Graphiant Demos with Vinay Prabhu

Folks, bu Chief Product Officer at Graphite. So the first demo that I'm showing, this one's the only one that's prerecorded, and it's just, so I stumbled through copy pasting IP addresses in the form. So essentially what we are showcasing here is the data exchange capabilities.

You'll see that in, uh, manifest itself in three broad steps. Essentially, how we have eased up logistical and communication, uh, functions is by first publishing a sub, uh, publisher inventory, essentially services that you want to showcase on a marketplace on the graph in fabric. Second, you're essentially getting a subscriber inventory, all your B2B partners who might want to consume this service on demand.

And third is the invitation and an invite to these customers to really consume this application, our service as they see fit from the location of their choice. All of these really think of it as a switchboard. You get to build an inventory match up subscribers to it on demand.

You can have any, to any communication securely with the routing, security and policy constructs all built in by graph, taking the logistical overhead for the operator away from them, and essentially allowing them to really start pause, disabled service on demand. So like I mentioned, AI is a massive distributed publisher subscriber problem at a global scale. All of this data exchange needs to happen in three easy steps.

The first step, like I mentioned, is creating the publisher inventory. Raffin allows you to create a service, whether it's an application or a peering service. In this case, we have an AI connect app.

Think of a Stargate or a GPU service that I want to advertise to my business partners. I essentially let the producer define the location of this application and the prefixes that they want to advertise as part of the application. Once you create the service, it's as simple as creating a subscriber inventory.

They can be either graph or non graph, it doesn't really matter. You can connect to the graph fabric via either methods. In this case, we are taking a graph customer.

You fill in the customer details, essentially the customer name, the email they want to use for getting the invite and the sites they want to consume it from. The third step is you just match this customer or the subscriber to your inventory of publishing app published apps. It's as simple as it's matching up the app name or the service name to the customer.

In this case, we are saying, Hey, you have a Salesforce company that's going to con consume the AI connect service. That could be a Stargate GPO firm. You really want to connect those services up.

You allow them to consume this. You essentially set up the prefix. You want to see essentially handling all the NA capabilities to obfuscate any overlapping IP addresses.

In that form, you see an invite. This is an in internal app, could also be an external app like Salesforce, but it could be a company. It doesn't matter.

There's not specific inventory of apps that you support. It just whatever, Whatever apps that you want to publish into your publish uh, publishing inventory. So think of it like your personal marketplace.

Now, just like you have your iPhone, uh, app application marketplace, you have this for business to business connectivity for any app that you develop, build, and operate. And so as you're doing this creation process, how are you connecting this connection to the governance side? How am I labeling this, uh, so that it can, my policies can apply to, to what I do?

Really good question. So what we do here is we completely isolate this publisher and subscriber communication on a separate slice on the network. In fact, they have separate pairwise keys as well.

So there's different security associations for every subscriber and publisher communication. So think of it, you've got your own private VPN tunnel across the globe for every publisher and subscriber co combination. And since we are doing routing at the source, we can really give you visibility of who's communicating where, how across the globe itself on the infrastructure.

Okay, but if this particular connection is transferring, say, financial data, um, how does the compliance folks know that I'm setting up a new PCI compliant or non-compliant connection? Really good question. It'll be part of the next demo that I show you, the visibility piece.

So it's a great segue for me to jump into the second half of it. Uh, but yes, you'll, you'll see the application identified. We, you've tagged it.

You've named it as a, a, a compliance or regulatory financial app. Here you'll see that reflected in our data assurance dashboard, which then lets you drill down into every session that has ever taken place on for that consumption of that application. So we, we will dive into that in the next section really shortly.

So like you can see here, what we have sent to the end user is a email, really a secure exchange of what they need to consume. They can accept the invite, choose where they want to consume the application from and how they want to consume it in their network so they can really solve their netting issues over there. They can choose the sites, the land segments that are allowed to consume the application or service.

And all the security and routing policy constructs are auto-generated by graph for you. And that's it in three easy steps. What took months to really build out takes minutes on the graph portal.

You don't need a de militarized zone setup. You don't need, uh, emails exchanged and text exchange of pre shared keys flying over the internet. You don't need all of that for a non graph customer.

