As AI platforms become real product interfaces, SaaS companies are under pressure to open up secure agent entry points without rewriting their stack. In this session, Frontegg CEO and Co-Founder Sagi Rodin shows how AgentLink lets you expose existing REST or GraphQL APIs to ChatGPT, Claude, and Gemini in minutes.
The demo walks through spinning up a hosted MCP server, mapping APIs into agent ready tools, connecting identity, and executing live reads and writes across multiple SaaS systems from within AI platforms. Guests from Cisco share what this unlocks and why natural language driven workflows are becoming the new default.
Learn more about AgentLink:
Super excited to be here taking this livestream for our launch of AgentLink here from the lovely Mountain View, California. And as you’ll see in a few minutes, we’re taking it global, so we’ll have guests appearing from the other side of the world as well.
Super excited. We will talk about AgentLink, the new exciting product from Frontegg, and I will actually demo the first part of of our product, the connector, the agent connector. That sounds exciting, and I’m sure, you know, it you’ll love it. You’ll love it. We have some amazing stuff coming up, and we have three days of that. So every day, we’ll introduce something that is unique and interesting and should provide a lot of value to SaaS organizations.
And will also intro Oded from Cisco, and and we’ll have some audience Q&A. And finally, kinda the cherry on top, Aviad, my cofounder, my partner in crime for the last ten years we’ve been working together, will join me in talking a bit about the product road map coming ahead for the AgentLink and the Agent Connector that we’re introducing today.
So great to see everybody joining us. We have quite a lot of people here. And first, Oded, welcome, and thanks so much for joining me here.
Thank you. I’m excited to be here.
Yeah. And, you know, before we kinda start and I’ll introduce AgentLink, you know, I thought about kinda an ice breaking question. So maybe something that is not AI related that is happening, and it’s fun.
Yeah. So just have three kids. My youngest one is a my daughter, she’s the youngest. She’s seven. And last year, she had a robotics class where she needed to use the microbit, and we have two microbits at home. And yesterday, I offered to her if we can sit together and build something cool with microbits, and she accepted.
And we built a hot and cold game with two microbits that were transmitting with radio frequencies. And she was super excited, and she dragged the blocks. And she she wanted she understood the flow, and she wanted to go far and see all the range of the radio frequencies. And it was it was really fun. So kids still like coding.
Kids still like coding. And, you know, what’s fun about that story is that you could have done the same thing probably, like, you know, twenty years ago, and Yep. It would still apply.
So I find it super cool that you can still make things work in the old way, and and the kids are still connected to it. Right? Because, you know, the the new stuff, that’s that’s that’s, you know, that’s I’m I’m I’m worried about about some of the the the new innovation that is happening and how it’s gonna affect the the the kids. One of the nice, you know, interesting conversations that I’ve been having is whether kids, you know, would still learn coding or they will just write prompts and stuff like that. So, yeah, so thank you for that story. That’s a that’s a great opener.
And we will talk to you a bit about stuff that are AI related in in a few moments, but let’s start by showing a bit on what we’re doing here today and and kinda present I promised my marketing guy not a lot of slides, so I promise there won’t be a lot of slides here.
But let’s introduce Frontegg AgentLink, and we’re securely opening up your products to Gemini, to change p t, to clothe, any AI agent. So this is actually what is happening here.
A bit about the company quickly. So Frontegg, as you know, exists for many years now, six years. We’re start stopping to count soon. Hundreds of customers, great investors behind us.
You know, we we’re super proud to have customers that are fast growing startups, but also enterprises, and they’re already using the product. They’re super excited about this release as well.
And to tell you a bit about the story, so, you know, up until six months ago, everything that our customers were talking to with us were about securing their web interfaces, their mobile apps, their APIs. But lately, it’s all about listen, Sagi. How can you help us secure this agentic interface that we’re starting to open up? We introduced the some chatbot or Copilot within the application, and we’re starting to see some messy stuff that are happening with it.
And our security team is not happy, and our product team is not happy. And how can you help us with that? How can we track the users that are coming in, what they can or cannot do within those application? So it seems that in the last few months, there’s a lot of conversations that are happening with our existing customers and also with some prospects that are coming in to listen about our identity product, and the discussion is about AI agents.
And I think that that’s kind of the story. And if I need to take this story, rewind back, you know, about and I think it’s maybe more on the macro level for the move and the shift that the evolution that is happening in product interfaces. So, you know, we started with terminals and mainframes and d b two and all that fun stuff in the sixties.
And and then, you know with CD ROMs and installing software on their computers with a license key. And in the early two thousand, there was a major shift to the cloud. So all of the software started to move to the cloud, and then mobile interfaces were introduced. So a lot of, you know, software companies, SaaS companies were basically shifting or adapting their existing interfaces to mobile interfaces.
And then we have, obviously, APIs. So in the last few years, a major shift to support APIs, so kinda APIs and automation machine to machine became a standard. And the exciting thing that is happening now, and that’s a big shift, is the AI operating systems are starting to get into our workspaces, our workplaces as well.
