About two years ago, before Agni was called Agni, I was at a security conference in Berlin. Someone gave a talk about a major email provider that got fully compromised — not the users, the *infrastructure*. The attackers didn't need passwords. They just... read everything. Because the keys were right there on the same machines as the data.
I remember walking out of that talk and calling the other two people who eventually became co-founders. "What if we just... didn't have the keys?" It sounded almost stupidly simple. Like, could you even build an email service where the server is literally incapable of reading the mail?
Turns out, yes. But it makes everything harder. You can't do server-side search. You can't recover forgotten passwords — we can only nuke the old key and start over. You can't train a spam filter on message content because you don't see the content. We had to build an entire SMTP edge that encrypts incoming mail *before* it touches disk, using the recipient's public key that we don't have the private half of.
Is it overkill for most people? Probably. But the thing is, once you've architected it this way, you can't accidentally screw it up later. We literally *cannot* comply with a request for plaintext. Not because we don't want to — because mathematically, we don't have it. That's the only kind of privacy guarantee I personally trust.