The experience is exactly the same. The email that goes out to you, you get a, a one time username and password to really go in and download your security VPN profile based on the vendor that you are operating at your end. So you essentially, you, we, we see it in the financial sectors where the credit unions that really don't have the IT capabilities to really configure these boxes, all they want is a configuration that they can pick up securely.

Uh, and the publishers don't really want to sit and keep contacting each of those satellite offices constantly and keeping, uh, the logistical overhead. The key rotations happen automatically. You get notified, you come here, pick up the config, paste it on your Juniper Cisco box and you're good to go.

So I have a question for you. Yes. Um, Kevin Myers, this, um, connectivity to your box, that's EBGP, it's their a SN to your as SN essentially is that It's all routing.

Yeah. Whether it's I four IPV six, that's the handoff Yes. To the customer.

Exactly. Okay. So, uh, and then we are using the BGT BGP attributes to really give, give preference on where you want to land that service, through which gateway do you want to consume it, so on and so forth.

So, gotcha. You'll have, uh, all the load balancing capabilities for really consuming the application from the closest gateway point you want to enter the graph in fabric. I think you heard someone mention that SRV six is involved in your transport.

Y Yes. So we use a form of SRV six to really show you the visibility piece of it, but yes, I'll show that, uh, capability as part of the observability demo. Gotcha.

Okay. I'll wait for that. Thank you.

Okay. With that, jump into the second piece of the demo. So what we have here is Graphene data assurance dashboard.

Essentially what it provides is the visibility of all the applications that are running on the infrastructure, who's producing them, who's consuming them, and how they're being set in motion across the globe. What's enabling this is a real time telemetry streaming pipeline that's built by folks from Netflix, Disney streaming, uh, and Oracle, who can really mu that data. We wanted to bring that streaming experience into the networking world as well.

So what we can do is really consume millions of records with subsecond latencies from all the network elements in our infrastructure. 6 million records that have flowed into the system are munched in real time compressed down to 3,700 applications that are actually running on the platform itself on this network. And these 3,700 apps really compressed down into 15 malicious applications that you need to quarantine or really get out of your network with a click of a button.

You can see those application endpoints, these servers that we deem as malicious. You get to see their threat score, their risk status. You get to see the resource category of why we think these are malicious, and the ability to really see the blast radius of who was consuming it, which sites, which, uh, devices, which regions were involved.

And the ability to block this with a click of a button, let's say a network's running clean. Can, can you go back to that? Yeah.

So how, how do I look at the details under that particular one so that I can trust that I want to hit Go ahead and do this. Absolutely. So we show you even why we de as malicious.

We tell you all the threat intel associated with it. So you'll get to see which nation state actors were involved. Why we, what are the other threat intelligence platforms telling you about this endpoint?

What are the IOCs and why you're rating, Rating is the, So that I can actually trust the ratings that Exactly. So you are getting to drill that down right here. So what we have shown is the threat intel that we are sourcing for this endpoint, and you can really see all the factors that have gone into deeming this endpoint as malicious as well.

So it's completely transparent to the end user as to why we rate this application as a high threat in your network. Not That I don't trust you. No, No, absolutely you shouldn't.

That's why the observability piece. So follow up. Um, question for you on that.

If you're taking traffic that's routed in from your customers, how do you tell when you have like a spoofed endpoint in the threat assessment? So the traffic, so we have a very server centric view of the world, like, uh, our SL just mentioned. We are not decrypting any of your traffic, so we run it through a gamut of signals.

So, uh, I'll walk you through how we are getting to this compression ratio as well. So what we do is every net flow or IP fix record is augmented with a DNS, uh, packet, DNS header. If you really, if we can see the DDNS packet that's traversing our CPA device.

If we can't, we have the DPI engine that's actually, uh, tracking the sign. It's a signature database. So, uh, I like to say if it looks like a dog, smells like a dog, must be a dog kind of database to, to actually get a signature match on the application.

Then we send these records upstream to the north, north side of the communication here, where we are marking it through a gamut of signals so you don't have to really install this threat. Intel on the networking operating system itself. Go through frequent upgrades to get more threat intel down to the devices.