And I call it the new way to SaaS. Right?
So shifting from logging in to dashboards, counting on user experiences that were predefined for each and one of the products to moving to kind of those described and done experiences where the user is connected to the data, and, basically, they can create their own experience and get their own value for for whatever they need to solve at that moment.
And the fun stuff is that over the last few months, we see that a lie there’s a huge alignment happening from all of the AI platforms that we use at home, shifting from home usage to using workplace AI platforms. Claude added their own connectors. We will see that on my demo in a few minutes. OpenAI followed with shared GPT adding their own ways to connect applications from their marketplace or from custom model context protocol, MCP servers.
Gemini CLI supports it as as well. Microsoft has a big studio that you can connect a lot of the the applications to that studio. So a lot of movement in the industry that is happening in those areas.
And and I think that, you know, this is this is super cool. This is the vision. This is the new way to work.
You know, it’s supported by the analysts. So Gartner says that forty percent of enterprises will have agent specific tasks performed in the organizations by twenty twenty six. Right? So by 2025, only five percent of organizations have actually deployed agent activity for the business operation within their organization. We were talking about forty percent by 2026. So that’s actually happening now.
And and I’m I’m sure that it will, you know, it will rise throughout the next few years to huge percentages. There will be almost no organization that will not have AI in it. And this is where we introduced Frontegg AgentLink. So that’s you know, I’m I’m just thrilled to show you that product today.
We’ve been working really hard over the net last few months on working with customers, with initial design partners, and, you know, Oded will talk about it in in a few minutes. But I think that the product that you will see actually shows that a lot of value from day one. We want to make a seamless experience for our users. AgentLink allows any SaaS product to be opened up securely to any AI platform.
So you can be listed in any AI marketplace, ChatGPT, Gemini, Claude.
Any platform that accepts custom MCP connectors, you can be listed in a few minutes. And it’s actually a few minutes. I will show it soon.
So the principles that we try to work with during the the building of the product, first of all, use your existing APIs. Right? We don’t expect you to rewrite everything. We just want to leverage your existing APIs and start to shift them into AI powered tools very fast.
Second, any identity provider is supported. Right? So if you use Frontegg, that’s great. You get it kind of batteries included with with all the fun features that Frontegg has as an identity provider, but you can use any identity provider that is OpenID compliant as you’ll see in a minute.
Any type of back end is supported. So regardless if you have REST APIs, for example, Cisco had GraphQL, everything is supported, Lambdas or any type of back end.
We don’t want you to reinvent the wheel. We want you to continue using your existing stack. And the uptime is enterprise grade uptime. You know, that will become one of your main product interfaces. So it’s not a bonus.
It’s, you know, it’s not a cherry on top. It’s it will become actually one of the main interfaces through which your customers will connect and use your products. So it’s important that it will be always up and running, high security, and, you know, all the things that you expect from a critical infrastructure platform.
On the high level architecture, and I think that this slide and I’m sorry if it’s too small, but it shows how your customers from the left side will connect through their own favorite AI platform going through the Frontegg AgentLink product, and you will see that in a second, and connecting to your existing back ends when everything is audited, secured.
You can set up any policy. You have all the visibility, all the governance that you need, but not only through ChatGPT, Gemini Cloud. You can also get all of the data flowing through custom agents running through that same foundation. So whether it’s your internal agents, Copilot that you built in your app, or custom agents that your customers build, everything goes through the same interface and protected on the same back end configuration and guardrails.
Okay. So, finally, we will go and present the demo, and we will show you in the demo how I select an existing product, existing application with REST API, how we generate the connector, and then how we actually connect agent AI platforms to this application and make some fun stuff with it.
So that’s you know, it’s a I’ve people told me it’s crazy to do something like that with live and not recorded video, but we’re actually going to try and do it live. And, hopefully, everything works because there’s a lot of moving pieces here, but but I’m confident that that it will work. If not, I will blame my marketing guy.
So let’s start.
Okay. So several interesting things here, and we’re going to start with this application that we have here.
It’s an expense application, just a regular, you know, your b to b expense application, whether you care file expenses.
You can list the expenses. You can upload your receipts and all the regular stuff. And now what we’re claiming is that, you know, in a year from now, you will just upload everything. You will you you will query this type of applications, create reports from your chat interface that that will be used inside your organization. So let’s see how we connect, how we open up this application to AI platforms.
In this demo, we’re demonstrating how it can be used not with Frontegg as IDP. It’s important to for us to pass this message, but, obviously, if you’re using Frontegg as your IDP, you you get everything. Batteries included.
But any OpenID compliant identity provider is fine, and it will work fast. And here, we’ll show you something that works without Xero. You can use Keycloak. You can use Amazon Cognito or anything that you have that is protecting the users in your app. So next step that we will do is we’ll go and sign up to agent link. So we’re going to frontegg dot com, and I will sign up.
Okay.
