs exists only in applications layer. That is its networking is like in any other common application.
They don’t share goals, no peer-to-peer networking and no identity system. Emacs is an extensible text editor therefore those aren’t required.
That said someone creating a peer-to-peer application on Emacs isn’t a far strech, and I won’t be surprised if it exists. Afterall Emacs has few clients for various protocols written in pure-ELisp.
Inferno is the successor to Plan 9 that similar to its ancestor never became popular and remained an obscure project known to few. It is a distributed operating system that can run hosted or baremetal, utilizing a virtual machine, and written entirely in its own language.
Received from its ancestor are 9P, Venti, and Fossil. 9P is the network protocol (and virtual filesystem), Fossil is the filesystem offering versioning feautures, and Venti is a storage system that enforces write-once policy. Therefore they’re similar to Ames, Clay and Arvo’s event-log.
In contrast to Urbit, it actually runs baremetal (rather being theoretical plausible) and has ports to other operating systems (Windows and even embedded systems) rather being coupled to Unix. No identity layer exists but instead the common stack is used. 9P isn’t peer-to-peer protocol but only allows transparently connecting machines. Also Venti is only used for data rather permanently store the imperative steps that lead to its current state. Like before, they don’t share goals.
Nevertheless a peer-to-peer domain and identity system (such as Armes and Azimunth) may potentially be implemented atop 9P and therefore Inferno. This will make Inferno, an Urbit competitor without the clean-slate aspect.
GNUnet, according to its official site
[It] is an alternative network stack for building secure, decentralized and privacy-preserving distributed applications. Our goal is to replace the old insecure Internet protocol stack.
The applications in the suite require an underlying operating system, meaning it is meant to function on application layer only. But it also defines its own name system based on peer-to-peer networking and its own decentralized identity provider. Philosophically they share the same goal. That of decentralizing the internet utilizing a global namespace for identity.
It is interesting but, even though started in 2001, unpopular project usually overlooked in disqussions about the decentralized ecosystem. A thought worth exploring is whether Urbit can run on the GNUnet name system, and whether it could be used to provide ownership (identity).
Nix is a functional software deployment tool. The important parts in this case, is that it implements a functional programming language, which draws elements from Haskell, and a build system. The similarities are
A domain-specific language is made because it’s thought to be the ideal design in this case. Worth mentioning is Guix, a Nix-like system, uses Guile, a Scheme (Lisp) dialect, instead.
The build result is a pure function of the input recipe, like Arvo is a pure function of its event-log.
In genera Nix fits nicely with Urbit. The devs probably thought so themselves because Urbit is nowadays using Nix as its build system for the Unix side of things. I’m expecting a Hoon-based Nix-lite system will eventually find its way on Arvo. Or maybe, like Guix, Urbit remade using an existent general-purpose language.
Before proceeding, as I’ve wrote in another post, in
$HOME, I keep executables in
~/bin and variable data in
meaning Urbit instances (ships) are kept in
~/var/urbit. Be free to
use whatever directories you like.
The official site has a good usage documentation, covering everything but spread in various page. The following will be a summary as I see fit.
In order to connect to the network, two pieces are required. The system (stack) and an identity. Let’s start by getting the system. Easily retrievable pre-built binaries are offered. Meaning that running Urbit is as simple as doing
wget https://urbit.org/install/linux64/latest -O /tmp/urbit.tgz tar zxvf /tmp/urbit.tgz --directory ~/bin --strip=1 rm /tmp/urbit.tgz ./urbit
Now, let’s get an identity. As mentioned in previous section, there’re multiple identities. To get started you can create a comet, an anonymous urbit (also called ship).
$ urbit -c comet ... ; ~zod is your neighbor ; ~wanzod is your neighbor
This takes a while, and the last line will probably differ. When is finished loading (or booting), its console (called dojo) will be shown. The prompt is the following.