To avoid that, we run it through a URL database that's around 240 million URLs updated every hour, a threat Intel 10, uh, a database of 10 million ips updated every year, A CDN IP database around a hundred million ips that updated every year. So we are trying to track one, trying to identify the URL endpoint you're trying to talk to, trying to identify who the server side is in the communication, and then match it up against all of these threat intel databases to really figure out who that endpoint is, where it's located, uh, from the spoofing question, right? I could be, uh, a provider in let's say China, who is pretending to own an IP that really belong Like AWS or Azure's very common spoof for an attack on the network.

So we have seen That, we have seen that, uh, uh, at our customer sites actually, we've seen an A-W-S-I-P actually picked up by a threat actor o on, uh, for, for a couple of hours, uh, uh, uh, a nation state actor for a couple of hours. So that that outer header we are matching up, it's almost like a zero day attack. We mark and sweep through all the ips that we have ever seen in your network to really see if this IP that was deemed as safe has now, uh, been marked as malicious in intent because of new threat intelligence that has come in, even if it was spoofed, uh, to A gotcha.

And I'm assuming that goes in with the geofencing doing the geofencing you're doing in your policy that if I suddenly see origin traffic from, you know, China or Russia or wherever it is Exactly, and my policy is to be continental US or whatever. Yeah. That that's, you can't, You can't go there anywhere Regardless of whether or not it's spoofed, it falls under the, that policy, It falls under that policy, and I'll showcase that.

So you're going through a gamut of signals that really are enforced at the controller level. What it does for us is we, it takes the burden off the infrastructure guys, the IT guys. You don't need to set an upgrade.

Every router, every time you get new threat intelligence or new signatures. In all of this, intelligence is streamed real time into the platform. Any new threat intel, any new rules that come in that you want to enforce, get enforced at this control layer without really affecting it, is it, Is there any remediation that is done for these 15, uh, potential threats?

Yes. That's, you could block all of these with a click of a button. So I wanna really block this.

All I do is, hey, block, I don't trust this anymore. I nearly want to quarantine this out across the network infrastructure. It's not specific to a device.

It could be across all land segments, all sites, all regions. Hey, I don't want this endpoint to ever be, uh, consumed by any of my hosts. And, and what if I, if the now I've, I've said, I, I agree that that's a threat.

Yeah. And so the things that fall in that same bucket in the future, I can automatically stop. You can automatically, and They won't be showing up on my dashboard anymore.

Exactly. Exactly. And then the follow up question is, I screwed up and I shouldn't have automatically done this.

How do we see that that was what's screwed up? The connection that's randomly not coming piece. So that, so you'll see, you'll see that in the block bucket.

So I'll, uh, segue into how we categorize the intent of the application, but you're giving me great segues. So let's say you're running clean in your network, right? Uh, hey, I have all my security and compliance issues sorted.

There are no threats. 3,700 apps are still a lot of applications to monitor. So what we do is we neatly bucketize them based on their intent.

Broadly speaking, you have the enterprise SaaS apps. These are the O 360 fives of the world, the, uh, sales forces of the world, the public internet apps, these are the general apps, the CNN dot coms of the world, the YouTubes of the world. You have the AI apps.

You'll have folks on my corporate network going to chat GPT and open AI this morning. So yeah, so you essentially get to see large buckets that you can ignore because you fairly know what the intention of the application that's running. You have the unsecure apps that you really wanted to quarantine any way that we've just showcased.

And the blocked application, things that you might have intentionally or unintentionally blocked or dropped in your network really showcase up in this block bucket. These could be policy constructs that you pushed to get something malicious out or there these could be acls that you accidentally pushed and they're blocking legitimate traffic. So you start seeing what these network issues are, which applications are getting dropped, and for what reason and the ability to unblock them right here because you get to see which policy is actually dropping them.

Then you have your crown jewels, the apps that we were talking about, uh, your exchanging with the B2B partners or within your enterprise e essentially things that I want to exchange on this data fabric. Let's say I have the Google Cloud, the Gemini service that I'm consuming for my AI application. I get to say exactly where it's being consumed from.

I can enforce compliance, but before I enforce compliance, I need to know where these are running. So what I do is I view that apology on that global backbone. This is the almost uber-like field that I was mentioning.

You get to see both spatially and temporarily of how this application was consumed. You get to see for the past day in a temporal fashion that this application was consumed within the North American boundaries. You get to see the exact locations that consumed it.

