Introducing NautobotGPT: Your AI Co-pilot for Network Automation

Hi everyone. Thank you so very much for joining us today. We’re very excited to have all of you joining us for our first ever NautobotGPT webinar. We’re going to give folks a couple of minutes to get uh get in and get settled. So, please bear with us. Uh but again, thank you for joining us and we’ll be starting in just a moment. Again, for those of you just joining us, I want to welcome you uh to our first ever NautobotGPT webinar. Thank you so very much. We had an incredible amount of interest in this webinar.

Um so, obviously, I think everyone’s very excited about what NautobotGPT and AI can do to support your adoption of Nautobot and network automation journey. We’re going to give folks just a minute to get in and get settled. still have a lot of people showing up. So, I want to go ahead and give them a second to get in and get settled. Excellent. Uh seems like the pace of folks joining is just starting to uh to sort of level off. Uh I expect we’ll get a few more over the next minute or two.

Let’s go ahead and uh bring up uh a quick poll while people are waiting. Uh just a quick uh poll uh for those of you out there uh just sort of try to get a feel for where you are in your network automation journey and where you are in your adoption of Nautobot. So please take a moment or two uh give us your feedback on the poll uh and then we’ll come back to that in a couple of minutes. While we’re continuing to wait, I’ll cover off a couple of our housekeeping items here. Uh first, the chat is disabled.

Uh if you want to ask us a question, please use the uh Q&A function across the bottom of your screen. Uh that’ll get us the notice that you’ve got a question. Give us a chance to respond to it. Whether it’s a question that you want to ask uh either myself or Tim about the subject today or some other issue you may be having with the webinar, you can just put it into the Q&A function. Uh, and we’ve got folks watching that to respond and take care of any issues you may be having.

Um, in addition to that Q&A function across the bottom, we also have a reactions button down there. We would love to see you folks use the reaction button throughout the webinar. Give Tim and I some feedback. Let us know what you’re finding interesting uh or anything else that might be going on uh that we that you think is insightful or giving you information that you like or want more information on. Certainly would appreciate you using that uh that reaction button across the bottom. Fantastic. All right.

Uh looks like we’re uh about where I expected us to be uh participant wise. So, we’re looking good there. Fantastic. We’re gonna go ahead and get started. Thank you uh so very much for those of you just joining in the last couple of seconds. Uh there’s a poll up on the screen asking you to share where you are in your network automation journey. Uh where you are with your adoption of Nautobot. I’ll give you guys another five or 10 seconds to respond to the poll. Uh and then uh we’re going ahead and kick this thing off.

So, please uh give us your poll responses now and uh and let us know where you are with Nautobot. And with that uh that being shared, uh let’s go ahead and bring down the poll and uh kick this thing off. Uh thank you again everyone for joining us. Very excited to have you join us for our first ever NautobotGPT webinar. We think there’s a lot of great information in here. Get you up to speed and a little more information on NautobotGPT in particular, but also share a bit more about what you can do with it and how you can access it.

Today we’re going to be talking specifically about the automation uh adoption challenges, what sort of drove us to try and build GPT to help folks out, help you overcome some of those adoption challenges. Then we’re going to dive right into an introduction to NautobotGPT itself and its capabilities. And then Tim will walk us through a demo, little follow-up closing out the loop on some more information on where Nautobot’s going, NautobotGPT is going next, how you can get access to it, and then an open Q&A.

And again, use that Q&A button across the bottom if you have any questions. Certainly that Q&A at the end is where we’ll dive into any of the big heavy questions, but if you have any questions along the way, happy to take those sort of in uh in stride with the presentation here. My name is Jeff Bradbury. Again, thank you for joining us today. I’m the vice president of marketing here at Network to Code.

And joining me today is Tim Schreyack, our director of strategic transformation and the guy who’s been doing all the heavy lifting, showing you NautobotGPT, walking you through the demo and sharing his experiences as somebody who spent years helping our customers adopt Nautobot and network automation. So, he’s got a real subtle touch in understanding how NautobotGPT helps folks overcome some of these automation challenges. So, we talked about this a little bit right, automating anything is complicated and it really has two main ingredients.

Let’s go to the next slide, Tim. We’re really looking out here. How do you automate anything? And really it comes down to do you have domain knowledge about that thing you’re trying to automate and do you have the technical capabilities you need to operationalize that automation. Really two big uh uh overarching uh objectives there. And when we dive into the network automation space on the next slide we can talk very specifically about what those two ingredients look like.

It’s the domain expertise around networking and network management and network operations that you absolutely have to have as well as the right tool set, the right development capabilities uh and the right data models that you need to execute that network automation capability. And again, this is all built on that data management layer of having the right validated robust data set to run your automations against.

So for network automation, two key pillars here that network management expertise as well as that development expertise and we essentially built NautobotGPT to support those two pillars to help people move forward with their automation efforts. So when we talk about NautobotGPT helping everybody on the team, we do mean everybody. So NautobotGPT is not just for those folks who are hands-on keyboard. It helps newer or less experienced engineers and developers get into network automation. Uh part of this is learning by doing.

As you’re sitting down and you’re working with NautobotGPT trying to do something or trying to learn something, NautobotGPT is there to answer questions to show you how it did something or why it did it. And you wind up sort of learning by doing because NautobotGPT is walking you through the steps you need to do to access Nautobot jobs, access the data models, define a device, all the things you need to do to make Nautobot work. It’s guiding you through that process uh with its knowledge of Nautobot and its knowledge of Nautobot automations.

