00:00In this video, I'm going to show you the new release tool from OpenAI called AgentKit, where it's going to be very, very similar to the no-code AI automation tools like N8N or D-I-Zapiers, and we're going to walk you through how we're going to create our first AI agents using this tool, how we can be able to choose our models, how we can be able to add our different tools like knowledge base and MCB servers that we can integrate onto this automation workflow, and by the end of this video, you will know exactly how you can set this up to create your first automation workflow and your AI agents using this tool
00:28So with that being said, if you're interested, let's get into it
00:31All right, so before we jump in, a quick intro for those who are new here
00:34My name is Eric, and I have spent years as a senior software engineer at companies like Amazon, AWS, and Microsoft
00:41And I have started this YouTube channel to share everything I have learned along the way, from AI and coding to automations, web 3, career developments, and more, all broken down into practical tutorials that you can actually follow
00:55So, if you're ready to level up, make sure to check out my YouTube channel and hit that subscribe button
01:00Now, let's get Back to the video
01:02So, first let's take a look at how the AgentKit from OpenAI works
01:03So, here you can see that this is the OpenAI agent platform, and basically it's very similar to the Nan or DeFi where we can be able to use a canvas to be able to build our workflow, our AI agents on that canvas without any code
01:15And the process is very similar where in the building process we might have like an AI model which is chose by the open AI like GBD 5 or GBD 4.1, and we also can be able to have a list of tools that we can choose from like web search or be able to call the database or be able to interact with any other tools, or we can Also, be able to pass any prompts for the models, and also guard rails on how we can be able to monitor the workflow, and eventually we can use that to deploy this whole thing
01:39We can also be able to further optimize this as well, but I think what really separates the OpenAI AgentKit from other platforms like LangChain or Dify is that you not only can be able to build the agent workflow with the canvas in a visual first way, but you can also be able to start with a code first to be able to build the agents using their agents SDK to be able to build the code first, and then you can also convert it back to the canvas mode to be able to see the agent's workflow visually
02:03So here you can see below here, it sets the SDKs with Node, Python, Go, right with type safe libraries, and you can also be able to visualize that with the agent builder to be able to drag and drop different nodes and be able to translate that back to code if we want to
02:16So clearly you can see that both of those things have increased their flexibilities in terms of how we want to build our AI agents
02:21Now here, in terms of the built-in tools, definitely does not have a lot of option compared to other tools like Nan or D5
02:26Clearly you can see that we're using the web search
02:30functionality from OpenAI, and also file search, image generations, code interpreters, computer use, and connect to any other MCPS, which pretty much are the tools that we have
02:39So with that being said, let's take a look at how we can be able to use this
02:41So first thing first, I'm just going to click on login, and here I'm just going to select the API platform
02:46All right, so once we click on the platform here, you can see platform.openai.com, right? So here inside of this, if we were to scroll down, or we can also just search, but here, usually inside of the agent sections, we have a build agents section, right? So we're just going to click on the subsections, and then we're just going to select the agent builder
03:03So here, we're just going to click on open agent builder, or you can just simply go to platform.openai.com/agentbuilder right here
03:10So here you can see that we have the options to create a workflow
03:13So build a chat agent workflow with custom logic and tools
03:15So in that case, we're just going to build a workflow from scratch
03:18All right
03:20So right inside of our new workflow here, you can see that we have two nodes
03:22One is the start node, and the other one is my agents
03:24So here, this is the start
03:26So that's going to be the start of the workflow, and then here is my agents
03:30So here, inside of my agent, you can see that we can give it a name
03:33For example, we can call this Bob, for example, right? Or it could be like a specialist agent, like an AI email writer, right? So for example, like AI email writer, and then here we can also be able to set the system prompts on the instructions for what this large language model, or this AI agent, should do, right? So that's going to be the system prompts, and then we also have our memory
03:53So what is the past conversations that we have with this agent? So it's going to keep track of that for the past conversation on the chat history, so that's kind of like the memory on the past conversation that we have so far
04:04And then there's also the model, so for example, we can use GP5, we can use GB5 Pro, right? We can also use some other models that we have um options from, right? So these are all from um Open AI