You get to see the fiber links that were taken for this communication and all the telemetry along that path. You also get a spatial view of the world of all the sites that use this application. You get to see the host that use this.

You even get to see every session this host ever had with that endpoint. And with one click, you get to see the exact path taken by that application for its consumption. The host sitting in someone's home office is taking the at and t provider through us.

San Jose Pops, taking the at and t provider through our US West pop that's directly peering into the Gemini service. That's sweet. Yeah.

Well, sure. Uh, so what I'm seeing here is you're extending your value proposition in that people seem to test the water with NAS services and do partial. You're making a play for global because you give them all this extra value and simplify their life, which is a big thing lately, just not enough people and time to go around.

Absolutely. So three things that we really want to do well, any to any connectivity end-to-end observability and absolute control. We want to do those three things really well.

And those are the three fundamental pillars that we want to showcase. You can get connectivity into your business partners, into your cloud providers, into services, like I just showed you. You get to see how that application is moving.

It's fully auditable. So I like to say there are two personas that really consume this. The network persona really thinks of this as mean time to innocence.

The application guys always want to blame the network for the application performance. Within three clicks, I can prove or disprove whether it's the network or not. I can show you the exact telemetry all along the path.

I can show you the client and server details. And no more than three clicks for my security persona. I have absolute auditability of every session that's gone over this global network.

So I can really get my compliance and regulatory affairs in order and enforce my security rules across the planet. It, And when that ticket comes in from a particular client, I can search it from that direction as well. I assume that Yes, exactly.

I've got a ticket from a client had a problem with this server application at a particular time. Exactly. You can pull up that, You can pull up this and you'll see both a spatial and temporal form of that data.

So from a security, uh, persona, I can quickly say, Hey, this data has not left the US boundaries. At least I'm safe on that front. I go to the second order problem of the performance itself, but it doesn't matter which endpoint is, whether it belongs to you or to your business partner, you can really track it in the same fashion.

So sometimes from a security perspective, the decision isn't, um, cut that client off right away. Sometimes it's, we need to understand what's occurring first. Yes.

How deeply can you drill into what, obviously you can, you can look at sessions, but can you look deeper than just simply the session data for, uh, you know, for a client that's, that's connected. So from a connectivity perspective, you can pause the activity. You don't need to disconnect him completely, or you can mo redirect that traffic to your scrubbing endpoint to really see what's happening.

From a payload perspective, we are not decrypting your data. That's fundamentally not what we want to do. Sure.

Uh, and it stems from the fact as if you are decrypting data, we strongly believe that data should stay within your private boundaries. If I'm sending it out to someone else's club to really decrypt and scrub the data, it's only a matter of time that gets breached. There's no way, uh, one, one environment safer than the other.

So we are not decrypting the payload. So if there's ma, if the question is, is the malware or is it a phishing attack in the email, we wouldn't know. We are not looking into that.

You're Not looking that deeply. Okay. That, that deeply that Answers the question.

Thank you. Yeah. But we are pairing with partners that can really scrub that data.

So, hey, I want to move this data. I need to scrub this need to see what, uh, more details into it. Just like I've peered with, uh, the Google services over here to pair with your secure access provider to really scrub that Is, is there a product, um, or, or something a customer's built that you're typically displacing?

So you're using this instead of talk about what those are. 'cause I can imagine if you Absolutely. Um, so they are, think of the NA space.

People like to put things in buckets from a connectivity perspective. This is, this replaces your SD-WAN solution. So you have your SD-WAN vendor of the world from a visibility and observability perspective.

You need another artificially probing kind of observability tool. Like say, say the TEX and the thousand eyes of the world. You for your cloud exchange to get to your cloud, you have a third provider, a third vendor per se take say the packet fabrics and the backup ports of the world.

So essentially what we see in the networking domain, at least in the enterprise space, is three or more vendors building a single enterprise network. It's a very vendor centric model, and the enterprise network is pretending to be one network. It's three or more networks pretending to be one, add a B2B partner into the mix.

It's three or more vendors he has. So you essentially have six or more networks trying to exchange data. That's why the second order problem of another observability tool comes in place.

There's no way to see how that traffic's moving. Mm-hmm. All you can do is guesstimate what's happening through probing that, those three networks that brings you to those tools like the kentex, the thousand Eyes of the world that have a space because of the vendor product, if you really unify it as a fabric, now you can see not through artificial probing through your actual data, how that data is in motion across the planet.

