There's a thought that's been rattling around my head for a while now, and this week it finally got loud enough that I gave it a name and a repository.
To set the scene: I have been on the AT Protocol for years. My handle lives on it. My website is built on it. This very blog post is, at the exact moment I'm writing this sentence, a site.standard.document record sitting in my own repository, addressed as at://did:plc:ofrbh253gwicbkc5nktqepol/site.standard.document/<rkey>. That address is real. It is as precise as an https:// URL. It is an IANA-registered URI scheme. And if you copy it into any browser on this earth, the browser will shrug and offer you search results instead.
That's the itch.
The protocol's beautiful core idea is that everything has an address: every post, profile, blob and record, nameable with an at:// URI, fetchable from whatever PDS holds it, verifiable against its CID. The infrastructure is right there. The one piece of software that exists specifically to follow addresses, the thing whose entire job is resolving URIs, the address bar, has never been told any of this exists. So today every path to an at:// URI runs through a gateway or an app's web client, which politely resolves it for you. It works! It also quietly reinstates a trusted middleman into a protocol designed, carefully and on purpose, to not need one.
Just over a year ago I wrote that I don't really care if Bluesky dies, as long as the AT Protocol survives. raindrop is that argument taken to its logical conclusion: the protocol deserves to be infrastructure, and infrastructure means the browser speaks it natively.
What "first-class" means
raindrop is a Firefox fork with first-class support for the AT Protocol and its URI schema.
Concretely: type an at:// URI into raindrop's address bar and it resolves the way a browser should. The browser resolves the DID. It finds the PDS from the DID document. It fetches the record. It renders it, and later versions will check the CID so what you see is the exact data version, with provenance (which DID, which PDS, whether verification passed) shown in trusted browser UI rather than a web page doing an impression of trusted UI. No gateway in the middle, and no sneaking off to one if resolution fails: if the PDS is down, raindrop says so, because honest failure states are a design goal, not an afterthought.
I have a history of shoving this protocol into strange places (two separate Nintendo consoles, a raw C library, a PDS in C++), so please trust that I considered every option short of this one first. This isn't another client. It's the substrate underneath all the clients.
Why call it raindrop?
Because the ecosystem already named itself. The collection of apps, servers and tooling built on the protocol is known as the Atmosphere, and an atmosphere's whole business, the thing it exists to do, is rain. If Bluesky is the sky and the rest of us live in the Atmosphere, then the protocol deserves a browser that treats its records as weather: small, precisely-aimed drops falling straight from the author's cloud (their repository) to wherever the URI points them.
Which is, conveniently, also the architecture. A raindrop's defining property is that nothing intermediates it: it leaves the cloud and it lands. It doesn't stop off at a gateway to be signed for. In a world where everything routes through someone else's cloud, at:// is a drop that lands exactly where it is addressed, and I wanted a browser that understands that about it.
Also I live in Britain, where rain is the one national resource we keep in surplus. Sustainability matters.
"Why not just an extension?"
The first question everyone reasonably asks. Extensions for at:// already exist, and I have genuine respect for them, but they demonstrate the problem more than they solve it: with no scheme support underneath, the best an extension can do is catch the URI somewhere other than the address bar and redirect it to a web viewer. A gateway with extra steps.
The annoying part is that the pieces raindrop needs to change are precisely the pieces extensions cannot touch:
there is no WebExtension API for registering a URI scheme, so an
at://string typed in the bar is resolved (as a search query) before extension code ever runs;the URI parsing itself needs fixing, because a DID in the authority contains un-encoded colons, which breaks WHATWG URL parsing, and the parser that has to cope lives in the engine;
fetching an arbitrary PDS from page context means every PDS operator on the network must cooperate with CORS, which the protocol should never have to ask;
and provenance UI has to be browser chrome, unforgeable by the content it's describing, or it's just decoration.
The one-line version, which now lives in the repo's architecture doc: an extension could deliver at://-as-gateway. Only the browser can deliver at://-as-infrastructure.
Where it stands, honestly
If you're hoping to download a browser today: not yet, sorry. raindrop is at the design stage, and I'd rather open with that than with a countdown I can't promise. (I am one person, chronically ill, with a Mac mini that already works for a living. Timelines flex when my body does.)
What exists today is the repository, and I've tried to make it worth reading: a roadmap that deliberately keeps the first milestone small (scheme handling, DID resolution, PDS fetches, a plain generic record view, so that every at:// URI on the network is viewable from day one, correctness before polish), a working architecture draft of the whole request flow, and, because forking Firefox deserves adult supervision, the legal homework: raindrop builds on Mozilla's open-source code under the MPL but uses none of Mozilla's trademarks, and the same care is taken with Bluesky's. There's a whole document about it. Trademarks are the least glamorous thing I have ever researched for a project and I'm weirdly proud of it.
It's at github.com/ewanc26/raindrop, the roadmap is public, and design challenges are wanted, genuinely. I would rather be argued with early than agreed with late. If you know gecko, or Firefox build configs, or you've maintained a fork before, there is an issues tab and it has my full attention.
I call my blog a garden. The Atmosphere has grown a whole crop of clients and tools in it over the past few years. raindrop is my attempt at weather: rain on the open web's address bar, from the protocol itself, with nobody's gateway holding the umbrella.
Rain is free. Opinions are welcome.
raindrop is an independent project, not affiliated with or endorsed by Mozilla or Bluesky. Firefox and the Mozilla logos are trademarks of the Mozilla Foundation; AT Protocol, atproto, Bluesky and the Bluesky butterfly logo are trademarks of Bluesky Social PBC.