Transcript
A (0:00)
Foreign. Welcome to reshaping workflows with Dell Pro Precision and Nvidia, where innovation meets real world impact in high performance computing.
B (0:19)
That's Logan GTC lost the day 2026. Not even sure anymore. I'm here with Iman of Protopia AI. So Iman, give us a little background who you are, what you do at Protopia AI and just a brief overview of the company.
C (0:35)
Thank you Logan. My name is Iman. I'm the CEO and founder at Protopia AI. We are an inference privacy layer for AI agents, AI factories and we enable high utilization multi tenant compute on inference backends.
B (0:52)
Okay, so you've said a lot of words, let's break those words down. So let's talk a little bit. I see a dancing claw kind of in the background. Everyone loves crustaceans. Tell us a little bit about. Let's start with the privacy piece. What does Protopia AI do from a privacy and security standpoint for AI deployments?
C (1:09)
Yeah, so the little visual you're seeing in the back, we call it SafeGlaw. We actually did a live stream on Nvidia's the developer livestream on Wednesday last week about this where we introduced the notion that in Open Claw Agentic systems a lot of people talk about the only way for keeping data private is to try to host the actual language models that the agent's going to try to talk to locally. But what we're showing using our privacy preserving technology that we call Stained Glass Transform, which is what jacket the claw is wearing there, is that you can have your OpenClaw agent talk to a hosted model without needing to actually send the sensitive information in plain text to that hosted model. So let's say your Agentix system wants to go through your emails and find things that it's going to notify you of. That classification of what email you need to know about doesn't need to expose your emails to the hosted model. And so that's what our product enables. Stained Glass Transform is something that runs on your local agent in order to connect to the hosted model. So that's an example of privacy for agentic AI using hosted models.
B (2:17)
Okay, so we've got kind of a follow up, right? So say, you know, I don't know if I would let OpenClaw have access to my corporate emails. That's another, another thing. But let's say I did. Like we're kind of talking about, you know that point, like is there is an encryption, like what do you actually like the layer in between? Like what are you exposing to Ultimately give the agent and that, that that host. Well, I'm not hosted model but like let's say you're using like ChatGPT, what is it? What's the newest one. 5.2 or whatever you've opened that you have it connected to an API. Like what are you doing to protect that data? Is it limiting the data that's sent? I mean you don't have to share any like proprietary information. I'm just more curious.