Let’s complete the sign up with the code that we received.
And I’m getting into our Frontegg portal. And as you will see in our Frontegg portal, the experience that you’re getting to the new application when we’re making you know, the the purpose here is to make your application agent ready. Obviously, you can do all the great stuff that Frontegg allows on other fronts. We get an AI native experience.
So it was very important for us to build an experience that is truly AI native just like we used to get kinda on on our day to day work. And and so this is an Frontegg agent link AI assistant, and it will guide me through the onboarding process. And it asks me about my application. So what kind of application I have.
So I’ll just write that I have an expenses app.
And what happens now behind the scene is it actually spins up an enterprise grade MCP server, and that’s it. It’s basically ready. Now we want you to be able to test it out before you actually connect your real APIs, and this is why we created the portal for test environment. And you can see here that it’s a full test environment.
It can work with any model. Right now, it works with GPT four, but all the models will be supported. It’s kinda give you a taste and a feeling of how it will be used by your end users without actually connecting it to to Clotch, GPT, or other AI platforms. So you can kinda get the sense of it.
It gives you a few templates of what is supported. So in this case, let’s ask for a list of my expenses, and it will produce an answer. This is a mock answer. So, basically, it mocks our APIs just by the by the type of application that I provided.
Nothing really is connected here, but you can kind of get the sense of how it will feel after we connect the real backend to it.
Okay. So this is the request that was made, and I can make a I can ask other questions and and play with it. Super cool. But let’s maybe connect and, you know, turn to the real stuff.
So now the next step is to basically connect a real back end. And for that, I will open up the interface of the schema upload. We can use REST APIs. We can use GraphQL, connect any type of back end behind it.
You can upload the file. You can paste the code. Here is a small example, but I’ll just go ahead and paste my open API spec.
And it’s a standard open API spec. You can generate that from your back end APIs.
It shows the different objects that are used on our back end, and it shows all the endpoints that we have here. So list on expenses, great expense. We will upload it, and AgentLink analyzes that open API spec file and and proposes some tools that we can expose. So as you know, AI and agents understand tools. They don’t go straight to APIs, and these are the tools that it suggested to create. So I created I chose five tools here.
I’m saving it, and that’s it. Basically, we already have those tools inside. And now the next step, it asks for what is the authentication provider. So what is the identity provider we have? As I mentioned, you can use any OpenID compliant identity provider. Everything is supported. At this example, we’re using alt zero, so I’ll pick alt zero.
And next step, if you’re not using Frontegg, you need to do an additional step of generating the trust between the IDP and Frontegg. And for that, we need the client ID, client secret, and well known URL. You can export it from the back office of your identity provider. So I’ll just paste the well known URL, the client ID, and the secret.
Don’t try to hack it. There’s nothing fancy here behind it. It’s just a demo application.
And now it’s configured.
And the next step that it asks, and that’s kind of the final step, is actually how to connect to our API. So what is the URL where our APIs reside? And this is hosted on Vercel right now, the demo application. Cool.
So we see that it set up the API base URL, and that’s it. We’re ready. So we’re going to the dashboard. The MCP is up and running, and I’m going to the dashboard.
And we can see here there’s no data yet. Everything is configured. And, also, we can see that we have the assistant, the AI assistant working with us on the left side so we can perform all of the activities, all of the configurations, fun stuff that we will see tomorrow on the second day release.
And we can configure everything from the left side, but also if we still want the point and click traditional experience, it’s available on the right side as well. So we try to create this hybrid approach of user experience. So awesome. Everything is set up here.
We have everything up and running. And I think that what we’ll do on the next step is we’ll go and try to connect that to one of the AI platforms. So this is the this is the the real the real thing that is happening. We’ll just copy paste the MCP gateway URL. So this is the live URL where everything is hosted.
And I will move to let’s start with JetGPT.
So I’m in ChatGPT currently working in developer mode.
It will be opened up for all of the modes soon, and I’m creating a new MCP. Oh, let’s pick some icon for our expense app.
The name is expense application. I’m pasting the live MCP URL, and I’m basically creating my MCP. Okay. So let’s see how that works.
The first thing that has happened okay. So we see that connected. That’s kinda the end user experience when they connect the first time. They will need to log in with their user.
Obviously, Frontegg as a gateway needs to understand who is the user that’s connected, that is trying to make the calls in order to determine which kind of tools they can use, which kind of guardrails and policies we have configured on those type of users. RBAC is fully supported, so it’s important to know who’s the user. And I will log in as the administrator to do this first connection of the MCP.
And after I’m connected, basically, I can see that it was successful because I already can see a list of tools.
So I can create expense, delete expense, you know, all the five tools that we created on our gateway, and and that’s it. So now let’s choose auto so it will be quick, and let’s we we’re using our expense application that is connected, and let’s see list all my expenses. So that’s, you know, that’s super exciting. That’s the first query we’re doing from our AI interface.