A big step forward and helping somebody who may not be greatly experienced in either the development side or the network management side, getting them up to speed and walking them through how to do those sorts of things. Second of all, you’re interacting with NautobotGPT with pure human language interaction.

And in response, you’re getting very technical responses, uh, very detailed level responses to your network management questions as well as to your development questions, getting actual code back to help you do some of the automations you’re trying to do. And then lastly, anytime you run into trouble, anytime you run into something that doesn’t work properly or you get an error message, lot about GPT is right there to help you troubleshoot and figure out what’s going on.

So great tool for taking a new user and getting them up to a capability where they’re able to do stuff on their own. Now for those folks who are already experts and there’ll be many of you in this uh webinar who may be experts in Nautobot and and doing this stuff GPT helps you as well. It can help you accelerate everything you’re doing. And in fact our own network to code experts are using NautobotGPT this way and it is enabling them to do so much more so much more quickly.

So, first of all, just that overall accelerated development that it just allows you to do so much more quickly. Second part is it allows you to go from ideation to execution in literally seconds. Um, you have an idea, you want to try something out, you think there may be some capability or some way to make something happen that you want to have happen. You simply ask NautobotGPT in your language.

It could come back with a prompt and tell you this is how this might work or here’s a code snippet you might use to test out this theory uh or here’s uh you know the underlying uh data models that you would need to connect to make this work whatever it is that you’re asking it it’s helping you to rapidly prototype those solutions and lastly it’s that trusted net devops partner that companion who sits next to you you may be deep into the middle of coding something or writing something or you know really in the zone of of working through some sort of logic problem and now you need to know something about what the data schema is you need know something about how this API works.

NautobotGPT is there to answer that question for you, to keep you in the zone so you’re not code switching back and forth trying to do multiple things to work through your problem. It’s that companion that’s there to help you accelerate what you’re doing. Great tool for expert users to help them do more more quickly. And then lastly, as I said, NautobotGPT isn’t just about helping those with hands-on keyboards. It really does function as a key tool to support management and leadership as well.

Of course, overall the goal here is to accelerate project delivery and get things done more quickly. No doubt about that. And that’s certainly a big outcome that management and leadership is looking for. But from a more tactical perspective, it’s letting you get more value out of your automation investments. You’ve got more people who are able to do more things more quickly. You’re getting a lot more value out of your network automation investments. a lot more people contributing to your network automation outcomes.

It’s just all around a way to get greater value out of every individual on your team. And then lastly, it allows you to shape your network automations around those specific uh business strategic initiatives, whether that be cost savings or compliance issues or whatever it is, not by GPT can be there to help you shape your automations to those specific outcomes you’re looking for.

I’ll note one more big part of that automation ROI piece here and that is the query capability especially the ability to talk to your Nautobot data allows management and leadership to ask and get answers to their own questions that would typically take an engineer to go get an answer for them. You now have a way for those uh in the management level leadership level to ask those questions to get that their own information back without having to involve an engineer every single time.

So it’s a great way again of democratizing the uh the automation capabilities across the team. So now with that background I’m going to turn it over to Tim and let him really walk you through introducing you to NautobotGPT highlighting its capabilities and most importantly get you to that demo. Thanks Tim. Take it over. Thanks very much Jeff. Great to see everyone. Uh love seeing such participation here today as we introduce this amazing new feature.

Um, every time we announce major features, we always get a lot of you joining the webinars and you have really crushed it today. So, thanks. Glad to see you all out there. Um, well, let’s take a quick peek at that uh poll that we did earlier. Let’s just see what the results were. Kind of curious to see what the population is that we’ve got here across our audience. Uh, so where are you in your automation journey? So, here we see uh about 46% of you have not installed uh Nautobot yet. So, great seeing all of the new faces out there. We love it.

Um, and then we’ve got the next largest group here. It looks like we’ve got it installed and are actively using kind of a significant group in there as well. So, welcome to all the newcomers. Uh, love to see you here and hope that you find what we’re talking about here today interesting. And we’ll go through this amazing new feature that we’ve got with NautobotGPT.

So, we’re going to look at a few more slides just to give you a little bit more details about how you might use it and where it can uh really help you as Jeff was explaining at a high level there. And then we’re going to get right into a demo and start showing you how we interact with it as we move forward. So let’s go ahead and get into it. So NautobotGPT accelerating automation. Uh basically what we see is you know the two biggest blockers to automation success are knowledge and time.

So automation requires deep expertise across systems, data models and integrations, right? And even when you know exactly what you want to do, the effort to implement it can be really really intense. You know, often you know scaling with the value that it delivers. And that’s why we built NautobotGPT. is going to break down these barriers by helping you learn faster, code less, and achieve more. It it’s like having a team of senior automation architects on demand basically guiding every step of your journey.

And whether you’re just starting out or working on the edge of what’s possible, NautobotGPT is your ladder that you need in order to climb and really hit that gas pedal so you’re ready to accelerate. So you can see, you know, we’re really looking at how we can move forward uh with this journey and help you no matter where you are in your process. So what makes GPT different? You know, other LLMs have things to offer, but what really differentiates NTC’s LLM here?

So, you know, while it is a chat GPT-like experience in terms of ease of use and what sets it apart is the content that it can access and what it can do. So, NautobotGPT is powered by NTC’s internal repositories, not just what’s available publicly. Uh, it also has our expert knowledge and purpose-built agents for tasks like Nautobot job creation and Nautobot data exploration. These capabilities are exclusive to NautobotGPT and tailored specifically for realworld network automation. Um, and that’s really a key element here.