Bug alert: There’s a bug where lingers in Eyre state. That is seen
http appearing right of
>. Wait until this eventually finishes
> will be used and below will be the result. The ship
is a persistent process. You can abandon the ship (pun intended,
abandon is not an actual term) with ^D or entering
|exit into dojo.
In the previous command
comet is the name of the directory that hosts
the ship (instance) state. This directory is called pier. Though
designed to be portable, moving it must be done when the ship is
down. Also, as networking is stateful, two copies of the same ship
cannot run at the same time.
To start the comet again, run
$ urbit comet
After starting it, rather using the terminal, a web app named
Landscape can be used instead. Check the line starting
http: for the
port used. Then on the web browser access
localhost:<port>. At the
+code and copy-paste the code that appears in the “Access
You can check connectivity with rest of the network running
> |hi ~zod >= hi ~zod successful
Automatic updates can be enabled running
> |ota (sein:title our now our) %kids >= kiln: activated OTA from %kids on ~binzod kiln: downloading OTA update from %kids on ~binzod
The status of automatic updates can checked with just
> |ota >= OTAs enabled from %kids on ~binzod use |ota %disable or |ota ~sponsor %kids to reset it
Filesystem synchronization may be initialized running
> |mount % >=
This creates a
home directory inside the pier populated with few
standard child directories. After making any modifications, changes may
be synchronized to Clay running
> |commit %home >=
This workflow is exactly like git where you checkout the repo, make
modifications and then commit those. The directory containing the
“repo” data is
.urb inside the pier.
%/web was automatically mounted and served on localhost. This
made it simple to statically publish a site or run single-page
apps. Formatting was done with Udon, a flavored Markdown dialect. This
functionality was eventually replaced with aforementioned Landscape.
Actually the dojo isn’t just a console (shell) but a REPL allowing you to evaluate arbitrary code written in Hoon.
> (add 2 2) 4
Moreover it isn’t the only prompt available in Urbit. Like an operating
system (such as running
fdisk in a Unix shell), apps may provide their
An essential such app is
:chat-cli or just chat. To enter chat press
^X. The prompt is the following.
Unfortunately it appears comets are not allowed to join channels anylonger. Better go back to dojo for now. This level isn’t yet available for your character.
To go back to dojo simply with ^X. You can see all commands and available tools, alongside an one-line summary running
Bug alert: There’s an open bug right now where if you enter
~sampel-palnet/chat-name in chat you will forced out and be prevented
re-entering it. To fix run the following (taken from the GitHub issue).
> |unlink %chat-cli > |link %chat-cli
For anyone interested to learn more, Neal Davis taught a course last semester on Urbit at University of Illinois, cleverly named Martian Computing. The materials have a public mirror on GitHub. They are what lectures notes you’ll expect to be. That is perhaps the easiest way to learn in more depth everything discussed here.
After that, or before that if you feel adventurous, check the official documentation. To learn Hoon programming check Hooniversity. The source code of every Urbit component is available on its monorepo. You’re warned that codebase is opaque at first glance.
Also take a look at post by Erik Newton and post by Clark in Popehat for philosophical views and a commentary on how Urbit can be used to create a decentralized network. This is an also important part which I didn’t discussed in this post. For a critique of this network see post by Francis Tseng. Similarly to me, it objects the identity system.
Closing up, Urbit is one of the cool, cooler, coolest projects happening. If you pass over the arcane jargon used and the cryptocurrency relation (Ethereum; soft coupled), it is an interesting approach to how computing infrastructure could be if it was designed, rather evolved4.
The artwork used in the blog posts is amazing. ↩
Reading the Urbit’s documentation was (and is) like reading an occult grimoire, and reading Hoon code like reading a incantation. Maybe I should’ve called Urbit a mystical computer. Maybe both are correct and Martians are wizards. ↩
Initially rather astronomical terms, feudalism classes were used. This has generated, especially when considering Yarvin’s personal political writings, some bad press. One should note that universe structure hierarchy captures perfectly decentralized networks. ↩