#OPG
The part of OpenGradient I keep coming back to isn't the privacy pitch.
Its the signed reply after.
Nice little answer.
Most chat apps still run same old deal. You type something sensitive. They hand you a policy. Everybody pretends that's the hard part solved.
Cute.
OpenGradient Chat at chat.opengradient.ai feels wrong in a better way.
Client-side encryption first. Identity gets shoved into OHTTP relay path. Fine. Content only unwraps inside the OpenGradient attested enclave. Then reply comes back as a TEE-signed response.
Normal looking answer.
Not a normal path. exactly.
Thats the part I keep reopening. I'm used to "private AI" meaning a nicer policy page and a calmer font. OpenGradient Chat makes reply feel different in a worse way. Worse good, I mean. Because now I'm not asking whether company behaves. I'm staring at relay split, client-side encryption, attested enclave flow, and @OpenGradient private inference path trying to work out what part of this thing was even allowed to know I existed.
Alright.
Say somebody drops a draft deal memo into OpenGradient Chat. kind of note nobody sane wants sitting in a normal AI log trail. On most assistants... question is whether the platform deserves trust. On OpenGradient, the uglier question is better.
What part of the path could even read it?
Thats a nastier question.
... where normal chat starts looking cheap.
That's the split.
The answer still looks ordinary.
route behind it doesn't.
And once OpenGradient Chat turns reply into something tied to client-side encryption, OHTTP relay separation, and an attested enclave, privacy stops sounding like policy and starts sounding like route design. Hardware. Cryptography. Receipt-like proof. Boring words. Good. Boring is where this usually stops lying.
Thats the part I keep getting stuck on.
Policy promise. Cute.
OpenGradient made the route answer back.
That's a weird thing to get used to.
@OpenGradient still hands back a normal-looking reply anyway.
Good luck unseeing what had to happen before it got there.