So it’s trying to find the tool, the relevant tool to fulfill this request. It asks me whether I want to allow using the tool. I allowed it, and we should get an empty list now because there’s no expenses. Perfect.
If we go to our expense application, I will log in to expense app. We see that we indeed have no tools right now, so that’s that’s just perfect. And now let’s do the second connectivity, and that is to Claude.
So let’s do the same thing that we did with Chat GPT. We already connected there, and we can make queries.
But now let’s connect to Cloud. So as you can see in Cloud, I already have some applications connected. My Google Drive, my Google Calendar, my Notion is connected, and now we’re going to add our expense application.
Okay. So, basically, we’re doing the same thing here.
We configure the app. The next step that we will take is connecting the application in the same way the first time we connected it. So the administrator will do it for your customers the first time with the MCP that you’ve opened up. And perfect.
We can see that it connected. We can see the configuration. We can see the five tools listed in cloud as well. We will allow to use all of them.
Again, that’s a onetime configuration that your admin will perform.
And when we’re going to tools here, we can already see that your expense application is listed along with Notion, Google Workspace stuff, you know, any application that you connect to it. Let’s remember that in a year from now, all of the apps of your customers will be connected here, and and you can be listed here as well through this product that we’re releasing today. And, you know, let’s see that it actually works. So now I will do something. I will do a write action, not a read action. So let’s add an expense for team dinner from yesterday for four hundred dollars.
Okay. So now we’re trying to actually write something, perform an action on our expense application through our account, and it creates the expense.
And done. Added the team dinner expense. So we should be good, but let’s verify.
We’ll verify through going to Chat GPT and asking it to list our expenses again.
And now we’re expecting to see the the new expense here.
Asking it to, list all the expenses again. Yeah.
It called the tool.
And let’s see that we indeed get this one expense for the team dinner. Perfect. So we see this one expense. And if I’ll go into our UI, I should be able to see that over the UI as well.
So perfect. So we actually created an action from our cloud. We can do the same. All of the tools are usable.
We can edit everything. So any tool and any configuration and authorization and entitlement that I have as a user to to use with my expense application, I can now do through the chat interface.
And we did all of that in a few minutes, but this is calling one application. One of the big values that we have here is not only using one application, but AI platforms that are connected to many products will allow us to do these integrated activities throughout all of these products that are connected to our workspace. So I actually want to try and do something that is more complicated. Right? So I have an expense equipment file here. Okay? So that’s basically a report from October 25, so last month, on things that I purchased.
And another thing that that I have is I have my calendar. And from last week, I had a few lunch and dinner and stuff like that. So I actually want to take the dinners because they’re the most expensive, and let’s try to ask Claude to do something more complicated. So I will ask you to read the file with the equipment expenses and also get the dinners from calendar.
And you know what? Let’s add everything to our expense application now.
But not only that. After it finished, I wanted to generate a report, a new report on our Notion on our company Notion so my CFO can can have access to everything that was done here in my department.
Okay. So that’s you know, I wanna read the file with the equipment that was purchased in my department. I want to take the dinners from you know, that we had from last week, and I want to put everything to our expense application. I think we had only one dinner, if I’m not mistaken.
And I want to put a final report of everything into a Notion document. So let’s see how that plays out. So what happens here is that Clot starts to work on it. It it creates the plan, and, immediately, it goes to Google Drive to to search for equipment expenses.
And I see that it already found the files, so it’s taken the file. Now it’s trying to fetch the dinner events on the calendar. Great. It found one dinner as expected.
And now it goes and creates the expenses for the equipment, for the dinner.
So we have several expenses that are created here.
And, finally, after finishing this task okay. Now it takes the team dinner from calendar.
One dinner. See, a few expenses that were added.
And now it creates the page on Notion. So let’s see after it finishes.
We can see here the title and the content that it tries to put on on Notion.
And perfect.
Completed the task, and it summarizes the action.
Everything, obviously, here will be audited on your organizational Cloud account, and we have the full report on a new Notion file. So full expense report with everything, with the equipment, with the team dinners that we have here, and the the whole summary in Notion. So I think that this is a great way to demonstrate how the new age of work is going to happen without logging in. So we didn’t have to go out to four places, retrieve the data, put new stuff there.
Everything is happening through this single platform, and and Frontegg is there to make sure that everything is goes smoothly. Expense application is added, all of the new expenses here for the team dinner and for the equipment, and everything is smooth. So so that’s kinda the demonstration of the new age of of workspace operation with SaaS applications, and we’re so excited to power that up. If we go back to the Frontegg dashboard, we can see that everything obviously is presented.
We can see which kind of platforms are used. So what is my top used tool?
We can work with the AI assistant and ask ask it for some analytics and insights. We can create policy. We can see that it answered with the with the correct tool for creating new expense that were used five times, and everything is also monitored. So you can see here that there’s a log for every activity and every execution, and you can really get all the data here. Even if something didn’t work, the developers can troubleshoot it, and this make sure that everything runs tight. So so this is this is our demo, and, you know, things went pretty smooth. So I’m I’m excited.
