The IPFS family of technologies contain a Petri dish of great ideas. However, many of them have a high degree of optionality that can render interoperability challenging, and several were created in a way that does not always mesh well with other parts of the wider web stack.
LUCID is an experimental project to subset and apply IPFS technologies in ways that are readily interoperable, relatively easy to understand to newcomers, and that work well with the web as generally practiced by developers. It gets its name because it first applies that logic to IPFS's most foundational component, CIDs.
CIDs (Content IDentifiers) are an excellent and highly-extensible format for content-addressing. One aspect that is hampering their wider adoption is the wealth of options that they support and the accompanying complexity. LUCID is an experiment to see how small a subset of that option space we can get away with supporting while working on key non-IPFS goals. LUCIDs have two requirements:
- They are fully compatible with CIDs: any off-the-shelf piece of software that reads CIDs will read LUCIDs with no modifications. This also means that LUCIDs retain CIDs' extensibility, they just don't make use of it just yet.
- They support as few CID options as possible (though this set may grow as we explore the space more):
- Only v1, no v0.
- Only base32 multibase encoding (the
b
prefix) for the string, human-readable encoding. - Only the raw-binary codec (
0x55
) and (maybe) dag-cbor (0x71
). - Only Blake3 hashes (
0x1e
). - No blocks.
- This set of options has the added advantage that no varint processing is required.
This is a highly opinionated and arguably wrong set of options. No blocks means that big files are big and will need to be split at a separate layer if that is desirable (e.g. seekable video will need to use MPEG-DASH or something similar).
One valuable property of LUCIDs is that they can be string-compared. They only need to parsed when there is a need to access the codec or the hash.
Parsing and producing LUCIDs can be done using cid.js
.
Web Tiles are a simple format that can be used to build content-addressable web sites. At the root of a tile is a simple manifest that contains metadata about the tile, and a mapping from paths to CIDs (here to LUCIDs). Evidently, the path mapping can be autogenerated.
Advantages of tiles include:
- High security. It is impossible to load resources from outside the tile. This means that you can not only verify that the root of the tile you are getting is the one you expect to get, but also that all of its dependencies are as well. This also makes it private. This is a big improvement over web content.
- Web pathing. Paths in IPFS can sometimes be confusing, and require supporting UnixFS, which also comes with its own dependencies. In HTTP, the path is not structured as a series of directories, it is just a string. The tiles approach mimics that so that it has direct, simple, 1:1 mapping from paths to resources (and therefore to CIDs).
- Media types. A lot of IPFS content doesn't have a MIME type attached to it, and that is often inferred at runtime, for instance by a gateways. This is problematic and introduces known security issues. With media types built in, tiles avoid this problem.
As a result, when you take a web app or page contained in a directory and you wrap it up as a tile, it can just work and be safe, privacy friendly.
As currently implemented, this system does not allow calling any external APIs. There are ways of building that into a tile implementation, either by exposing a wallet API or by supporting intents/wishes.
The relevant code can be found in:
manifest.js
: manages the tile manifesttile.js
: manages tileswatcher.js
: watches a directory for changes, to keep producing a correct manifest for it
As a demo, the lucid-dev-server.js
script can be used:
Usage: lucid-dev-server [options] <path>
Run a dev server for Web Tiles
Arguments:
path path to the directory to serve
Options:
-V, --version output the version number
-p, --port port
-h, --help display help for command
For instance, running ./lucid-dev-server.js demos/beautiful-cat/
will spin up a dev server at
http:https://localhost:3210. Pointing your browser there will show a tile on the right as well as its
metadata on the left. If you then tinker with the code in demos/beautiful-cat/
, the browser will
keep updating the LUCID that it is pointing to as the root of the tile and updating the tile.
(Note: Brave blocks some of this, for non-obvious reasons which I have yet to debug.)
Caddy is a very cool server, that also happens to support a very flexible dynamic configuration API . What this command does is that, when pointed at a directory, it automatically configures a running Caddy instance to serve the content of that directory via gateway endpoints.