We are not looking at something that’s just uh you know let’s say marketing fluff if you will. This is how you actually do real work. And in fact we are finding our team our internal uh automation experts and developers are using this tool heavily to really help them accelerate the speed the time in which it takes to develop code and develop features for Nautobot and for our clients. So we can see now you will also be able to take advantage of this in your world uh and be able to accelerate your development on top of the Nautobot platform.

So if we think about this right as you look across this journey here uh it’s really designed to accelerate this process and if we think of it without NautobotGPT we have a lot of steps here in implementation right we have the platform setup we have data on boarding we’ve got job creation and application creation and then finally integrations right the different things that we might be exploring and as we look at that across that timeline it’s going to take some time when we’d have traditionally done that right and what we’re seeing is a drastic

improvement in speed with NautobotGPT where we are seeing this huge decrease in the amount of time that it takes to do things where something is no longer being measured in hours or days but now tasks are being measured in minutes and it really is showing a tremendous amount of value that we’re seeing here as we use it internally. So we know that you are also going to see this kind of benefit as you start to be able to use it as well. So we’re seeing a lot of benefit as we see this really just speed to market of getting work done faster.

So here’s where it all comes together, right? As NautobotGPT isn’t just trained on the docs. It’s modeled on the experience of our over 100 automation engineers and developers within NTC. That collective knowledge is now available to every user through a simple chat interface. So you can ask questions, you can create jobs, you can troubleshoot errors, and even soon you’re going to be able to interact directly with your live Nautobot instance. That is a coming future capability.

Uh we are going to demo it here today, so you’ll get to see what that looks like, but just so you know, that feature is not currently available and that will be coming very soon. Uh this is also just not a productivity boost, but it really improves what your team can achieve achieve with Nautobot. And again, just really hammering that home. This is not just for engineers and developers. This is for our managers, our architects, for our users.

Anyone that needs access to data or something that they might do within Nautobot’s via these natural language queries that we can now explore as we begin to use NautobotGPT. So now that we’ve seen what it can deliver, let’s take a look at how it works and what we can actually do for you today. So we bridge the skills gap and quickly deliver these results. In the old method, right, we might spend hours writing a new job from scratch or creating a job.

We might spend a lot of time trying to dig through documentation and determine the root cause of a problem when we’re troubleshooting errors. Or maybe we’re just trying to understand some original code that someone else wrote and how do we make a change to it. So we want to customize an existing job or even just where how do I create a report? I need to get some data out of Nautobot so that I can actually consume that data and use it to uh you know do my job. Uh how do I do that? How do I do I have to use some you know other tool perhaps?

Whereas now we’re able to do this all via NautobotGPT. We can generate a new job literally in seconds and we’re going to see that here in the demo today. Uh we can understand errors that come back from a Nautobot job or something else within Nautobot and understand what happened there, why did that happen and what can we do to resolve the problem. uh being able to add to existing jobs. Hey, take this current job and I want to add some new functionality. NautobotGPT can do that for us.

And then again, this idea of creating these customized reports, something we will also see here in the demo today. Really being able to get detailed, useful information back out of Nautobot so that we can proceed with whatever it is that we need to do, whether it’s operationally or managing the environment. Here’s some ideas of what these queries might look like. So these are just real some real examples.

We might have conversational Nautobot knowledge like how do I create a golden config feature to you know to verify my NTP configuration might be one thing we might want to just how do I do something uh versus actually let’s go ahead and create a job right so now create a job that verifies all active devices at a selected site as an example right so another another way we might work with it those code and error explanations hey you see a trace back in your error log copy paste that trace back plunk it into Nautobot and we’ll get some results back as far as what do I do how do I solve this problem.

And then of course, as we mentioned, is chatting directly with your Nautobot instance. This is where we see a tremendous amount of power coming out, particularly when we look at how we operationalize that data that we have within Nautobot. All right, so with that, let’s get right into the demo because I know that’s always the most interesting part for everybody. We want to see this actually functioning here and going live and what can we actually do with Nautobot. So let’s start off with this idea of reporting.

How do I interact with Nautobot’s so I can get information back that I might say use in my job? So let’s start off with a really common example like give me a list of the devices that I have in my Jersey City site. So give me a list of devices in Jersey City with their full names. And it looks like I made a little typo there. So let’s fix that before we proceed. Uh so basically what I’m looking for here is I want to get the full name of every device in my Jersey City location. So here we go.

We can see there we are transferring to the data agent and this is actually going to produce what we were looking for. Um as we see here in this agent of an agentic meth methodologies here we are getting out to the various and appropriate agents that might understand what it is that we’re looking for. Right? So here we go. All right. So here is the list of devices in Jersey City in their full names. Okay cool. That’s interesting. Uh how about you know can you make that a nice report uh in uh a nice report and uh include relevant details.

Let’s try that. Let’s see what we get back as we start to try to interact with this and learn how to use the AI. It could be very interesting that we could just have this natural conversation as we move back and forth. Uh so here what’s I find really interesting as well here is you’ll notice that what it is actually doing even though it’s going to produce a report for me is it’s leveraging the jobs agent.

So it’s actually producing a Nautobot job in the background that is then going to then produce this report that it’s going to present back to me. So pretty cool there. The way we’re leveraging the AI in and of itself to do things like this and here we go. And there we go. Now here is our report here with our list of uh devices. And you’ll notice it did give me the job as well that I might say, you know, want to save and and uh be able to execute later in the future as well. But here we can see this table coming down here with the report there.

Um and then say, hey, um you know what, that’s cool, but uh how about can you give me a make that a table and also include the end of support and end of life information. See what it does now. So now we’re going to expect that we’re going to get back a table and it should have some additional information in it. Right? So here you can just see hey as a manager or some unknown operations. I don’t have to know everything about all the data within Nautobot.