And let’s welcome back Oded and and chat a bit. Oded, great to isn’t it exciting?
It is. It was an awesome demo.
Thank you. Thank you.
Odette, I would love to, you know, talk to you a bit. You know, you’re kinda one of the first adopters of of AgentLink. And maybe to dive a bit on the before we’re diving into the things, you know, tell us a bit about, what you’re doing, why even, you know, considering opening up, the product that you’re working on to to agents and to AI interfaces, why it’s interesting for you. So give us kind of the the overview.
Okay. So I work in Cisco Identity Intelligence. It’s a product that gathers identity information from your identity providers and service providers and provides a single pane of glass of all the identities, human and nonhuman, in your organization.
And we collect a lot of information about users, groups, devices from different service providers and identity providers.
And we we let our customers slice and view and filter different aspects of that data.
We want to use agents because I think agents is is the future. It democratizes data. I mean, if my customer wants to see the Salesforce administrators that have signed in from a location that is different from their registered work location, they can do that today in our dashboard. But they need to manually go to the users table and filter by the Salesforce administrators.
And then for each one of them, go into their activities and filter with that cumbersome query language to find locations where the registered location where the sign in location is different from the registered working location, and it’s it’s just a hassle. And doing that through natural language questions and performing queries by natural language and letting the MCP client control the workflow, it’s it’s the future. It’s a lot easier than what I just described doing manually.
I agree. You know, I think that before up until now, we’re kinda relying on somebody to guess what is the the best experience for us. Right? So we have, you know, very talented people, product managers mainly that are working very hard to talk to customers, gather some information, then kinda find the the the least common denominator, right, within the product to to try and suit the user experience for everybody.
And what you’re basically saying here is that, you know, one user would want to retrieve the data in this way. One other the other user would want to do something else. Once you’re giving them the ability to work with the data, you’re kinda saying, listen. Everything is opened up, democratized as you said, and I can just go ahead and build my own experiences.
I can ask my own questions. I can ask the agent to do some recurring tasks, like, every week if I wanted, ask it to paste the results on Slack or whatever I want to get them, and I’m not relying so much on these, you know, predefined user experiences within the SaaS applications.
Yeah. Exactly. And even if we do get the the UI right, the user experience right, you need to train your customers and their administrators to use that specific UI and user experience. And and when you change that and you change their experience, you need to retrain them. And I think using natural language is definitely easier.
Awesome. Sounds great. And, you know, when you kinda first heard about AgentLink, what made you try it? You know, you have a team, that that built some amazing product.
And, obviously, there’s a build versus buy question here. Right? Why would we want to build it on ourselves, or why would we want to kinda delegate that that infrastructure to somebody else like AgentLink or other solutions out there? What were your first thoughts about that?
So it it it was a journey. We started off with, trying to build this ourselves in Slack. We added slash commands in Slack, and then we had to write the entire infrastructure of translating the user questions into API calls.
And it it it’s very clunky and cumbersome and time consuming. And we found that we didn’t want to do this ourselves. And we do want users to use Slack. We we would love it if we could do this in Slack, but we saw this emerging technology in MCP, which allows the users to do this in different IDs and MCP clients.
And we said, yeah. But this is what we want. This this is the bridge between the user questions and our APIs. And we wanted to use MCP servers, and we built an MCP server using the STD IO transport.
And it works, and it’s fine. But I think I feel that the STD IO MCP transport is geared towards developers and and environments because you need node installed or Python or UV or containers or you need some kind of of of execution environments.
Yep.
And we and and that’s a a big hurdle for for nontechnical people to get past. So we wanted something and and HTTP Streamable came up, and we wanted to use that.
And it’s all new. I mean, there is not a lot of we you didn’t have a lot of competition when when we started working with, AgentLink.
We did look at the the what, big cloud providers try to build now, and it’s it’s cumbersome. And we didn’t get exactly what we wanted. And we tried agent link, and it was super easy and super fast, and it and it just works.
Yeah. We try to make it you know, we could have taken an approach that is very wide scope like most of other solutions. There are great solutions out there that allow you to do a lot of stuff.
But talking to our existing customers from our customer identity solution, they had a lot of tasks in mind, but one thing that we caught is that all of them want to try and open their products even you know? And I’m not not talking on opening everything from day one. Right? But even OneFlow opening it up to to AI usage, and we build the product centered around this value. So we didn’t try to capture everything, but we try to make sure that this actually works perfectly and and very fast. And and we try to support also every type of back end that you know, your back end is mostly GraphQL. Right?
How, you know, how did the connector handle it? Did it pick up the schema automatically? Was it because GraphQL is more complicated than what I showed with the REST API. Right? That’s defined.
GraphQL is a great way to define your interfaces and and APIs.
And at it’s it’s very similar to what you’ve shown in the demo. On screen, you had a a tab for GraphQL. So if you click on GraphQL, you upload your schema file, and agent link parses the file and suggests all the GraphQL queries and mutations.
