Video ID: NivJPs8dAb4
YouTube URL: https://www.youtube.com/watch?v=NivJPs8dAb4
Added At: 13-06-25 21:16:42
Processed: No
Sentiment: Positive
Categories: Tech, Education
Tags: OpenAI, AI agents, no-code, programming, tutorial, vector store, web search
Summary
To use OpenAI's new responses API to build AI agents, start by creating a vector store in the OpenAI developer playground and uploading your YouTube transcripts. Then, you can train an agent using the OpenAI agents SDK and use it to search through your transcripts. You can also add web search capabilities to your agent to give it more autonomy. This is a powerful tool for no-code enthusiasts and developers alike.
Transcript
OpenAI just revealed that you can build AI agents using their new responses API. That includes web search, file search, and even the ability to control your own computer. we're going to learn in this video how to use these features to develop autonomous AI agents. Perfect for you if you're a no code builder. And I want to simplify it. So hopefully by the end of this video you'll be building your own AI agent using the new releases from OpenAI. Now we're going to use Cursor AI because it's a brilliant no code tool. And I've actually gone ahead and trained in all the documentation from OpenAI into my cursor rules. Now you can access this by going to settings sets up in the top right of Cursor. And then you go to rules over here. And when you go to rules you'll see there are project rules. Now I've trained in the Python Agents SDK from OpenAI and also the responses API from OpenAI. If you want to get these rules and use them for yourself, they're in the premium section of my community. Just use the link down below, join up and we'll also discuss how to use these powerful new features. Now let's get straight to it. Close down those rules. This project will now follow all the best practices from OpenAI. Let's say create an AI agent using the open AI agents SDK that searches through my YouTube transcripts. I'll upload those to a vector store, which, by the way, OpenAI just announced this cool vector store, which is kind of like a really cool Rag database. If you haven't seen my videos about quadrant in previous videos about making agents. This is pretty groundbreaking in itself. So I'll upload those and then it provides answers based on what I said in my videos. Call it Agent Magic and also make a nice web interface to chat with it. And as always, I like to ask for the valuation of simplicity, because often when I type something into cursor, it goes completely crazy and goes over the top. Okay, it says, I'll help you do this, and it's even reading the rules now, which is a great sign. It's paying attention to exactly how to use OpenAI's new release to make this agent. For me, it's thinking about it very hard, and now it's exploring what's already there. So you can go ahead and create my first AI agent. Okay, first, I'll run this command to get it up and running. Now it's creating a requirements file so it can install dependencies. Even the brand new OpenAI agents is included. So this really is listening to the rules. And now you can see it's creating a flask web application. This will essentially give a website where I can interact with my agent. Finally, it's created an env dot example file, which is where I'll put my OpenAI API key. Now, look at this. It's funny because it stops running the agent in cursor after 25 tool calls, but will resume and let it keep generating until it finishes my AI agent. It's even making something to create a Docker image of this app. This is absolutely insane. That's amazing. In just one prompt. All of this code here has been created. That's nuts. So I guess now I'll go and get an API key from OpenAI. Let's create something and we'll call this Agent Magic and just create a secret key. Then we'll copy that key and paste it into. Well, I'll copy this and make a proper env file. So let's rename that to dot envy. And then we'll paste the OpenAI key in here Okay. With that done, let's take a look at the readme to see what is mentioned here. now everything's up and running. Let's run the python command and see if it works. And I can't believe this. It is working. Okay. Wow. Agent magic, ask questions about your YouTube videos. What do you know? And remember, these are only sample transcripts at the moment. I'm here to help you find information from your YouTube transcripts. If you have a specific question. So this is actually working. Okay. Tell me how to make a great video. What is coming back? I'm experiencing trouble accessing the transcripts. Okay. Every time I try to ask a question, it tells me it's having some trouble accessing the video transcripts. Now, I think this is because you've got to actually upload these transcripts to the OpenAI vector store in the open AI developer playground. So let's try that. So here I am in the OpenAI developer playground, and I'm going to go to storage here. And you'll see vector stores. This is new vector stores up here. And then you can create your first vector store. So like I said at the start of the video, this is like a Rag database. I might not need quadrant or AI Mike anymore. I can just do it like this. So let's call this AI AI Mike open AI. And that's the take. Now this is important. This is our vector store ID that we're going to need to give to cursor. And I guess I can add files. So let's try that okay. Drag your files here or click upload. I've gone here to my transcriptions folder. I've got all of my YouTube video transcripts downloaded, and let's just grab a few of them, maybe back to the start of 2025 and let's see if we can drag and drop. Yep. It's uploaded all of those transcripts, so we'll attach. And now it looks like all of those will be yes. They're in my vector store inside my OpenAI account. This is great news. Now, even though this app seems to be working, I'm going to go to the chat here and say. The transcripts can't be found by the agent. I think we need to reference a vector store ID from OpenAI. Mine is. And then there's my, little vector store ID. Can we put that in the