The command is pretty simple:
Usage: caddify [options] <path>
Dynamically configure a Caddy server to serve IPFS
Arguments:
path path to the directory to serve
Options:
-V, --version output the version number
-e, --endpoint <url> the Caddy endpoint for its configuration API (default: "http:https://localhost:2019/")
-d, --domain <domain> the domain under which to serve IPFS (default: "localhost")
-p, --port <number> the port from which to serve IPFS (default: 443)
-o, --out <path> path to save the config to instead of posting it
-q, --quiet shush (default: false)
-h, --help display help for command
Running it to configure and serve from the local Caddy:
[lucid 15:12 (main)] $ ./caddify.js demos/beautiful-cat/
Configuration updated:
- https://bafkr4igra3bh6uuhmtpwiymbipaodx3lxwupn3absumn6do234xupooice.ipfs.localhost/ ➯ /index.html
- https://bafkr4ickhymp67xqfakp4qysyep5hgwmul5lucymb7ttxfbaqkbuap355i.ipfs.localhost/ ➯ /cat.svg
- https://bafkr4idcy33utsake6atvbagnojkn7odp7mdo6n7tvspd4ndnewphj67xu.ipfs.localhost/ ➯ /wtf.jpg
The high-level overview of the Nostr protocol has this opinionated statement about its architecture:
It doesn't rely on any trusted central server, hence it is resilient; it is based on cryptographic keys and signatures, so it is tamperproof; it does not rely on P2P techniques, and therefore it works.
I experimented with a Nostr integration because that aligns well with IPFS. IPFS is often thought of as operating atop a DHT, but that is only one way of doing IPFS. As per the IPFS Principles, IPFS is transport agnostic, and HTTP gateways are an excellent way of using IPFS.
In this repo, augury.js
is a minimalistic but usable Nostr relay server. Its usage is simple:
Usage: augury [options]
An experimental Nostr server that integrates with IPFS
Options:
-V, --version output the version number
-p, --port <number> the port on which the WSS server runs (default: 6455)
-c, --config <path> a configuration file
-s, --store path to the directory in which the data is saved
-l, --log-level logging level (verbose|log|warn|error)
-h, --help display help for command
The configuration file is a JSON file that has the same options as the CLI above (with logLevel
for --log-level
)
as well as an extra posters
field that is an array of Nostr pubkeys (formatted as they appear on Nostr events)
who are allowed to post to that server.
If you run augury.js -c some/config.json
, then you'll have a Nostr relay server running. Because of some
annoyance with how web sockets work, it's better to run it not directly on localhost
but using a local
wildcard domain name. I use dnsmasq
to resolve augury.bast
locally as well as a Caddy server configured thus:
https://*.tile.augury.bast, https://*.ipfs.augury.bast, https://*.augury.bast, https://augury.bast {
tls internal {
on_demand
}
reverse_proxy http:https://localhost:6455
}
(You might get away with internal
instead of on_demand
.)
This Augury Nostr relay server has extras compared to regular Nostr relays:
- It has an extension to NIP-94 that supports a
cid
tag on resource metadata to capture the CID for a resource. - It has an extension to NIP-96 that conveys the
fact that a NIP-96 endpoint is IPFS compatible (it adds an
ipfs: true
field).
These features were designed and implemented so that they add IPFS support to Nostr while creating the smallest amount of difference possible. And, importantly, no change to the Nostr client is needed. Essentially:
- NIP-96 works as usual.
- The preference is to expose IPFS CIDs at subdomain gateways (
https://<CID>.ipfs.<DOMAIN>/
), which NIP-96 doesn't immediately support (it expects the resources to live as sub-paths of the defined API URL). In order to bridge the two, this exposes the usual operations on the sub-path of the API URL, but redirects (308) theGET
operation to the corresponding IPFS subdomain gateway.
Inside coracle
is a lightly modified Coracle client that can
be used to interface with this Nostr relay. (The only modification is that Coracle bails if you lack a Dufflepud
preview server, this falls back to just using the original file.) Run it with yarn dev
in the subdirectory.
(Another Nostr client should just work as well.)
Configure your local relay in the client:
Create a post with attachments:
The images are available at a subdomain gateway URL: