With the release of Germ DM for Android, our new cross-platform app relies on a complementary set of server capabilities. We've designed this architecture to support the mobile client without compromising on privacy. Most of these capabilities can be hosted on independent infrastructure, and we're interested in exploring that.
These capabilities can be separated into functional domains of:
Key Monitoring
Messaging AppView
Mailbox operator
Key Monitor
The basis for public key trust between two users Alice and Bob who haven't previously exchanged secrets, is that they can obtain authentic public keys for the other user.
Publishing on atproto allows us to publish Germ messaging public keys, but there's an important caveat. The PDS holds the atproto signing keys, and — like a certificate authority — can publish any keys it chooses.
This is a well-understood problem for websites and secure messengers; the solution is some variety of a transparency monitor:
I don't trust keys I got directly from the issuer; I wait for them to have reached a third party - this monitor.
Germ has never operated a key directory — we don't have the authority to publish keys and attest they belong to a particular user. And the problem here is more like certificate transparency than messaging key transparency. There's no single Germ key directory to monitor. Instead, every PDS can issue keys for their user(s), forming a decentralized key directory. But atproto's self-certifying data gives us a neat solution.
Germ is now a monitor for Germ keys published on atproto in our com.germnetwork.declaration lexicon. We operate a key monitor that subscribes to Jetstream to get an up-to-date view of the decentralized Germ key directory. Germ cannot issue keys. Rather, it is only an authority for the latest key it has seen for a particular atproto user.
This is a generalizable approach for any key you want to publish on atproto.
Germ Messaging AppView
Our iOS app was spending a lot of effort keeping the user's social graph up to date. Seems like a job for an AppView!
The client mainly needs two kinds of information:
Who can start conversations with whom, synthesized from inbox settings (in the Germ declaration), and follow and block records in the bsky.social lexicon.
Germ's Messaging AppView assembles this information for quick retrieval, and provides a diff service for devices to efficiently catch-up a local cache.
This is public data, so consumption of this doesn't leak information about a user's private activity. Our AppView includes the key monitor as a component, but they can be separately deployed.
Mailbox Operator
Finally, to bootstrap your messaging sessions, you need two kinds of mailboxes
DID sender
People you've never been in contact with, can send you mail, authenticating as their DID. Your mailbox can apply inbox settings to these senders
Pre-authorized sender
The core of our Comm Protocol, allowing you to reduce metadata on the wire by pre-exchanging rotating mailboxes with people as you message them
To authenticate DID senders, the mailbox operator has to resolve the DID to a device-bound key, and check if the sender is allowed to start conversations. As a mailbox operator, you can operate a Germ AppView, or you can yourself consume the one that Germ operates. This means you can operate a mailbox with fairly low complexity, without consuming a stream of records updates.
What's Next
We're working in public on these functions:
https://github.com/germ-network/atproto-pmr
Key Monitor
Are you publishing keys on atproto? Reach out if you'd like to collaborate on key monitoring for atproto.
Independently Hosted Mailboxes and AppViews
The mailbox operator sees some limited but necessary metadata about users' ongoing conversations and message requests. It's also a gatekeeper for incoming conversations. It may be desirable for an app view's moderation decisions to extend to message requests.
Users should be able to customize this behavior by choosing a provider they're aligned with to provide these functions.
There are a few steps before we'll be ready for that. We can probably make the public API's formally defined in lexicon. We'd also benefit from a discovery mechanism so that when a user logs in the app can detect if their PDS offers mailbox hosting. Reach out if you're interested!