AI DevOps Project #3 - AI DevOps Teammate in 15 Minutes (Free & open source

Added: 2026-06-02

[00:00:00]Hello everyone, my name is Abhishek and welcome back to my channel. In today's video that is the third video of our AI DevOps project series, let's go ahead and build AI DevOps teammate.

[00:00:15]Trust me, you will be able to build this in 15 minutes of time. We will be using concepts like rag, MCP, skills. But still by the end of this video, even if you don't have mastery on these concepts, you will be able to build this. I'll show you the tools that you can use. In fact, you can use pre-tools, open-source tools, and even enterprise tools. I'll talk about all of them in the video. So, let's start the video by understanding problem statement. Then I'll explain the different tools as I told you. And finally, I'll pick up one of these tools and perform the demonstration. It's going to be very exciting. This is something every organization should have. So without wasting any time, let's get started.

[00:01:04]First of all, what is this AI DevOps teammate and why do you need this?

[00:01:10]Today if you look at DevOps teams they have different people within the DevOps team like you have junior DevOps engineers you have senior DevOps engineers and sometimes architect as well in the team or the DevOps team.

[00:01:27]Now what most of the people do today so they make use of AI it has become a common thing they usually go to chat GPT or within the organization if charg is blocked they go to the co-pilot whichever they have access to and they ask questions to the co-pilot now this is good using AI should definitely be encouraged but there are two downsides to this two very important downsides one let's say junior a DevOps engine asking AI maybe for some Kubernetes command problem one it can give you obsolete information maybe it can give you information for Kubernetes 1.29 29 version very common with AI if you don't specifically provide it with the version it will give you the version that is popular on the internet maybe it can give you old command old version and if you're asking for a manifest there is a problem the manifest is deprecated or the API version used in the manifest is deprecated second problem and the most important security issues right this is the reason lot of companies are running into uh AI security attacks these days. People make use of AI but because of the prompt that they provide or because of the instructions that they provide to AI assistant, it can leave you with commands or it can leave you with code.

[00:02:57]When it comes to DevOps, it can also lead you with leave you with manifests which have security issues. So how do you overcome this?

[00:03:06]So that is where you need an AI devops teammate. So instead of everyone using different things, one engineer in the team using chat GBT, other one using GitHub copilot, other one using something else, but you have to suggest or what we are going to learn in the video, we will build an AI DevOps teammate that will be internal to your organization and internal to your DevOps team. What advantage would we get, Abhishek? It would solve these two problems, right? The first problem again it would solve the absolute information problem because we are building it. We will make sure it returns the latest information and updated information.

[00:03:52]Second, we will ensure it does not leave any security issues within the organization and it has the context of the organization. Context is the main thing. So within your organization, let's say you ask a question related to Kubernetes deployment. In the organization, you're using Argo CD, but it gives you the information of flux CD.

[00:04:17]What's the point of it or to be even precise? Okay, Argo CD is flux CD. Uh Argo CD and flux CD is still something you can say abishek I can explicitly mention Argo CD. Okay, that's fine.

[00:04:31]Let's say within your organization within the Kubernetes cluster you have developed a proprietary cubectl plug-in.

[00:04:39]Now this is not available on the internet. This is not public.

[00:04:44]You ask a about this. It does not have the information. So what is the point?

[00:04:49]Right? It should be with the security.

[00:04:52]It should give you the secure output.

[00:04:54]Along with that it should also have the context. Right? So our AI DevOps teammate that we are going to build it will come with three things. It will always give you the latest information.

[00:05:07]It will give you the information that is secure and it will give you the information that is in accordance with the context of your organization. The compliance context, security context or any internal proprietary tools context.

[00:05:23]Cool. Now the big question is okay Abishek I got the problem statement it is clear because everyone of us are running into it but how are we going to do it before tools before I explain the tools let me first explain you the concept how can you do it so you need three important concepts here one you need agent skills second MCP P third rag.

[00:05:58]So when you build an AI using these three concepts today in 2026 then your AI is going to be very robust and even the AI DevOps tool that we are going to build the AI DevOps team.

