#opg $OPG @OpenGradient
I used to think privacy and AI were two things that just could not coexist.
Either you use AI and accept that your data is exposed, or you avoid AI altogether and lose out on everything it offers.
That was the deal. Take it or leave it.
Then I started digging into what Open Gradient actually built with Veil, and it completely changed how I think about this.
Here is the part that stood out to me the most:
In a normal setup, one party always knows two things about you at once — who you are, and what you asked.
That is the actual privacy leak. Not the AI model itself, but the fact that identity and intent are sitting together in the same place.
Veil breaks that connection on purpose.
Using Oblivious HTTP, the relay only ever sees your identity wrapped in encrypted text.
The TEE enclave only ever sees the prompt, never the person behind it. For anyone to expose you, two completely separate systems would have to collude.
That is a much harder problem to create than to solve.
Then there is the verification layer.
Every response gets signed inside an attested TEE enclave and checked on your own machine before your agent even sees it.
This proves the AI ran on code that has not been altered or tampered with. You are not taking anyone's word for it. You are checking the proof yourself.
What really sealed it for me is how easy the integration is. No rewriting your agent. No new framework to learn.
Just one environment variable, and your existing OpenAI-compatible setup is suddenly running on private, verifiable inference.
This feels like the kind of infrastructure people will look back on as the moment AI privacy stopped being a tradeoff.
Have you ever thought about who can see both your identity AND your prompts at the same time?