I can just start to interact back and forth with it uh and get those results there as I’m expecting. So really allowing me to go from nothing to being able to expertly get data in and out of Nautobot as quickly and as easily as possible. So, we just give that a minute there to produce this result for me depending on the level of detail that we’re asking for. Here we go. Starting to get these results back. And first, we’re going to get again that job that just kind of explains how I might do that.

And then it should give me a report which it decided not to do. Uh but the again the joys of live demos as we see here in the uh in the uh in the AI deciding to to produce a report that I could show me this results as opposed to the report. Okay. So, as we can see, here’s some ways we can interact with the AI. Let’s take a look at uh how I might learn about using the AI. So, for example, how do I create a compliance feature? Like, how do I do something in Nautobot?

So, how do I create a compliance feature to check my uh NTP status, let’s say, right? So something that we might commonly do um I want to learn how I might interact with the NautobotGPT and be able to learn how to do something within Nautobot . Right? So here we’re starting to see it producing a lot of information for me around how I might do that. So let’s take a look at what it gave me back. It’s basically saying, “Hey, here’s what you’re going to do. You’re going to go into the UI. You’re going to create that rule.

You’re going to create some uh structured uh compliance rules here. We’re going to use some rejax matching. uh we’re going to map that to the appropriate uh places and then we’re going to run the compliance. So, step-by-step instructions for what I might do. So, here we can see uh as we look across that, you know, those exact details there and then even some nice examples for how I might structure my YAML uh for inputting data into that rule. Uh and then finally, you know, the detailed summary here, right?

So, really cool and even some little extra information here. Hey, maybe you might also want to check your NTP permanently, right? So giving you a little extra information for how you might make it better from initially what you asked for. But now so let’s say we created those rules and now we’ve got our compliance rules and they’re running and you know me I I want to kind of see what’s going on.

So you know show me a summary of uh show me a summary compliance report for all the devices in my Jersey City uh location would be an example right something but again we’ll just stay focusing on there. So, let’s take a look at what I have in my uh Jersey City location from a compliance perspective. So, again, we’re going to see it transferring out to those different agents as it is going to read in that data from my Nautobot instance uh and provide that back to me. So, we’ll just give that uh the GPT a minute here.

And we do see lots of questions coming in. So, that’s great uh as well here. And so we can see uh quite a lengthy list of details here that we get back from our summary compliance report of you know here is the uh feature and is it compliant yes or no right for those particular things but they say okay that’s cool um could I get a uh you know show me a detailed uh compliance report for uh for the for the devices that right so so for all those devices in Jersey City.

Let’s take a look at a detail and uh actually start to drill into these specifics. So as it will produce uh now what we’re looking for is to kind of see a little bit more explanation about what’s happening in each of those uh locations and those devices and what is actually going on in the compliance side. Again transferring out to the agents as we wait for this to process here in the back end. Sometimes the GPT does take a minute longer than others as it uh depending on the detail that is pulling out of the database uh can take a minute.

So we’ll just wait for that. There we go. So here now we start to actually start to drill in and see some information here about exactly what is wrong uh on the devices here. And I think this one is really helpful just dynamically producing these kinds of reports where now I’m seeing not only hey for this specific device you know the feature and that it’s not compliant but what the problem is. Um, yeah, there’s some missing stuff. There’s some extra stuff, for example.

So, let’s take a specific example of, you know, hey, I want to look at a uh a specific device. So, let’s just try that again, but now we’re going to just look at a specific one. And I am going to copy paste that in there just for speed. Uh, and we’re going to look at one of those uh backbone routers that we have in Jersey City. So, let’s look at it specifically and see how we can start to interact with it and improve our ability to get details out of this. So, cool. Okay, we saw some summaries. Now we saw a little bit more detail.

Now let’s take a look at a specific device uh and those what’s going on there. So now we’re going to pull that same very similar data, right? But specifically at that backbone router and then we’ll look at how we might interact with that. So again, you know, we kind of saw a little bit of this before, but we said, “Hey, that’s cool, but okay, so now I know the missing and the extra stuff, but what could I do from there? How about uh include remediation steps, right, uh in that report? So, what would I do to actually go fix this now?

So, here we’re looking at what I might do to fix these problems. So, here we can see first it told us, you know, hey, the ones that were compliant there. And then on this one here, we need to remove the IP domain name and we need to add a different one. And then if we scroll down here into the details, we’ll notice here we’re actually getting the commands from the uh compliance remediation that I would need to just copy paste and go ahead and put that on the device.

Now, of course, uh this is AI, so we need to be reviewing everything as we look across it. Um but this is coming out of the compliance engine, so we should expect a high level of accuracy. Uh but it always is going to require some review to ensure that it is doing exactly what we expected it to do. Uh but you can see just how much time this is going to save us uh very easily being able to interact back and forth uh with these types of things.

Now, one of the things that starts to get cool about this as you get better at your prompt engineering is you can start to provide uh details to the AI that get you more specifically to what you’re looking for. So, here’s one that I ran earlier and I got a little bit more detail out of. I said, hey, you know, show me a detailed compliance report. Um, include only the failing checks and the exact remediation commands I need to enter as a device formatted as a table.

So, I’m getting a little bit more specific, a little bit better at my prompt engineering as far as how I might interact with the GPT. Uh, so just great way to uh again interact with it, learn how to use it, and learn how to uh specifically get the things that you’re looking for back from GPT. Okay, so now let’s move over to the idea of job creation, right? This is I think one of the most powerful functionalities within NautobotGPT is this idea to be able to speed up how we do things.