If I can see how that data moves, it helps you to really enforce compliance. And that's a problem in the enterprise space too. You'll see almost any enterprise network look like a hub and spoke.

And why, because I can't have a unified policy construct across three or more networks. I need to get my data to a scrubbing point. A hub facility where I enforce policy.

Two problems emerge from it. A single point of failure, you need to keep those data, data centers up, the tunnels, sprawl, everything goes to the hub. So you need to buy a massive router that can hold those tunnels up and keep them up and running.

And second you fat finger one policy into that policy construct. And that's it. You, you are out of luck, right?

Data is moving out of your jurisdiction before you know it. Once you get it into a fabric, you can really see how your data's moving and you can enforce it at all the edges that exist in your infrastructure. You can block it out of where you don't want it to be, and you can consume it from where you want to consume it.

So just a question on that, uh, Dustin, how much customability do you guys allow for policies? You know, coming from an MSP space, there might be certain times in the day where we don't want customers accessing certain applications, certain services. What sort of custom ability do you guys offer, uh, within, you know, that regard?

So, uh, from an MSP perspective, essentially we have route level policies, programmability of the backbone that's unique to graphene itself. You have, uh, I like to put it in again, three buckets, right? Uh, you have the path assurance, the performance assurance and risk assurance.

These are the three kinds of policy constructs. I like to think of the risk assurance stay that hey, I wanna stay within compliant boundaries within geopolitical boundaries. This is a HIPAA data.

There's GDPR data. My data is my most sovereign asset, and this is where it should recite. And my, uh, laws should apply wherever the date data moves, the performance assurance policy, essentially.

Hey, I have x-ray files that I can't really compress. I want bandwidth to be the highest priority across the backbone. I don't care about the latency.

So you can get performance style of policy constructs and the path assurances. You can even control what links and providers you want to go on. Let's say I am part of the defense sector and I don't trust some kind of equipment coming from a particular nation or state.

I can control exactly which providers, which links I use to exchange that data on the platform. So you can really move through all these dimensions automatically based on your pipeline. All of these are exposed as APIs.

So people integrated into their APIs. Our customers typically put it into their pipeline. As applications are developed, the policy constructs are auto-generated, ports are opened up where they need to be, and they restricted where they need to be so that the consumption of a new application is completely agnostic to the developer of the application itself.

So the last piece of the puzzle in the interest of time is the enforcement of the compliance. Now that I know how this data is moving, I can really enforce PO policy on that data. I can say across my sites, across the enterprise for all the land segments, whether it be corporate or guest wifi.

I want a topology selected that could be hit for a lack of a better term, GDPR or hipaa or some topology that I want this data to stay within. And, and this has a log option, so I can do log it first, so I can look at what's gonna be, what's gonna happen Locked exactly Before I Yes, that's You. You can just log all of all the policy constructs that I mentioned.

Yeah. Not That I don't trust. Yeah.

So, uh, essentially all of the policy constructs have the logging option. So you can really simulate what's gonna happen, uh, beforehand. With that.

And in the interest of time, I will dive into the last demo if the, unless there are any other questions on, I have one, one quick question for you, Kevin Myers. The, um, so you've built the box, you've built a transport network, and I realize you have a virtual appliance as well, so that you enter the Grapht network to be able to take advantage of the data assurance and all of that. What happens if the grapht network has a bad day?

What is that? You mentioned that you're trying to get beyond being a transport network and have the solution exist outside of your own transport network. Customers are establishing a B2B peering with you.

Do they just route around it and they lose that capability? What's the plan for that kind of a scenario? So, For our backbone networks, we want to get the best of both worlds.

That's what Raffin was built on. SD-WAN that just using the internet, throwing the packet on the ether of the internet, hoping it reaches the destination. And the MPLS style of the world where you have absolute control and SLAs that you can guarantee because you own the infrastructure.

From our backbone perspective, we have a resilient backbone. So we have redundant links, we have redundant pops. Uh, all of the edges, in fact, auto, uh, uh, scale out and scale back based on the availability of the nearest backbone node.

