AutoGen Studio 2.0 Tutorial - Skills, Multi-Agent Teams, and REAL WORLD Use Cases (NO CODE)

we have a brand new version of autogen Studio that just dropped it has new features expanded functionality and we're going to build something for the real world and this video is going to be a little different I haven't built this agent team myself yet so you're going to build it with me and you're going to see my mistakes let's get into it the team we're going to be building today is going to take a YouTube url grab the transcript for it and then convert that transcript into a compelling blog post and a tweet thread this is something that I need it's a real world use case and we're going to build it together all right so the first thing we're going to do is install autogen Studio from scratch now I had an older version of autogen studio and then when I went to upgrade to the newer version of autogen Studio everything broke and I kept getting all these database issues and errors and it was just a nightmare and what finally fixed it was uninstalling autogen studio with this command pip uninstall autogen studio and not only that I actually had to go into my python site packages and manually delete the auto gen Studio folders so if you run into any issues while upgrading to the new autogen studio be sure to do that and I already made a beginner tutorial about autogen Studio I'll drop a link to that in the description below so the first thing we're going to do is create a new cond environment condac create DN AG python equal 3.11 hit enter so I already have an existing environment we are going to remove it and then accept we're going to install everything we need then we're going to grab this Command right here cond to activate AG paste it in hit enter and now we know the environment is active because it's right there next we're going to install autogen studio and this is super simple pip install autogen Studio then just hit enter now once you're done with that we're going to run autogen Studio ui-- Port 8081 and you can specify any port you want sometimes if you close autogen studio in the wrong way it'll say this port is busy and then you just change it to 882 or 883 or anything that you want so I'm going to hit enter on 881 and there is that issue so I closed down Port 8 881 in the wrong way last time so here is the error that you will see so we're going to change that to 882 now and now it works I'm going to hold down command and I'm going to click on this URL to open up local host in my browser all right and here it is the newest version of autogen studio so what is different about it if you go to the build tab you'll see we have new tabs over here for one we have this models tab now you can predefine different models to use with your agents and you can mix and match as you want you can have specific specific models tailored for specific agent use cases it is brilliant so here you can see we have GPT 4 GPT 4 preview and the BLS Zephyr 7B Alpha if you wanted to add a new model we'll go do that right there but we don't need to do that we're going to be using chat GPT 4 today and we're going to actually be using this preview version the other thing that's new and I'll get to it in a second is the fact that you can have agent teams greater than just two agents in the previous version you could only have two agents a sender and a receiver now you can have agent groups where you can group together multiple agents in addition to your user proxy now I haven't tested this yet I'm hoping it works but if it doesn't you're going to see it so switch over to open AI if you don't already have an account go ahead and sign up come to the API Keys section we're going to create a new key I'm going to call it autogen studio hit enter and it's asking me to verify that I'm a human so this is something I'd usually cut out of the video use the eror to rotate the object to face in the direction of the hand interesting all right verification complete all right now I'm going to copy this yes I am going to revoke this API key before publishing the video I know you all are sick of hearing me say that but if I don't I get a bunch of comments about not showing my API key so we'll come back to autogen Studio we're going to click on this model right here and then we're going to enter our API key I don't believe we need to do anything else for GPT 4 everything else should stay the same because the chat GPT T API is the default then we'll hit enter model created successfully now the first thing we want to do is create a skill and the skill that I need created today is a function to go grab the transcript from a YouTube video now how would I go about writing that I'm actually going to use chat GPT to write it for me now just to make sure it is in the format that is similar to something I already know works I clicked over to the skills tab I'm looking at this fetch profile skill clicking into it and this is pretty similar where you pass in a URL and it goes and gets something so this is probably what I'm going to want to do something similar to so I'm going to go ahead and copy it I switched over to chat GPT I'm going to paste it in and then under it I'm going to say here's an example script make something similar to this but instead make it accept a YouTube url as a parameter it goes and gets the transcript from the URL via the YouTube API and then Returns the transcript let's see what we get okay so it looks like this method is going to accept a video URL so that's right so far