So we’ve seen some examples where the GPT has given us back the ideas of jobs in responses to other queries. But let’s specifically start looking at creating a job. So in this case we’re going to ask GPT to write a job that will updates interface descriptions based on a selected device and the connected device. So something really common that we do in network engineering, right? We put our descriptions out there based on what we might be connected to.

We say, you know, hey, this interface is connected to the, you know, the pier switch or, you know, maybe the spine or the leaf, for example. So in this case, we’re asking NautobotGPT to write us a job that’s going to go out and do that for us. So make sure all of our interfaces uh are provided correctly. Right?

So as we can see very simple query very simple prompt into the uh GPT and we get back you know uh some decent lines of code about you know 59 lines of code come back to it and as we look at this we might scroll through it and say ah we missed a few things uh it’s not letting let me see oh exactly sometimes depending on what it decides to do in each case it’s not letting the user in this case for example select the device that they want to run it against right maybe we want to be able to have a drop down that says that So we can learn how to improve our prompts.

Uh and here’s one that I developed longer that starts to get more details, right? So we’re going to say, hey, create a Nautobot job and let the user pick the device and then if it’s cabled, update the description to connected to. So you can start to see tremendous amount of very just natural language query that’s going to result in code.

So this is going to take this idea of I need to solve a problem uh and I could spend several days or maybe a week writing some code and troubleshooting some code to do that or I can prompt NautobotGPT to write that code for me and while it might not give me exactly what I was looking for it’s going to get me 80 to 90% of the way there. Uh so this means first of all just taking a look at hey we have nice logging and details provided in there. There’s good comments.

uh everything is done according to uh good practice and uh NTC recommended coding standards and of course if you have your own coding standards within your organization you can add those requirements in as well. So we’re making sure we get back good code first of all u and secondly it’s just going to save me so much time right here we can see a really well-executed job that now starts to become really useful and providing all of this detail even then you might think okay cool uh before I go ahead and you know load this into Nautobot and do all of the

things that I might do to actually execute it let’s just see like what what is it going to look like so like show me what the user form uh will look like so you know just starting to understand and really be able to work with it and confirm that okay, it’s doing what I wanted it to do. It’s looking correct. Right? So here it’s uh giving this idea of these are the things that the user is going to be prompt for.

Uh these are the things that the user is going to be able to do and so I can understand hey that user form is containing the details that I was expecting. Um if not you know maybe we can make some adjustments to it. And then here I think starts to really get interesting. um is there anything else uh anything I could do to make this better to or to improve this right like what else could I do what what else might I want to do in such of uh a job like this so right away it’s coming up with looking at all of our best practices that we have here in network

to code and things that we often do in our code that we develop for clients which is how do I perhaps maybe I should include a dry run option uh maybe I uh you know only want to update uh you know, show only updated interfaces in the output or maybe offer a verbose option. Right? So, just some great ideas of how I might improve what I am doing and and and in making my code better and more consumable by the folks that are around me.

So these are some of the really cool things that I can do with NautobotGPT as we look at this idea of being able to interact with not only the data that we have within Nautobot but be able to develop, create and execute code more efficiently. Um and even to this idea of you know how do I now run this? So here’s an example of okay what’s next?

uh how do I install this job in my not so now what do I do next so I created that code uh what what were my next steps right so here again I don’t need to go detailing looking into Nautobot documentation I can very easily find out exactly what I need to do so hey I need to save the job I need to put it in the jobs root folder adjust my files and permissions register load it etc right so step-by-step instructions for how to actually proceed here from where we were.

So just really amazing as far as being able to speed to market and get things done more quickly. Okay, that brings us through the a really interesting demo piece here as we were looking at those concepts and we’ll now take a look at kind of what’s coming up next for NautobotGPT. And for that I’m going to hand it back to Jeff. Thanks Tim. Really appreciate that. By the way, I see a ton of questions rolling here through the uh the Q&A section. Uh thank you for using that. continue to put those questions in there.

I know Chris is out there actively answering as many as he can. Uh we’ll get to any of that that aren’t easily answered by Chris in the Q&A as well. So, lots going on there, but we did get one Tim uh in the in the Q&A that asks, can we do a query against multiple sites? I thought that was since you just did the demo. Maybe you could talk to that very quickly about how you handle multi-sight queries and things of that nature. Absolutely. So, definitely for sure.

Yes, you can absolutely query against any of the data that’s in uh Nautobot uh with that you have access to. So yes, no problem as far as how you might mix and match that data across multiple sites. Absolutely. Thanks for that, Tim. So where are we going next with this? So what everything Tim just showed you is either currently in the GA release that we just put out on May 28th or it’s in the agent 2 release that’ll be coming out in just a couple of weeks, which is the talk to your Nautobot data release.

And I’ll get to that in just a sec uh in a second. But we do have a big plan for where NautobotGPT is going. And here it is for you. So you can see uh in the big scheme of things, as I said, the current GA release, which is most of what you just got shown, is the job code generation, the conversational knowledge about Nautobot, and the code and error explanation. That’s all done right now. Uh and that was done very specifically because there is no customer data interaction with that agent uh delivery.

Again, everything that’s in there can uh can be done without touching your specific customer data. So that’s why we released agent one and we want to be very sensitive to those folks who are very cautious about AI. I see a lot of questions in the Q&A section about uh data protections uh and data security and that’s why we opened up our initial GA release to not touch your data at all.