And we were the and it was super easy using GraphQL and not REST.
Yep. Yeah. That’s taking an advanced one advanced technology and connecting it to another advanced technology.
Looking ahead now that you have, you know, your APIs can talk to agents, what kind of, you know, user experience do you think that it unlocks for you? What is the, you know, number one experience that you think will be used through chat chat GPT or Clot or whatever your customers use?
So I have two types of customers. Some do more threat detection. And in that case, they can ask the the ask in in cloud or their or or whatever they use, who are my riskiest users, and they can get an instant response about who are the users or nonhumanite, users, that they have in their organization and drill down to each of them to find out why they’re considered risky. And we have, compliance and posture type of customers want that can ask questions like, give me the list of the apps with the most unused users or send create a report with the users that are assigned to use Entra but are not using it.
And and then we can provide that information really quick, and they can through the MCP client, they can generate the report or send that to a different MCP server to visualize the information differently, which is also a key benefit.
Yeah. For sure. Right now, it’s kinda the chat you know, the the chat interface, which is something that is boring. But, obviously, that will evolve, and we will get more ex kinda product oriented tailored experiences on UI as well. So stay tuned. There’s a lot of stuff that will evolve along the next weeks and months with the product.
Oded, it’s it’s great to have you, you know, working with the product.
It’s a it’s a, you know, a pleasure for us to to support Cisco in in this endeavor, and great to see that, you know, organization big organizations are taking an innovative approach and trying to work before the market and not kinda chase after the market. And I think that this is what you guys are doing, and it’s it’s great, you know, to see from the side, but also great to to work with you on that.
Yep. I got lost for a second. Yeah. Thank you.
Thank you so much for being here, Oded, and you can you can stay as we will work with we have some questions. Okay. So let’s yeah. You have we have some questions from the audience.
Do you wanna read this out loud?
Yeah. So the first question, hey, Oded. With an organization like Cisco, you all have a huge API footprint. Do you see AgentLink as a driver to uncover new ways for Cisco to partner or your partner network to drive growth and new opportunities? So, basically, is it opening up business opportunities for for the for for your product?
I think Cisco the the Cisco security business group does have a huge API footprint, but in this case, we’re integrating one of the products from the entire suite. So it’s Cisco Identity Intelligence.
And I think that using tools like agent link, you can link different products from the Cisco suite and then ask the MCP client if this is a risky user, then tell Duo, which is the Cisco MFA, to do step up authentication on the next activity. So we can also drive policy across your Cisco suite products using tools like, like an MCP server.
Yeah. Everything is connected. And, and as I showed, you know, you can ask, for task. And let’s remember that, you know, these activities are usually recurring. Right? The the one that I showed today, for example, it’s something that the user would typically you know, a CISO or a security analyst would want to do every week or get a report when they, you know, start their work week on Monday.
And we can just set up an agent to do this every week. They don’t have to remember or type it every time. So so I I completely agree.
Any other questions that we have? Let’s see. Okay, Odette. What kind of risks are you worried about when letting agents into your product?
So letting agents into my product is a lot like, letting it it’s it’s like it’s invoking APIs. So it’s a lot like humans invoking APIs or service accounts invoking APIs.
The thing that I am worried about is that I’m not really worried about it, but but I noticed that NCP clients tend to generate more, operations and queries than than different Right.
Than user than human users because it asks the question.
And if it it isn’t happy with the response that we got, it It will try again.
Ask the same question again or slightly different thing. So the number of of invocations of APIs might be bigger. So that’s one of the things I I worry about.
Yeah. And we get a lot of those embracing, you know, these new interfaces. You know, in security wise, it’s it’s it’s pretty pretty challenging, and this is why we have day two tomorrow. So so stay tuned. We’ll have some some fun stuff to to show tomorrow on how we can actually reduce some of those risks.
There there is one more risk that that we did did try to mitigate.
We tried so we’re using user user delegated permissions. You have two models of of operations. Either a user signs in or use API keys for a machine to machine delegated permissions.
So we opted to use use of delegated permissions, and then we can still use role based access and minimize the risk of unauthorized users performing different tasks.
Yeah. Perfect. Because, you know, those endpoints, the APIs that I connected here, some of them already come predefined and protected by some controls, some entitlement authorization controls, RBAC, and other stuff.
So in order to allow you to continue using the same protection, we we made some some stuff there, which you guys already use, and I would love to, you know, show more of that tomorrow. I don’t wanna ruin the surprise.
But but definitely definitely something that we hear a lot on how I can actually enforce my existing policies on this interface as well. So so for sure. One more quick question.
And okay. So this is for me, actually. So if someone is skeptical, what is an experiment they can run on agent link this week to validate this is worth investing in?
You know, I chose to do a live demo to actually show you how that works. So that’s actually a few minutes to connect, and and it works.