giving me some documentation that's fine it's going to extract the video ID from the URL that looks good okay and it's going to use the Google api's YouTube API so hopefully this works we'll see we might need to iterate on this if it doesn't work I've seen another method written for this exact use case to go grab the transcript and it was a little bit different from this so I don't know if it's going to work okay so this one says the YouTube data API requires an API key which you can obtain from the Google Cloud console so this is different I don't want to have to do that let's try again show me another way to do it that doesn't require an API key there should be a YouTube API that you can hit directly okay and this looks more akin to what I saw when I did it last time so it's using this Library called YouTube transcript API and hopefully it's able to do it just with that and it says I need to install the YouTube transcript API with Pip but I'm going to try it without that that so I'm going to go ahead and copy the code and I believe autogen Studio when I paste this in will automatically know that it needs to install that Library so I copied it switch back to autogen Studio we're going to create a new skill I'm going to call it fetch YouTube transcript I'm going to take what's in here which is a sample I'm going to delete it then I'm going to paste in what I just had it has an example of how to use it at the bottom I might as well leave that in and then I'm going to hit okay so that's done so now we have our model and our skill created now let's create an agent we have our primary assistant and our user proxy but I want to create an agent team and the Agents that I want are one I want an agent to go and get the YouTube transcript and remove the timestamps and then return it the second agent I want is to take that text and then convert it into a blog post and into a Twitter thread let's see if we can do that so I'm going to click new agent the agent name I'm going to call it transcript getter for the description takes a YouTube URL and gets the transcript from a video I'm going to say Max consecutive auto reply just leave it as the default human input mode never and then the system message this is where I believe we need to actually tell it which tool to use you are an AI agent that uses the fetch YouTube transcript skill to get a YouTube transcript for subsequent processing we're going to be using this model now if you wanted to add a different model you do so right here now here for the skills this is where we need to add our skill so I go ahead and click plus here we go fetch YouTube transcript I'm going to go and say add skill and there it is then I click okay so now we have our transcript getter then let's create our second agent and this one is going to be content writer takes raw YouTube video transcripts and converts it into blog posts and tweet threads Max consecutive auto reply stays the same human input mode never now for the system message you are an insightful intelligent and witty content writer who is able to take raw YouTube video transcripts and turn them into blog posts and tweet threads we do not need a skill for this guy and we're going to leave the model the same as before so now we have our two agents right there now the last thing we need to do is create a workflow so we'll click over to workflows and we're going to click new workflow now this is something new previously when you clicked workflow it just automatically opened up and you can only have two agents now you can have two agents and you can also do this thing called group chat so we're going to use group chat because we're going to have three agents the content writer the YouTube scraper and then the user proxy so we're going to hit group chat we're going to call this YT script to content workflow description so we're going to say takes a YouTube url gets the transcript of the video and then creates a blog post and tweet thread from that transcript summary method we're going to leave as last which is the default the sender is going to be the user proxy and then the receiver is going to be the group chat assistant but we need to fill out the group chat assistant so go ahead click into it now we we have two primary assistants here so I'm going to go ahead and delete both of those and we're going to add the content writer and the transcript getter I'm going to keep it the same name I'm going to leave the description default and then for the system message I'm going to be very descriptive okay I'm being very explicit here I actually don't know if you need to be That explicit in the system message cuz typically a system message you're just setting up the role you're telling the agent who to behave as not necessarily describing what you want them to do which we will likely do again when we actually submit the prompt you are a helpful assistant skilled at okay coordinating spelled wrong coordinating a group of other assistants to solve a task the task you will solve is taking a YouTube url having an agent use the fetch YouTube transcript skill to get the transcript from the YouTube video pass that transcript to a Content writer and then having the content writer create a blog post and tweet thread based on that transcript now it's interesting the group chat manager also can take a skill and I don't think we need to give it a skill because the agent itself has the skill we'll find out ah and here's a bug that I've noticed so if you remember I I removed these primary