That being said, Agent 2, which is coming out in just a couple of weeks, um does allow you if you choose to to open up a readonly capability to your customer data. And again, I’ll get to some of those very specific questions uh in the Q&A portion when we could talk about some of the security precautions we’ve taken and how we ensure your data is protected.

But for agent 2, we are going to open up, if you choose to allow it, a readonly capability so you can do a lot of the things Tim just demoed, which is asking uh questions of your Nautobot data. Beyond that, we’ve got future expectations where we’re going to start to then allow a little more interaction with your data set.

um the the agent coming later this year, we’ll start to do data validation and support for our SSOT um API integrations allowing you to validate data and validate data sync from other uh source of system of record uh data sources, making sure that the data is correct, it meets all of your standards, it’s compliant, etc., etc. Beyond that, we’re going to allow uh the AI to start to interact with some of the other uh Nautobot ecosystem applications. I’ve seen a couple of comments about Golden Config uh in the uh in the Q&A.

that next uh uh uh agent uh will start to enable interactions with some of our uh more uh successful ecosystem apps including Golden Config, allowing us to make adjustments to the data model and and make some suggestions from the AI on how to support uh some of our uh support and use some of our ecosystem applications.

Much further down the road, we’ll get into a place where we can actually start to do data changes uh a data change capability and then ultimately much further down the road, they’ll be at a place where we may have some AI ops capability where we’ll allow the AI to make some automation recommendations and then go ahead and execute those recommendations. There was also a question about when will NautobotGPT be integrated into N Nautobot itself.

So, you can actually use it within Nautobot as opposed to using it as a co-pilot that sits off to the side. We’re going to expect to make that integration happen later this year. Somewhere around our third release or fourth release is where we’ll start to see uh bringing NautobotGPT into Nautobot itself and you can actually call up NautobotGPT and use it from within the application. Um so this is a very a very considered approach that we’ve done recognizing the um natural caution about allowing AI into your proprietary data.

And so we’ve taken an approach that allows you to move at the pace you’re comfortable with and at the depth you’re willing to go to allow AI to uh access uh your data in order to support your network automation efforts. Hope that answered at least at a high level and we’ll get to some of the more details as part of the Q&A. So big question we get all the time uh how do I get NautobotGPT?

So right now and continuing through the end of July, we will be reaching out to all of our Nautobot customers who currently have uh secured access to our tier 2 application licenses. Those are our um subscription-based licenses that allow you to do some of our more advanced uh applications uh u um integrated into Nautobot. Um NautobotGPT will be part of that tier 2 application suite.

And so for our current Nautobot cloud customers who are tier 2 application users, they are getting NautobotGPT right now and we’re rolling out accounts to those customers. If you are currently a Nautobot cloud customer but don’t have access to tier 2 applications but want NautobotGPT, the easiest way to get NautobotGPT is to go ahead talk to your account managers and add Tier 2 applications to your cloud solution. Uh and you’ll get NautobotGPT as part of this initial roll out.

If you are a self-managed Nautobot user and you are a premium support customer later in Q3 once we get all the cloud customers taken care of first you will start to get your NautobotGPT accounts. So again if you’re a self-managed uh Nautobot user and you are using one of our premium support packages uh you will get GPT uh a NautobotGPT account assigned to you later in Q3.

If you’re a self-managed user on a basic support package and want NautobotGPT, again, talk to your account manager, we’ll find a way to get you onto a premium support package and you’ll get NautobotGPT as part of that premium support package. And then later this year, uh for those who just want to go and access NautobotGPT on its own without worrying about getting it through our tier 2, um apps via a cloud customer uh or our premium support customers, we’ll make it available for direct purchase.

um for those customers who choose to go down that path. I believe that takes us into the Q&A. There we go. We’re into the Q&A. Fantastic. I’m going to uh quick take a look here uh at the questions, Tim, and then I’ll figure out maybe if I can group a couple of these together. Yeah. of while you do that I know there was one question about hey uh so if I have two engineers that ask a similar question let’s say days apart are they going to get the same response so uh as with all AIS the results may vary over time and even if you ask the same question

multiple times right you might get a slightly different response back what it will not do though is let’s say that I query about some data within Nautobot right that that the details of hey is this device compliant for example there There will be no hallucinations about that, right? That’s based on the data within your Nautobot. So, that will not vary. What may vary is the actual style or look and feel of the report or maybe exactly the code that it writes if you were to ask it for a NautobotGPT job for example.

Uh, but you know the results will remain consistent. Um, but as with all AIS, we do need to make sure that we are checking it, that we’re reviewing it because that is just the nature of AI. Language models do tend to vary slightly over time. Great. Hey Tim, I’ve got a great question here. Um, the question is this user wanted to be able to make sure that their configs were compliant and could possibly make corrections or notifications if something wasn’t compliant. I think you sort of demoed that, but you want to talk about that a bit more.

Yeah, absolutely. So, I think this is one of the really great features as we start to look at using uh the AI is being able to get to those details that we need specifically for the case that we’re looking for. So in this case you know it’s hey how do I you know say compliance and then remediate that compliance. Now of course you can do that within the golden config app. So if you’re using the gold config app that functionality is there. However um it might be difficult to get exactly the data you need.

So maybe what you’re saying is hey I want to make sure that my devices within a particular data center are compliant on their SNP settings for example and any that aren’t show me what I need to do to fix them. You can absolutely prompt it like that. And that is essentially what we did here during the demo as well to show this idea of hey give me a report now show me the details of a specific device or group of devices um and then tell me how I might remediate a specific problem.