The redundancies on the provider at the edge, the redundancy on the provider links facing, uh, the internet on the backbone itself. There's redundancy on the transport that's connecting the backbone links across the globe itself. So there's almost three levels of redundancies and the device having intelligent to auto really move connections based basically.

I Guess that, that's really what I'm asking. 'cause every, you know, I'm sure you have plenty of redundancy built in Yeah. To the architecture, but you know, every network has a bad day.

Absolutely. So if you do, and you have an issue there, you're, you could write a policy that says, I will use the internet as my underlay to fail over if I so choose. Is that, does the customer make that decision, or is that something you would plug into the graph box and make that routing decision?

What does that look like? The customer could make a decision on that if they want to. The policy construct allows it to do so, but what we do automatically within the graph software is to try and get you to the next available pop next available link that to, that gets your traffic on this private backbone.

We want your private data to stay on the backbone and not really just go, uh, onto the underlay fabric. Okay. So at, at the moment, there's, if, if that is unavailable that, that it's, you're just gonna drop the peering to the, to the client or it's No, So it's, uh, think of it as, uh, two legs in completely different parts like Seattle and San Jose both.

Oh, I mean, I get it. I, I build service providers. Sometimes you have things that affect the entire, you know, you know, you have an issue with the IGP or bgp.

I guess one of my, the, or the real crux of my question is, is it, what is the, if if that's unavailable for whatever reason? Yeah. If the redundancies are all unavailable, yeah.

What's the expected behavior? The Expected behavior is if you have an underlay network that can really, uh, route the traffic out. You could have a policy construct saying, Hey, this is my preferred, this is my backup preferred.

And then, hey, this is okay, just fail safe. So you have the, like the, so of, So you have the ability to define a policy. So even though it may not meet the data requirements that I have, you'd rather have connectivity over the data assurance for a period of time or not.

So, So people implemented the other way around generally is like, uh, if it's a compliance requirement that this data doesn't really leave my jurisdiction, I'd rather just drop it even if I have underlay connectivity. Yeah, that was really, that's what I was getting at is, uh, you can basically say, I would rather it be down than be non-compliant. Non-compliant.

Exactly. If you, if that's your decision. Exactly.

Okay. So that, that's how, uh, the, and this is because the source is routing it. Unlike the internet where there's hop by hop routing where each one's making a decision independently, the source decides whether the entire parts compliant, if the label stack that they're gonna put in onto that packet header really is compliant for your, uh, regulatory needs.

If you don't find an end path, an end-to-end path, you're just dropping that And that's why you're using SRV six. That's Why we're using. Okay, cool.

So The last piece of the puzzle, Gina, introducing graphenes intelligent network assistant, like I mentioned, she runs my standup meeting, buying back time for me, things that used to take me 60 minutes, multiple teams, multiple touch points, multiple dashboards, ask Gina to guide me into how I can access that data, what data is interesting. So she's not someone I'm just querying. She's giving me insights into what I should query and lets me run tasks that are important to me.

So saving me time, essentially by running my standard meeting in real time. But Gina is going to really tell me what data is interesting, what she has access to, and what I should be asking. She's going to give me insights on the operational state of my network.

She's going to tell me what needs my attention from a security and compliance perspective, the malicious actors that I might want to get out of the network. She's going to give me insights into the compliance reports that I need to publish for my security officer to consume and show me trends of where we can do better from a compliance and regulatory point of view as well. All of this action and activity happening autonomously as if she were a member of my staff.

These questions typically take me through all the demos that I was showing you right now, drilling down into every session, trying to figure this data out, getting those reports in, talking to members of my backbone team that are managing the backbone network, the devices, teams that are really monitoring the end devices and my customer success teams that are really monitoring alerts that are coming across the planet. Gina has given me all of this information in 60 seconds. I know it's a blink and you miss it demo, but that's the intention of it.

She's buying us time back, which is really valuable when you're trying to operate a global enterprise or a global network. You're Creating an, uh, an AI agent based on your physical, what your physical employees analysts are doing, doing originally. Yes.

And so you're, you're taking the, the processes that they've validated over over the previous year. Yes. Uh, and now you're automating that into, into ai on, on the data.

Uh, exactly. That's, that's here. That's here.

Okay. And now obviously this has been trained on the, the data that those, those people have generated to this point, right. How often do you do the retraining and how often is the new data, uh, pushed in?