So the experiment that I would do is I would sign up. I would connect just one API, one flow to to agent link and try to see how that works through the AI platforms and and see if you like this this experience. So we have it free for until the end of the year. You can sign up and use it for free. So that’s that’s a great benefit. And and I think that, you know, it’s worth worth testing out and see if it opens up some interesting opportunities for your business, for your SaaS application, for your customers.
Okay. Perfect. So, Oded, thank you so much for for joining me.
It was a pleasure and very insightful, and hope to continue, you know, providing value to to you guys and and have you supporting us.
So thank you so much.
Thank you.
And now we will have my cofounder, Aviad, joining from our office in Tel Aviv. Hey. Aviad, welcome.
I know that it’s it’s late over there, but, you know It’s never too late for such an exciting demo.
So yeah. Thank thank you. Thank you. So, you know, Aviad and the team back in Israel worked super, super hard, so this demo goes as smooth as as it did. A lot of hard work over the last few months onboarding, you know, the first customers to it. We have some amazing customers such as Cisco already onboarded and using the product, so it actually works in production.
And, Aviad, I think that the the kinda interesting question that people are asking and, again, let’s not ruin the surprises kinda for tomorrow because we still have things that are already released, obviously. But but on the connector side, let’s talk a bit about what is coming next and how we think about those things.
Cool. So, yeah, we we spent, like, you know, the last couple of months, you know, gathering feedback from customers such as Cisco with the dead and and a lot of and a lot of exciting, you know, design partners on building out the agent connector.
And and when we, you know, we moved along on building out the solution that you just demonstrated live, we kinda understood from from other other design partners that, you know, that’s pretty cool that we build, like, a complete, you know, API to MCP connector.
But but some of our design partners are actually looking to gather some of the APIs and to gather logic jointly into a code based.
So that’s part of our, you know, road map, building out code based tools.
Actually, our, you know, our partners will be able to code their ways, our own tools on the Frontegg platform, and they’ll be able to actually code everything around what a tool needs to do, maybe orchestrate several APIs, make maybe orchestrate several several if else logics, but everything will be resides on Frontegg and we’ll be able to connect to your general CICD flows.
That’s that’s that’s super exciting. We’re talking about you know, you kinda talking about it from creating a tool. But if we think five years from now, we might get a different way of building back ends. Right? We’re talking about democratizing, Goddard said, democratizing data.
So this democratization will go through giving you full access to to the data. Obviously, it should be guardrailed and protected.
But once you get that and you can really trust that, the the opportunity is endless.
Well, hundred percent. Yeah. Yeah. I mean, we are kinda talking about a new way of producing software, which, you know, didn’t exist even six months ago.
So so, yeah, that’s that’s a complete revolution of of of the way that we see software being developed.
Great. Next thing we see here is remote MCP tools. So tell us about that a bit.
Yeah. So it was funny because, you know, we we went ahead with with a concept of getting your APIs and being able to kinda build in an MCP layer around the APIs.
And there were several layers that was already there, okay, such as existing MCP servers that were already there and needed that enterprise grade layer on top of it, which is what Frontegg does best.
So we we were able rather than getting your APIs and wrapping wrapping it around with MCP, Getting your existing MCP and your existing, you know, HTTP based MCP, Streamable MCP, and wrapping it around with our enterprise level guardrails, etcetera, and just, you know, making sure that you are able to connect everything that you already built without you needing to rebuild everything on top of Frontegg.
I think that if we when we talk to enterprises, you know, there’s a big need for, you know, for having this freedom to to deploy it anywhere. And Right. There’s, yeah, there’s a big kinda ask around data and where does it go and whether it’s used by models and stuff like that. So super important to give that. Hooks, that’s the thing that we you know, when you’re building a critical infrastructure, and I think that it’s a journey that Aviad and I know, we can write a book about the importance of of of Hooks.
You know, you pick an approach which is quite opinionated, right, to how this should work and how this should be used, but then you also want to give the freedom. So you’re trying to find the sweet spot.
Tell us a bit about the the the Hooks capability.
Yeah. So it’s funny because, you know, we we we both of us, we have been in the business for six years now, and and we know the developers like to have the flexibility of everything around their logic and everything around the flows.
And hooks actually met us building out TIM, and now it met us building out agent link because we kinda understood that the way of proxying flows and the way of proxying, you know, transform request and and even adding headers that the the remote servers kinda expecting to get is is is kinda crucial with the way that proxies work.
So Hooks, you know, we we are so fortunate to use, you know, everything with the infrastructure that we built for our customer I’m product and leverage that for for the agent link product Because the ability for our customers now to to add anything with regards of additional headers and even transforming API requests and transforming bodies of request when they are proxy into their endpoints is crucial for them to kinda keep the way that their SaaS was connecting to their APIs, and now the same works for for for their MCP gateways.
Yeah. That’s giving the developers the freedom. You’re talking about AB testing here, and I think that once you introduce a new interface, you want to measure things. At the end of the day, that’s the live business of our customers.