assistants and I added other assistants but when I start typing it reverts back to these primary assistance so remember fill out everything first and then modify your group agents so I'm going to delete these again and I'm going to add the content writer back and the transcript getter and then we're going to hit okay and okay again let's double check that it actually does have the correct agent so I'm going to click here click there and it does perfect okay so now we're going to get out of here and let's test it out I hope it works we're going to go to the playground we're going to click new session and we're going to select YT script to content and then hit create so here we go take this YouTube URL and I paste it in one of my YouTube videos use the fetch YouTube transcript skill to get the transcript from the YouTube video remove the timestamps if any then create a blog post and tweet thread based on that transcript now the one thing that's kind of difficult about autogen is it doesn't have explicit delegation options I can't explicitly say this group chat agent should delegate to the content writer or the YouTube transcript getter you can't do that and I think we just have to be extra explicit in our descriptions to get it to work that way all right enough talk let's see what happens okay so we got an error and it says that the API key is not there which is interesting so I'm going to go back to build I'm going to go to models here it is and it is there let me see what that message is one more time so I click here and I'm just going to say retry so the API client option must be either set by passing environment variable okay that's weird I don't understand why I have to set the environment variable since I have it set in the model right there but that's okay we're going to do it anyways so I'm going to switch back to the terminal going to go go to a new tab and I'm going to type then I'm going to do cond to activate AG then I'm going to type export open AI API key and then equals and then the API key and then hit enter okay so it should be done now I think I have to restart autogen studio so I'm going to do that so I come here and I'm going to click contrl C okay now we're going to restart it there it is going to go back to autogen studio now going to refresh the page going to go to playground and let's click retry nope let's try that again I'm going to quit out of here again and I'm going to do export then the API key one more time and then we'll start it up and let's refresh the page and let's see if it worked that time it did okay so for some reason going into a new tab made it so that it didn't actually see the API key which is weird so just keep that in mind if you run into that issue now we can tell it's working because we can see these three dots bouncing right there so switching over to the terminal I can actually see the code and the back and forth between the agents as it's going so it is saying yes use the fetch YouTube transcript skill so that's good and it does know to install the YouTube transcript API now one thing to note is that in autogen it doesn't actually show you the back and forth conversation I do know that there's an open ticket to make it do that so you don't have to look at the terminal while it's going but here it is so there it got the YouTube transcript that's perfect okay it's done look at that all right so here it is let's take a look look at what we got so here's the code output okay we got all that that's great let's look at the actual agent messages so the user proxy says take this URL so that's me typing take the URL use the fetch YouTube transcript skill get the transcript remove the timestamps create a blog post and a tweet thread then the transcript getter to the content writer is actually saying to accomplish this task we will follow these steps so the transcript getter is actually planning out what we need to do the transcript getter also installed everything we needed writes a p python script to use that skill that we gave it and then here it is so here is the actual transcript from the video and then interestingly it didn't give us anything after that so it didn't actually output it okay so interesting for some reason it didn't actually output the final blog post and tweet thread in the UI itself however if I switch over to terminal here it is here's a suggested structure for the blog post and then it actually writes the blog post and this is pretty good and then we have a tweet thread so here's tweet one tweet two uses emojis uses hashtags I mean this is great now the one thing I don't understand is why it didn't actually output that information into here but I don't think I'm going to be able to solve that in this video we do know it worked though so that's it now you know how to use the new version of autogen Studio you know how to create a skill for your agents to use you know how to assign that skill to the right agent you know how to create the agent team and you know how to execute the task and it did so pretty darn well I'm happy with this result if you want to see me create more of these tutorials with real world use cases let me know in the comments below if you enjoyed this video please consider giving a like And subscribe and I'll see you in the next one

Channel: Matthew Berman

Views: 96,000

Keywords: autogen, autogen studio, chatgpt, gpt, gpt 4, gpt4, microsoft, coding, tutorial, open source, llm, ai

Id: 4ZqJSfV4818

Length: 15min 11sec (911 seconds)

Published: Thu Jan 25 2024