Moving Past Indicators Toward Behavior

Well, so AI Telemetry, I think, is a good place to... start because like, I think that one of the, one of the things that I've noted and I've talked about it a couple of times on this podcast recently is that I see this sort of like blurring and meshing together of everything, honestly, because if you go back 10 years, I would say, okay, well, you know, you, you, you've got vulnerability management tools, you've got endpoint protection tools, you've got firewalls, you, you have all these different tools, like all the various segments that, you know,

Gartner has deemed worthy of magic quadrants or whatever, you know, so it's like,

and they each had their own thing.

And then organizations would,

you

know, buy some set of those things and figure out how to stitch, stitch them all together.

As we've gone into generative AI into agentic AI and everything is AI.

What I'm seeing is that a lot of these vendors are like, okay, well, as part of us trying to introduce AI into the mix of what we're doing, whether we're an endpoint protection company or a vulnerability management company or whatever it is we're calling ourselves, we're actually looking at all the, we want as much telemetry data as possible.

And so we can make decisions across.

And so ultimately I feel like where we're headed is that there is no there won't be a distinction between these different tools as much.

It won't be like, well, that's the vulnerability management tool and that's the endpoint protection tool.

It's just, I have a tool and it does security and it uses all of the telemetry.

Yeah, that's actually kind of where Kerbal's headed, right? So as a company, we are founded off of telemetry.

And when you think about all the products and services that we offer, You start with what we're known for the most, which is stream and edge, right? And so that collects all of the telemetry.

And when we think about AI telemetry specifically, right, we approach AI and observability as a telemetry problem.

So LLM apps, GPU infrastructure, shadow AI, all of those things throw off scattered and inconsistent data.

And the same with all those tools that you just mentioned, right? So Kribble's job is to unify it into one layer.

And so that breaks down into a few different pieces and, you know, we'll go into, uh, all the things that, uh, AI telemetry, uh, gives us.

And then, you know, the difference between throwing it into a data lake and, you know, piping it to a simps, there are significant differences in the data that you want to pipe there and the visibility that you have.

But, um, Kribble specifically like stream and edge, we're collecting that telemetry, right? And then we normalize it, we tag it, and it's consistent across all those sources.

And then we have a Kribble guard, which is in-flight redaction, essentially.

So when you think about PII and PHI and credentials and, you know, just exposed things that you don't want in all these other systems that you're piping it to, it basically strips that or redacts it before it gets to the end source, right? And then we have Cribble Lake that stores the full history of all of your AI data and telemetry very cheaply, mind you.

And then we have Cribble Search.

So if I don't want to route my telemetry to my SIM, for example, and those log sources are more regulatory and compliance, I can pipe those to a data lake to Cribble Lake.

And then I can use Cribble Search, which basically federates queries across lake And I can, as an analyst, bring that back up.