It’s not a side hassle. Right? And and you wanna see, you know, what’s the best experience for your users, which tool works better, especially with the nondeterministic way that the AI platforms are going to use your application. Right?
Because when you build the UI the UX the traditional UX know, UIs, you basically you could control everything.
But now the model is making their own way to use your app and their own tools, as we saw, that they think are most appropriate are being called, and we really need the way to to measure the success of those.
So definitely, Hooks will allow that. Let’s talk about orchestration, another way to maybe, you know, enable that, but on on steroids. Right?
Yeah. Orchestration is really, like, you know, building out agent flows on steroids. You know, everything that we talked, you know, code based tools, remote MCP tools, hooks is purely developers. But then when we spoke with product managers, they said, listen. You know, I I really dig on everything that you just said, but I don’t wanna get the developer into the loop. They have so many tasks to complete.
And then, you know, they they already built my APIs. And when I I wanna introduce a new tool, the APIs are already there. I just wanna orchestrate them. I wanna orchestrate them. I wanna, you know, maybe introduce an external API.
And I I don’t really need an a you know, my developer to build a new API into that to protect in this new API because this API is gonna be, you know, consumed only by this AI interface that we just talked about.
But then it it was kind of a challenge because, you know, we were kinda used to having developers build everything. And then when product managers kinda came into that and they said, I’m not sure I’d you know, I need to bother a developer for it. You know? Right. And then when we talk about orchestration, they kinda expected to have everything that we built up until now with Frontegg, you know, have them drag and drop, configure and everything. The developer did the integration, and then, you know, it it goes goes without saying from that moment on.
So so that’s kinda what we’re working on now, allowing a nontechnical persona to kinda build everything with regards of how tool looks like, how tool behaves, what are the inputs and outputs of a tool, and what is the logic of a tool without, you know, involving a developer into the loop.
So so that’s that’s that’s for me, you know, being able to remove the burden of a developer. Being a developer myself, that’s that’s super exciting for me.
The new age of developing software, you called it, and I and I love it.
I think that Oh.
You know, as a developer from from age of fifteen, that’s it’s crazy. It’s, like, you know, twenty and something years ago. It’s it’s everything is completely different, and you can create value for customers so fast.
And the only worry that we have is is not about how fast you’re doing it, but about how you can actually make sure that it’s production ready.
And I think that this is what we’re talking about. Right? We’re kind of working with this evolution that is happening in the industry, but guiding it through guiding our customers through this evolution so that they can actually make it work with their existing products.
Because companies that exist for a few years, like Frontegg, you know, we have something to lose. Right? We have a stake in the game already. We have customers. We cannot just, you know, open things up without without controlling it, without guardrailing it. And and I think that this is what we also try to enable for our customers here.
So great stuff, and we’ll have more of it tomorrow. So we’ll have you Yeah. Tomorrow as well. So if you kinda you know, long days for you during this week.
Avia, thank you so much. Before you leave, maybe we have one more question. So time for one more question.
Okay. So the question is agent security is a huge conversation right now. What threat models were top of mind as you built it? Okay. So super interesting. And I think that also we need to answer it carefully without stealing the show from tomorrow Yeah.
Because it’s going to be Yeah.
We won’t steal the show from tomorrow, so we have a lot of kinda stuff to present tomorrow.
But yeah. I mean, you know, I always say that the part of bridging the gap between an AI interface and an API is probably, you know, the twenty percent out of this one hundred percent. You know? It’s the easier part of doing.
Opening it up safely is the hardest part of putting out the, you know, the the right guardrails, making sure that everything is observed and everything is is is, you know, kinda kinda monitored, and that you are able to make sure that, you know, an AI agent won’t delete your production database such as, you know, cases that we’ve seen, you know, within the last couple of months is probably, you know, the most important stuff. We we, you know, we have a we kinda have a way of making now a way of we have to open our product up to AI interface. It’s either we do it or our product dies, but it’s either we do it safely or our product dies.
So Right. It’s kinda, you know, it’s kinda a battle that we we kinda need to take into consideration.
So we need to do it for sure. We need to do it safely for sure. And and I I believe that, Sergei, that’s probably something that we’re gonna cover tomorrow.
For sure. This is the plan for tomorrow. So thanks for that question. And and, you know, that’s a big statement. Let’s open it up quickly but securely, and we’re here to support that. So thank you for that, Aviat. And with this, we will wrap up this this day.
And I want to thank Aviad. I want to thank Oded from Cisco joining us and working with with the product, helping us to find, you know, real problems and solve them and not just solve something from the top of our visionary minds.
So, Oded, thank you so much. And thanks everybody for joining us, and we’re super, super, super excited. I hope that, you know, it’s visible about the release of agent link. You can go ahead and sign up. We have a free sign up until the end of the year. We would love to hear from you about your challenges and and try to help with that.
And that’s it. We’ll see you tomorrow on another day of agent link release. Thank you, guys.
Thank you, guys.
The Complete Guide to SaaS Multi-Tenant Architecture