[00:06:14]We will add skills to it. We will add required MCP servers to it and we will add the rack to it. Abishek what are these things? Now I'm very new to AI. I don't understand. I'll explain this in simple words to you. You don't have to master this in today's video. In fact, even the output that we are going to build, you don't have to master this.

[00:06:37]You don't need to know the in details of it. Only the things that I'm going to explain if you understand it's good enough.

[00:06:44]What are skills?

[00:06:47]So agent skills is a concept or agent skill is a concept where you make AI right or you help AI understand what kind of output are you looking for example the same thing within your organization you might be using Kubernetes 1.33 so within the skills which is an MD file which is a markdown file you will tell agent we are using 1.33 version we are using argo OCD right we are using isto right this type of information you can give to AI so that whenever it is ready to respond before it performs the analysis it goes to the skills.mmd it sees what instructions have you provided and according to your instructions it will give you the output it will not give you the output for kubernetus 1.29 29 but instead because you're using 1.33 it will modify it search and it will give you the result for 1.33.

[00:07:52]So whatever you require you define that in the skills.mmd.

[00:07:56]Then MCP what is MCP? Basically you can provide list of tools or you can use MCP servers that are available on the internet. For example Kubernetes MCP server. What it does instead of your agent going to different sources on the internet, it will not search for answers on Reddit. It will not search for answers that it already has. Instead, it will go to the Kubernetes official documentation and give you the output. So this way you can eliminate security issues, right? For example, you ask your agent a question. It goes to Reddit. Maybe it does not have the information of it. It goes to Reddit. It finds some information and it gives you the information back. The problem in this case is that what if the Reddit answer has some security issue, the Reddit code has some security issue. If you use MCP server, you can eliminate that because the MCP servers tell agent if you're using Kubernetes MCP server, go to the Kubernetes official documentation and read the information from Kubernetes official documentation and give the output to the user. So your output will be secure.

[00:09:18]Okay. Then finally, what is this rag?

[00:09:22]Retrieval augmented generation in simple words understand it this way within your organization you have some internal documentation okay maybe you have some internal tool or in this case you have a internal kubernetes plug-in just for example AI will not be aware of it because AI is aware of public information it is trained on the public information using the concept of rag you can provide this documents so you can take the documents ments you can store them in vector database right and what AI does if it does not have the information it will go to your vector database it will look for the document if the information is available in the document it will give you the information back so what rag does anything related to your internal organizational context like internal documentation internal tools it will help AI understand that concept and give you the output. So that's why we will be building our AI DevOps team with skills MCP and rack. In fact, any agent that you are going to build, not only this thing, but if you're building something within your organization, make sure you use these concepts. So these are AI concepts, three AI concepts that are very useful in 2026. Maybe down the line we might have a new concept that can integrate all of this. But right now this is how you build AI agents in a robust way. Now let's come to the actual thing. Abishek should I learn all of these concepts now? Like I have to learn how to implement rag. I have to learn how to implement skills MCP server not needed.

[00:11:11]Today things have become very easy. All that you need is a workflow automation tool.

[00:11:20]Workflow automation tools are very very useful for DevOps instance.

[00:11:26]For example, you have workflow automation tools like nit, sim.ai, retool. Of course, there are many new in the market at this point, but these are very reliable and trusted. Abishek which ones to go with and what is a workflow automation tool in simple words of course I'll show you in the demo but in simple words a workflow automation tool you don't have to write code you don't have to uh give it a lot of steps you just have to tell it the flowchart we call it DAG but in simple words layman terminology understand you just have to provide it with the flowchart which tools that you want to use what is the input what is the output and that's it.

[00:12:12]It will build the entire workflow for you. In this case, it can build the complete AI DevOps team made for you.

[00:12:20]But Abishek, which ones to use? Okay, you suggested three options here. Which one should I go with? Okay, I'll give you the options when to go with Win Tool. First thing, if you want something robust, if you want something that is there in the market for a while that has revolutionized this space, go with an item.

[00:12:43]Or let's say you want to go with opensource solution. I want to go with something open source, something free for my organization, then you can go with sim.ai.

[00:12:55]Basically, it also comes with integration. So you can run sim.ai they are using docker compass integrate it withama and integrate it with your local models. If your organization has something like a DGX spark box or if you have local compute within your organization maybe on premises servers this is the best option to go with.