Uh so that absolutely you can do that drilling down into those now give me the commands I need to copy paste on the device um in order to remediate and get to look like the gold standard. So I absolutely can do that. Great great use case for NautobotGPT. And let me let me jump on the back of that Tim because I think you gave a great answer. I want to kind of maybe set it up uh more holistically for this particular question.

You know, as you noted, if you knew you were interested in making sure configurations were compliant, right, you could go use NautobotGPT to write a job to go test those compliance features, those those compliance uh attributes and then just as Tim did in the demo, pull down a report for those things that are not compliant, ask for remediation of those things and do that all manually.

So absolutely you can use NautobotGPT to do it that way as long as you were willing to stay on top of it and and and know what you needed to check and when you needed to check it. To Tim’s point, we have a Golden Config app in our app ecosystem. It’s our most popular app in the ecosystem. It was actually designed to solve this configuration drift and configuration compliance challenge.

Anyway, if you’re if you’ve got that deployed, you have that capability already in the third or fourth agent model that I as I mentioned, we’re going to expand N NautobotGPT to be aware of the um app uh ecosystem applications and be able to provide guidance and support for using those apps uh and and extending their capabilities and making sure you’re using them completely and robustly. So there’ll be a place where you can start to use NautobotGPT to deploy those apps more effectively um in the near future.

Near future being uh this year or early next year um and then of course further down the road ideally there’ll be enough knowledge in the AI and enough experience with using the Golden Config app that the AI may now we’re talking mid to later 2026 be able to make those sort of recommendations to you. the golden config app can solve this for you. Would you like to right would you like me to develop a roadmap for deploying golden config and addressing this problem with you using this solution?

So there’s a large arc there of how you can solve this problem today, how you might be able to solve it with the golden config app uh tomorrow and how the AI might be able to help you solve it much further down the road. Um I know we’ve been getting a lot of questions in. There are others, please do fire them into the Q&A for us. We do have a few minutes left here that we could answer any further questions. Jeeoff, did you mention there was one that was of interest? Yeah, I’m I’m going to answer.

This is a handful of questions that came in and Chris is doing a great job answering these things uh u via the the Q&A uh feedback, but I do want to highlight a lot of the questions were around data security, data model protection, those sorts of things. Each NautobotGPT instance is a standalone instance for you. Nobody shares a NautobotGPT deployment.

So um this idea of cross-contaminating data um uh you know from one customer to another customer or from yourself to some generalized pool of information that the LLM is using is not the way GPT was set up. We fine-tune the LLM. We bring the LLM into our Nautobot instance. We deploy that unique instance for you. Uh right now it’s primarily deployed either via AWS um using the OpenAI model or it’s deployed using the AWS um LLM model.

Um so that’s kind of one of the first key ways we handle data protection and data integrity was to make sure that none of your data ever left your system. Um and that each as I said each individual NautobotGPT instance is owned wholly by you and used only by you and your data. So that’s one of the big ways we try to protect uh and ensure everybody has a uh uh a protected enclave of information that you can trust and be confident it is only going to be used by you.

Um, and just to dovetail off that, uh, Jeff, we actually got a question in that’s kind of related to that, which was, uh, would we be able to make Nautobot aware of our customizations or non-public apps? As a result of what Jeff Jeff just said, then yes, because we are making that specific to your uh, environment, then we can absolutely allow uh, GPT to have access to your customizations that you have within your Nautobot instances because it is specific to you. It isn’t shared across multiple clients. Exactly. There here’s a good question.

Um Tim, how many users can access NautobotGPT once you have a NautobotGPT account within a client? Uh to my knowledge, there would be no limit on that as far as the number of users. It’s going to be whatever you have licensed in your Nautobot cloud or Nautobot environment. So not a problem there as far as number of users. No, no limits. Yep. I think that’s a great takeaway there there as well is that when we give you your Nautobot instance, it’s for you and your company to use and we don’t put any restrictions on it.

As many users uh as need access to it, we allow you to maintain and manage that access, right? Uh amongst your users. Um great question here. I’m going to throw this one out to the group. We can all kind of maybe take turns at it. Since it’s specific to our environment, can it be self-hosted? I’ll start the answer here and then Tim, I’ll let you come in and clean up anything I don’t cover. Um the short answer is right now no. Right now we provide it only via SAS model.

Uh because of all of the additional work we do to fine-tune it and train it and maintain it. We do try to only provide it via SAS model. Tim, anything you want to add to that? Yeah, I I would just iterate that uh on that idea which is that it is something that we are going to maintain uh from a SAS model at this time because of the complexities of maintaining and operating an AI in an LLM. uh and we want to be able to make sure that we are keeping it up to date and improving it constantly for you.

Uh so it will be delivered SS model for this time for at at the moment. Fantastic. I think there was a question uh that I saw come through um earlier about uh you know c can we use different AI models and and do we do we consciously use different AI models for different capabilities within the uh within NautobotGPT. Uh Tim I’ll I’ll open that up for you. See if you’ve got a a thought or any guidance on that. Yeah, so these are things that we have on you know as we look to the future uh uh as we come out of the gate here.

We want to make sure that we give a good stable experience as we get started with NautobotGPT and of course we’re going to grow it. If you know anything about NTC um we are constantly improving everything we do. So absolutely that’s on the road map to provide that capability as we look to that future. Um and of course if it’s something that’s really critical to you as a business um we’re always open to discuss those types of ideas and something that we might be able to accelerate that delivery for you.

um if you’re super interested and that’s something that’s a requirement for you as a business and you want to partner with NTC to be able to deliver something like that. So of course we’re always open to that conversation for partnership as we look forward. Um and similarly there was another one coming in. Hey couldn’t we self-host if we have an M MCP server in the mix. We are looking at that adding that capability to NautobotGPT to include that idea of being able to do this A2A or MCP type con concepts.