04:14And then here we also have the reason effort, and basically we set it to low
04:18What it means for reasoning here is that basically if we set it to low, that will consume less tokens, right, because it will be less thinking time and faster response, so we can be able to choose that depends on our needs
04:29And then there's also tools, so we can be able to add additional tools like MCP servers, file search, or if there's like local functions, like a code function that we want to call, or if there's like a custom tool that we can call, we can also integrate it there as well
04:34Okay
04:40And then there's also the output format, so for example, we can output as a text, JSON, or a widget
04:45Okay
04:49So for demonstration, we're just going to choose text for now
04:50And then here, we're just going to click on preview
04:53Okay
04:55So here, I'm just going to provide a prompt saying that, please write me an email for taking a day off for work
04:57Obviously, I
04:59did not provide any instructions or system prompts, so I'm just going to send this request, and let's see what it does here
05:03All right, so here you can see that we get an error because currently we don't have any credits inside of our billing
05:07All right, so in that case, what I'm going to do is I'm going to go to the overview for the billing page, and here I'm just going to add some additional credits to demonstrate this video
05:13All right, so I just add $5 here just for demonstration here
05:17All right, so in that case, back to preview, I'm just going to say hi again
05:20Hopefully, this will work
05:22Awesome, so Now you can see that I just say hi again, and here you can see that this is the response we get from the agents, which here says, "Hi, how can I help you today?" Okay, awesome
05:29So once we have demonstrated that it is working, what we can do now is we can be able to explore some tools that the AI agent has
05:34So here, I'm just going to call it the LM here for the node, just not to confuse ourselves, and then here we're just going to click on tools and see what are some options we have
05:43So let's try to explore how we can be able to add our knowledge base into our agent kit workflow
05:48So in that case, what we're going to do is we're going to Select the file search option from the OpenAI, and here we're just going to click on file search, and here what we can do is we're going to be able to upload some documents, use that as a knowledge base to retrieve relevant information
05:59So, for example, I just upload a file for my channel information, right? So here, I'm just going to click on attach, and it basically is going to attach this information as a knowledge base for the vector search
06:10So what we're going to do is we're just going to make sure to save this, click on publish
06:14I'm just going to say that this is going to be a test workflow, and click on publish
06:18Awesome
06:20So the workflow is published
06:22We can always be able to use this, but we can also use that to save
06:23Awesome
06:25So in that case, let's try to test it out to see if it works
06:27So here you can see that I click on the preview for test
06:29So here I'm just going to ask a simple question like, what is the channel's uh main focus? So here I'm just going to add this question, and if you were to look at the MD file, which is this is the one, and here you can see that it contains the channel focus or the content focus here, right? So this is the answer, the channel name, the language, the channel type, so let's take a look at, you know, what it does here, so you can see that the LM here starts to do some reasoning, starts to search the files to find the answer for what is the channel focus, right? So these are some questions that it's asking
06:59So here you can see that we have the answer, which is that Eric Tech is focused on AI, automations, software development, career growth, growth, and productivity tools
07:04Awesome
07:09So here you can see that we can be able to attach knowledge base onto the large language model here for the AI agents to be able to retrieve those information
07:11All right
07:14So now we know how we can be able to use the knowledge search functionality
07:16inside of our workflow for the Agent Kit
07:18So now let's take a look at how we can be able to add our MCP servers functionalities inside of our workflow
07:24So let's say if we want to add search functionality, for example
07:28So let's say if I were to click on the LM
07:30So let's say if I want to add the search functionalities, we can click on tools and be able to use the web search functionality from OpenAI
07:35But let's say if we don't want to use that, we want to use like maybe a third party, um, MCP servers for web search, maybe like Tabi
07:43Then what we can do is we can click on the MCP server here to connect our desired MCP server
07:47So Here, we're just going to click on MC server
07:48So here, you can see that we have a list of servers that we can choose from, like Gmail, like SharePoints, or some third-party servers that are also accessible that we can use as well
07:58But obviously, right now here, you can see these are all the servers that we have, which is really limited
08:00So if we want to add a server, we can just click on add server here