[00:13:18]Finally, if you're looking for enterprise solution, Abhishek, we are not bothered about money, but we are bothered about enterprise level features, enterprise level security, then go with retool.

[00:13:32]So, these are the options. NAN, if you want to go with something robust, open source and free, go with SIM.AI retool, if you're looking for enterprise option.

[00:13:42]Of course with N as well you can integrate OAMA but it's not that stable when compared to SIM.AI.

[00:13:49]I know majority of you would be interested with opensource solution because you want to implement the same thing. You want to show this within your organization.

[00:13:59]We will go with SIM.AI for today's demonstration. But saying that it doesn't matter. You can follow the same steps in any of these things. In fact I made video on N10. I made video on retool. I made video on sim.ai as well.

[00:14:15]So if you want you can watch those videos for reference also. Cool. Now let's get to the point. How do we install download sim.ai and how do we build this tool in next 10 minutes of time. Just head to the sim.ai documentation. Just search for documentation sim.ai or link in the description whatever you feel comfortable. Go to the docker section.

[00:14:39]you will find the docker compost command. Take this docker compost command and you will have sim.ai running. If you scroll down, you will also find Ola integration here. So you can directly run this command to set up sim.ai with Olama directly. But make sure you have the local compute. You have the GPUs running on your machine.

[00:15:01]Otherwise, don't use this option because if you don't have GPUs, this will not be much use. There is option for CPU only.

[00:15:08]But trust me it will be very very slow in such cases. Abishek what can I do in such case? Then you have the hosted solution for sim.ai as well. See if you can get some free credits. I think if you install for the first time you might get free credits. Try it. With free credits you can build one to two workflows. That's good enough. Same with nan as well. Same with retool as well.

[00:15:35]Right now I'm out of India. I don't have my uh personal machine where I have a lot of GPUs.

[00:15:42]I'm on a different machine. So even I will go with hosted solution only. Cool.

[00:15:48]How do we do this?

[00:15:51]Just head to the hosted solution www.sim.ai or n retool. It looks almost the same.

[00:15:58]The process is also almost the same.

[00:16:01]Click on workflow section.

[00:16:04]Okay. Let's name this workflow. Let's say instead of building an AI DevOps teammate, we are just building a Kubernetes teammate just to make things simple. Kubernetes AI teammate.

[00:16:20]Perfect. Now what you have to do click on this Kubernetes AI teammate. You have an option here start the workflow. So you can use different triggers. I'll use just the simple start trigger. Then I'll go to the toolbar section and here in the toolbars I will just search for an agent. Okay.

[00:16:42]So I'll look for the agent option because we are building an agent.

[00:16:57]Then here you can attach this trigger in the block section to the agent. That's it. So you can see now we will this will start the workflow. This is connected to the agent. The agent already has the model. If you're fine with Sonet 4.6 use it otherwise you can pick up the model of your choice. If you're using Olama integration here in the model section you will find Ola and your local models.

[00:17:25]I'm fine with Sonet 4.6. I'll keep it as is. But Abhishek, where is the rag integration? Where is skills integration? Where is MCP integration?

[00:17:36]This is just an agent. This is just like chat GB. I can use the same thing. No, what you can do. So if you look at the options carefully, there is options for skill and skills option. So click on add skill and you can add a skill of your choice.

[00:17:54]I have Kubernetes skill I mean I just created it. Basically uh I've added four to five points in it like the Kubernetes version is 1.3334.

[00:18:05]If I ask you something related to cluster administration always return the information about these versions only.

[00:18:12]Then I provided I'm using Helm V3. There is Helm V4 now. So I'm just saying my organization is in V3. Always written information about V3. Then I'm saying we are using Argo CD, we are using engineix ingress controller not gateway API and we are using Prometheus and Grafana. So this kind of information you can put in the skills and you can provide that here. Okay. So skills.mmd is done. We have granted it access to skills. What's next? MCP. How do you integrate MCP?