So we've taken a, a different approach. We're not trying to train, uh, the model on our data itself. That means we'll have to exchange customer data with the model.

So we're not doing that. We're not doing that at all. The data's not used for, uh, training purposes.

So what we are doing instead is giving gene the questions that are interesting to be interpreted by Gina to the model. So generative AI to just interpret what the questions are, and then giving Gina access only to the APIs that are accessible to us as users of the portal itself. So she doesn't get direct access into all the databases that we have at the backend.

She gets as much or as little access as any user would to our portal. So you have our back capabilities, just like a user would have. So she's getting only access to APIs or GRPC calls that have access for a end user.

So what happens is she can now guide you to what she has access to, and she can answer questions, what you would've wanted answered through the portal itself. So you secure what you're really exposing from an end user perspective. We, we want to err on the side of caution, and that's why we are trying to keep it as a closed system in terms of how we are operating this.

And from here, we can start to dive deeper on particular Absolutely. Branch Absolutely you can, Using that same Yes. Process, You could keep asking her more questions.

What we've added on why I say she's beyond a chatbot, she's a part of your workforce is because she's carrying contextual awareness of what you've been asking across days, across months. So you can really ask her a question and she's going to keep that contextual awareness that, hey, this is a CISO who's asking me for a compliance report. So the questions around threats question, uh, uh, around compliance are more important to someone of that persona as compared to someone who's in product or in, uh, a backbone SRE team who wants to know about the alerts on the infrastructure or the CPS that are leaking memory for, for some reason.

So that contextual awareness makes her a member of your team that can really answer questions that are relevant to you and answer questions in the jurisdiction of the API she has access to, which are no more or no less than any one of the members on this portal And includes the public data that you have access to. Yes. It includes a public data.

Yeah. And Is Gina available only for graph networks or does it work with other networks as Well? It's available on the graph network.

So, uh, if we are kept it as a closed system, like I said, we, uh, the infrastructure, the architecture supports being an MCP server to allow other agents to come in. We want to see how that plays out. It's a switch of a button for us to really expose it to other agents, but we really want to be cautious about exposing our customer data.

Just like we don't want to decrypt any of your data. We don't want to give direct access to the data we are procuring on our end to other agents without, without vetting of sorts. So Gina is a closed system for today, but the architecture is exactly like an MCP architecture.

So enabling it is just a switch of a button. We just Gina today, right? Hmm?

You announced Gina today We announced it. Is it available today? It is available ga.

Okay. Uh, in fact, all of what you saw is a live demo. This is corporate traffic.

I was looking at it as real time as y'all were. I didn't have this precan. I was looking at folks running traffic answering questions as we, as we run through it.

So to put a finer point on what you were asking, um, so in a sense it has role-based access because it's the user who's logging in, and then they, they have role, role based access, and so they, Gina only has access to the data that that user has access to. So, okay. So kind of a, maybe another follow up on that question.

We've talked governance now we, now we see a compliance report. Uh, is there any way of maybe comparing where you sit currently against one of the security frameworks, for example, uh, maybe PCI compliance or any, any of those things? Is there any way of comparing your current configuration we'll say to, to those, uh, those reports that we, that they're often have, uh, ran in?

Absolutely. So what we call it the gap score in our terminology. So you'll see the GAP score and she'll tell you the trend that it's increasing.

In this case, essentially what it means is the graph wind assurance posture score that you see over here. Think of it like a credit report inspired from credit reporting. We are losing or getting points docked from a total of thousand points or bad behavior.

Okay? And we essentially then tell you what those behaviors are, so you can really start seeing how to improve your Post. But what about mapping that towards one of the frameworks that that many organizations have to map their security?

So what we do there is, uh, we let the enterprise really dictate what compliance needs they have and what it means on the network infrastructure. So you, you saw the, a portion of the, uh, enforcing compliance where you can select the topology. Mm-hmm.

That topology is actually programmed by raffin on behalf of the customer made available to all industries within that jurisdiction. So let's say GDPR is common for almost everyone in the European Union. If one of the enterprises says, this is what GDPR means to me, it's likely the same for everyone else.

We program that in our backbone. We program that in our gap score, and let you then, uh, check the trends of how you are, uh, how you are doing against that, that metric. Okay.

So.