EMV example and use it throughout the code? So it's doing a few checks. It's looking at the main code set up. Now it's looking at the main agent.py file and the vector store file as well. Very important. And it's reading my EMV. going to go ahead and update everything. There we go. Vector store ID needed. It's even adjusting my actual EMV file, which is cool. And finally it's editing agent.py. That's just to make sure Vector Store is called by my agent magic. Okay. I think it's done. So let's try and run that Python file again. And yes, it's booted back up and let's open it up. Okay. This is our second attempt to Agent Magic. So what do you know about open AI? Wow. OpenAI recently announced the release of ChatGPT 4.5, which is in research preview. This is exactly what I said in a video just a week or so ago. It is now directly calling the files I've put in my vector store, and a gently presenting them to me right here inside the Agent Magic interface. This is cool. Focusing on emotional intelligence and effectiveness. Wow wow wow. Okay, let's get more personal thoughts on Elon Musk. and it says here the transcripts don't have a detailed section directly discussing Elon Musk. Okay, that's good because actually, maybe I haven't mentioned Elon in 2025 so far. But now obviously in this video I have, let's try another question. And this one should be an easy one. What did you say about Zapier recently? Of course. I've done a video under two weeks ago about Zapier, using it for online trading. Recently, Zapier was mentioned in the context of an AI trading bots. Wow, this is cool. Literally, I can replace my whole stack I was using quadrant and all kinds of other automation tools to make an AI agent work. But now with the open AI agents SDK and responses API, I can throw all of my transcripts into a vector store right inside OpenAI, and then I can code something using the rules that I will provide in the premium section of my community, linked up down below. This is absolutely nuts. Let's try a little tougher one. Now, what is the name of your channel and see if it gets that one right again. Agent magic is searching through. The name of the channel is Creator Magic. It's a leading YouTube channel focused on AI powered content creation with consistent weekly updates. No way. Open AI agents are totally insane. And if you are a no code enthusiast using tools like curse, you got to try this right now. Okay. Let's next level this and add another a genetic feature from OpenAI's new release. It can do file search perfectly. Well, I think we've proved that. But what about web search? Yes. The AI agents from OpenAI have the ability to do that. So I'm going to add in here another prompt to cursor saying in addition to file search by the vector store, give the AI agent the web search ability. Clearly. Note this with a cool animation in the web UI showing it's doing an online search only do a web search if you want to present more up to date information for a query. So essentially we're giving the agent autonomy to decide whether it can just pull the latest from my transcript in that be enough or whether it's actually going to go ahead and pull stuff in from the web. So it's looking at how the agent is currently built and understanding how the app works. It's also listing everything that's in the directories here. Just while we wait for my AI agent to build, look at this. The logs console in my open AI API dashboard is filling up with all the questions asked of my agent. This is really cool because when your AI agent is interacted with, you'll be able to see all the interactions right in your open AI account. Okay, I think it's updated. So let's try running this Python app again. And yep, it's all up and running. So let's control click. And here we are with Agent Magic V3. So let's just say what is the very latest about open AI and see if it wants to search the web. Oh search the web. This is so epic. Now note I have not uploaded the transcript of this video. I can't because I'm still making it. So not only is it thinking about what's in the transcripts of my video, but it's looking for an answer. Oh my goodness. Okay, not very well formatted, but OpenAI have been actively expanding partnerships in technological offerings, blah blah blah, blah, blah. Okay. And a lot of sources not very well formatted. One little problem before we go on. When it searches the web, the source is not nicely formatted. Please make them look pretty. And the links work. And I'm pasting in a sample here. Okay, that seems to be working. Let's run the Python app again and open. Here we are. Okay, tell me some latest news from open AI immediately knows to search the web. I'm digging the animation that's being generated, by the way. Wow. That's insane. Look at this. Not only has it said all the latest stuff from OpenAI, but it's putting citations all the way through with direct links out to the exact news stories. These are real links that go to real places and all the sources linked at the bottom. This is pretty insane. This is Agent Magic, and that is an introduction to the new agents that you can build right now using the open AI agents SDK, which is Python, and also the responses API. But of course, you don't need to know any code because you can use a no code tool like cursor AI and build these apps yourself. Now they can be fully agent multiple agent processes and. You can monitor everything you've done in the logs, from web searches to searches through transcripts. They're all recorded right there inside your open AI API playground. This is really super cool stuff. I really hope you enjoy building and to help you with building, I have uploaded my cursor rules for building with the latest agents SDK and responses API, so if you want to grab them there in my community, just join the premium section and they'll be available for you to download. And we'll be discussing more about building awesome AI agents with these epic open AI tools. Let me know what you're going to do in the comments down below. And YouTube is showing a video on your screen now that you should watch next.