[00:18:47]Again it's super simple. Just go to the tools section, add MCP server and add MCP server of your choice. There are thousands of MCP servers. Literally for every DevOps tool, an MCP server is available open source. You can search for awesome MCP servers and find the best one that works for you. I'll not complicate the concept. Okay, we will uh skip the MCP for now. But this is how you add it. Finally, your internal knowledge base. Go to the knowledge base section.

[00:19:18]to the left side I added a knowledge base called Kubernetes pod investigator plug-in. So basically I'm saying in my organization I have Kubernetes pod investigator plug-in. This is the documentation in your organization definitely you will have 50 pages 100 pages documentation. Just go here add that in the knowledge base. Abhishek is it safe? See as I told you Sim.AI can be self-hosted as well. This is a hosted solution. My bad I told it wrong. This is hosted solution. You can self-host as well. So using Olama you can run it locally. In such cases there is absolutely no problem. Now if you ask me about this particular hosted solution and is it safe or not. I mean I'll tell you have to ask your security team because I'm not the right one to say. If you want that then you can go for retool which is an enterprise tool definitely secure. They come with sock to compliance and everything your organization needs. So basically ask this to your security team but local integration it's absolutely safe.

[00:20:27]Go to the knowledge base and upload the knowledge base. Go back to the workflow you will find the file section. Provide the file the same thing Kubernetes pod investigator.

[00:20:39]Perfect. So we added rag implementation.

[00:20:43]We added MCP implementation and we also added the skills implementation. Our agent is perfectly ready. It's that simple. Now all that you have to do is click on the deploy button.

[00:20:58]This will be deployed. Just give it like 2 to 3 seconds of time. Now this is live. Oh my bad. I provided I forgot to provide the input here. So you just have to provide input. For example, uh you can provide your name, string, uh description, you can just say hi or something and any default value that you want to leave. You can update this um it will create another version like v2 and even here within the agents we have to provide a message. For example, I'll just ask AI to generate the message.

[00:21:34]I'll say Kubernetes. This is a Kubernetes uh teammate that we're building. It will generate complete system prompt for us. I mean if you want you can write it but I have provided the system prompt. Cool. Now again let's click on update. This will create V3 version for us and let's ask it to run it. Cool. So now this is in the running state. But Abishek how do I interact with this agent? Okay, we created a Kubernetes AI teammate. How do I talk to it? Click on this agent. Okay, or click on the start section.

[00:22:13]You see that live option. Go to the live option. Then you have different points to communicate to the agent. For example, you can use the curl command and talk to the agent using curl or you can use it as an MCP server. You can use it for agent to agent. Now Abishek, I'm interested in regular chat GP style.

[00:22:33]Click on the chat section. Just provide name to your chat URL. For example, I will call this Kubernetes Vala agent and I will say it as Kubernetes agent.

[00:22:49]Okay. Output I want to see the content.

[00:22:52]I want to make it public within your organization. You can control the access using password. You can control access using email and even SSO. So again there are a lot of security options.

[00:23:05]This is the entry message. The first message click on the launch chat.

[00:23:09]Perfect. In 10 minutes you have your AI DevOps teammate ready. Now just ask it a question. For example, what can you do for me?

[00:23:23]Okay. It will respond with the instructions that you have provided. It will say I'm a Kubernetes uh expert. I can do these things for you. I can help you with your internal uh pod plug-in.

[00:23:37]Let's see. There you go. Welcome. I'm your Kubernetes expert. I can help you with the Kubernetes 1.33 Argo CD Engineix ingress controller. I can help you with the architectural design, deployment strategies. And it also says like I can do all of these things for you. If you want ask me my pod is stuck in crash loop back off. I can help you with diagnosis. I can help you with any issues with Argo CD. What would you like to explore? So just like your chat GD experience but with MCP rag and most importantly skills.

[00:24:14]So your AI DevOps assistant or your AI DevOps teammate is ready. Isn't this cool?

[00:24:22]Now go back build something by yourself.

[00:24:26]You can use sim.ai, you can use nit or you can use retool. Wherever you get some free credits or if you can host it on your local machine, go ahead host it and try it. I hope you found today's video insightful. Let me know in the comment section if you have any question about it and even if you like it, leave the feedback. See you all in the next video. Take care.