So that is something as well that you’ll see as we further release Nautobot . Love the way you all are thinking of being able to use this as we look to the future. Um keeping in mind right this is our initial release and these are the things that we’re coming out with to ensure we have a good experience for everyone that starts to use NautobotGPT. Great. And I’ll just add to that just very quickly. We in the development of NautobotGPT, we used almost all of the major LLMs um as we developed uh our own solution.

And of course the the marketplace is changing so quickly, the technology is moving so fast um that whatever we thought was the best answer today turned out to be a different answer tomorrow. Um so we have experience with a broad array of the LLMs out there and we’ve made NautobotGPT work successfully with almost all of them. As I said before, right now we’re primarily focused on AP open AI uh and AWS bedrock. Um but we have used anthropic models as well.

Um we are um looking into the MCP uh service space as well as Tim mentioned primarily to interact with other agents who do things that uh that maybe we don’t want to do ourselves but we want to use somebody else’s capability somewhere else. I’ll give you an example. In one of our demos, um, we tied into an MCP server that gave us, um, uh, directions, and then we found an error on somebody’s network as part of the demo.

We then were able to take that location data, convert it into, uh, location uh, directions, and then pass that to a field engineer to so they knew how to get to the location that had the problem, to get to where they were going, so they could go do uh, you know, a field repair. There’s a great example of where we’re starting to use MCP servers to do interactions with other agent models that allow us to bring capabilities in that we don’t want to develop ourselves or don’t feel that we are best positioned to develop and obviously we’ll continue to expand our use of MC MCP servers to engage with other uh agents as well as make our own agents available eventually.

So it is someplace that we know our engineers are looking and will be someplace that we expect will will further develop as uh that MCP server community begins to grow and people begin having agents interact with other agents across uh across different platforms. Uh I’ve got a question here on AWS pricing and and demand and do we expect pricing to shift on this? The short answer is no.

As we said earlier, once you have access to NautobotGPT and we assign a a license to you, an account to you, it’s part of a a larger suite of of software that we uh we’ve enabled. We don’t expect price increases to happen based on usage. That’s why we open it up to as many uh users as you want to use it is we’re really trying to make it a tool that facilitates your use and your development of your network automations.

So, no, it’s not priced on per use or how many queries or how many CPU cycles you use or whatever uh pricing model uh that the underlying LLM uses. We’re we’re treating it as a holistic solution that is going to be part of a larger software suite that you’re going to be deploying. Anyway, uh let’s see here. Thanks for asking. Quick follow-up question to the MCP idea was to keep your IP inside NTC uh and combine local data with it.

therefore leveraging your knowledge base um that could put it behind a p a pay wall for premium support package or by token count. Uh no, that’s not something we’re going to do. Again, the short answer is anytime we would open up to the MPC MCP server uh to access additional agents, we would expect those agents would be providing value to you. Uh and and we would not expect any pay wall situation to prevent access to or from that data uh if it was going to support your automation efforts. I hope I answered your question correctly.

If I didn’t, uh, feel free to ask it again and we’ll try to come back and get you a better answer. Uh, I’m just looking through here. Anything else? And look at the open questions here. I don’t see anything, Tim. Anything you see on the Q&A that that you think uh, we want a great job getting answers out there. Uh, again, if we didn’t get your question answered here today, we’re always happy to have a chat with you. Um, you can reach out to us. We’re absolutely willing to have those conversations and get you those detailed answers.

So, please uh you know, feel free absolutely to to set up some time with us. Uh there was one here that Chris answered about NautobotGPT mascot. Uh I love the idea. Uh so, you know, maybe perhaps we’ll do a contest, a community contest, and we’ll get some ideas in from from the community here for what NautobotGPT mascot might look like. But yeah, really cool idea. Um and the folks on our marketing team, they’re always coming up with great ideas for this kind of participation. So, thanks for that. Love it.

Uh love the uh the interaction with us there. Yeah. And I’ll just add I’ll just add one more thought to that Tim as we as you go on to the next slide just uh highlight to Tim’s point you know lots of great questions here as you you know leave this webinar you may have deeper thoughts that come back to you uh tomorrow or the next day or next week. The uh the key takeaway here from our side is you hopefully know how to reach us. Hopefully you’re engaging with with network to code quite regularly.

We’d love to have those follow-on conversations with you if there was a question you didn’t get completely answered or there’s another component uh to the question you want to, you know, take the next step to. Uh we always appreciate folks reaching out to us. Uh bringing your network automation challenges to us and working with you to help find solutions and help move you guys forward. We all want to see network automation be successful. Um it’s really why we exist. And so if you do have any questions whatsoever, please do reach back out to us.

Would love to engage and find a way to help you guys be successful. Tim, yeah. Absolutely. So, just a quick recap. We went through those challenges. We introduced NautobotGPT for you. We showed you what those capabilities were like. We did a live demo and of course we got to interact with the AI and you got to see how sometimes it can be inconsistent, right? And that is the nature of working with AI models. We have to get used to that and get better at our prompt engineering. And then a little bit about what’s next and of course the Q&A.

Uh so, thanks everyone for attending. Please do reach out to us for a demo. We’re dropping links to all things Nautobot and NautobotGPT for you into the chat and of course there’s some links here for you on the screen. Uh thanks so much for attending the webinar, making it one of our most popular, if not the most popular webinar we’ve ever done. So that’s amazing. Thank you very much. We look forward to seeing you on the next one.