08:04And here, you can see that we can simply just paste the URL and also the authentications to connect our MCP server with the agent kit
08:11So in this case, for this demonstration, I'm just going to use Tavi here for the MCP server
08:14demonstration here
08:16I'm just going to click on get API key, and the way how we can be able to use Tavly MCP server here is that it says here in the doc for remote MCP server, we can simply just copy the URL here and paste the API key here
08:25So here, I'm just going to copy this
08:27So in this case, I'm just going to come back to agent builder and paste the URL here, and I'm just going to get API key and fill it in
08:34So I'm just going to come to the Tavly API platform, and here you can see there's option for us to copy the API key
08:40I'm just going to copy this and combat agent builder, paste the key there, and then below here I'm just Going to click on "none" for the authentications, and in terms of label, we're just going to name it "Tavi," and we're just going to click on connect
08:52So here you can see this is going to try to establish the connections, and here you can see that these are all the tools that we have, like the search, extract, crawl informations, and so much more, okay? So now what we can do is we're going to click on "add," and here you can see that in terms of tools, we have our Tavi and our knowledge base for Eric Tech
09:09All right, so now what we can do is we can be able to click on preview to test it
09:12Let's say if I want to search for why is Bitcoin down today, and for your reference, I'm just going to say October 10, 2025, and here I'm just going to click on enter
09:22Let's see what the large language model here does
09:23All right, all right
09:25So here you can see they start to use the Tavily search functionality to search for the related terms for the query that we're looking to search
09:30So here I'm just going to approve this, and then the large language model here asks for another permission for the Tally extract functionality, so it's going to extract information from this site
09:38So here I'm just going to click on approve to extract those
09:40information
09:41All right
09:42So with couple approval for the tablet extract
09:44So here you can see that it starts to search for other sites, and here you can see that after it starts to extract and search the sites, it has gathered the results and found the answer
09:53So the short version is that on October 10th, after the US escalates its trade war with China, here is what happens, right? So the drop also came just days after the BTC hits its new time high, so the profit taking and the cool off phase added to the move, right? So that's basically what causes the drop, and there's also like detailed explanations on where
10:11It found the resource and so much more, right? So pretty much, you can see that we can use the MCB server tools instead of AI agents from our OpenAI agent kits
10:20So now, if we were to, you know, publish this, making sure that we save this, right, we can also save this in a new version
10:26All right, so back to the workflow, let's take a look at what are some other options we have
10:29So just like any other, you know, no-code AI automation tools, we have the option for if-else statements, right? So we can add an if, we can add a while loop, we can also add a user approval
10:38Uh, there's also guards, which can basically help us to monitor our system
10:42Making sure that there's no hallucination checks or moderations, right, the content, and be able to do some filtering for the response that the AI generates, and there's also like sticky notes that we can be able to keep our text or workflow much more organized, we can also do that
10:56And also, there's some data transformations like reshaping the data or setting variables, like set state, right, assigning variables or values to a workflow state, so we can also do that
11:06And uh, let's also take a look at the most powerful thing, which is also how we can be able to convert this, what we have so far, into code, right, into how we
11:10can be able to use this with the Agent SDK
11:15So here, I'm just going to click on code, and if I were to click on the Agents SDK here, you can see that's going to be the full code on everything that we have just done in the canvas
11:21So, so things like the MCP connections or the file search connections, and then here's the large language model, right? So, these are tools that we have added, the model that we're going to use
11:31This is the instruction for the system prompts, and then here is the model settings
11:34So, the reasoning we set to low
11:36So, pretty much you can see we can get the code for this for what we have designed
11:39inside of our canvas
11:39Okay
11:40All right
11:41So, pretty much that's it for this video
11:42So just to summarize everything, we went over how we can be able to set up our AgentKit from OpenAI, and also how we can be able to use our AgentKit to build our simple AI agent workflow, and how we can be able to use the tools and be able to use different models, and things like how we can be able to add our knowledge base, our MCB tools, and so much more
11:57So if you do find value in this video, please make sure to like this video, consider subscribing for more content like this, but with that being said, I will see you in the next video.
