AutoGen Studio 2.0 Advanced Tutorial | Build multi-agent GenAI Application!!

hello AI enthusiasts welcome back to skill curve Micosoft just launched a brand new version of autogen which is called autogen 2.0 we'll see how to set up download and install autogen 2.0 locally onto a system then we'll be using it like on a real world scenario all right so let's try to install autogen Studio from scratch okay so for that I will open this thing up which is an AA Powershell okay so the first thing we need to do is to create a new K environment for that I could say like cond create hyphen n then I could call it as Auto genen okay then python version we need to specify and I will say like python 311 okay so all right so we need to confirm it by typing y on the screen okay so I'll hit enter and it will create an environment for me so that's pretty much simple okay all right it's done okay so now next we need to like activate our autogen environment I'll grab this thing paste it right here and hit enter okay so now our autogen environment is active so first of all it was base and now you can see here it's autogen which means that our environment is active and we are ready to go now next thing what we need to do we simply need to install autogen studio for that I could say like pip install autogen Studio okay it's that simple you don't need to download anything you can simply pip install it to use and run this thing okay so I would simply hit enter all right so it's successfully installed you can see right over here okay so now to use the open AI models and other stuff what we need to do we just simply need our open AI a API key okay for that I simply need to go to like platform. open.com SL playground simply need to log in with my Google account make sure that you have the subscription for like open AI API key that would be really useful for using autogen okay so here I am inside of my playground and here I need to go to like API Keys okay and I need to create a new secret key I would call it as autogen okay and click on like create secret key so here's my API key I'll simply copy this API key come back to my terminal once again here what I'll do I'll simply give it a command like set okay and if you're using Mac you need to type in like export but for Windows it said open aior API unor key and that would be equals to our API key okay then I'll simply hit enter and now my API key has six F been set so now what I need to do I simply need to run the autogen UI so I'll say autogen ui-- port and that would be 8081 okay if your Port is like busy for some reason you could specify another Port like 882 883 but for now 881 is fine okay so I'll simply hit enter oops there's an error okay I guess it's autogen Studio okay yeah Auto Studio ui-- Port 8081 so I'll hit enter so it will probably spit up the server for me and it might take a while so all right it just gave me this URL you can see right over here I will simply copy this URL come to my browser and paste my URL in right here so you can see right over here that our autogen is working successfully so this is how the UI looks like okay so we have like a playground then we have a gallery and then we have a build section okay in which we can build skills models agents and workflows okay we'll explain this in a while now you have successfully created a virtual environment in which you have installed like autogen and it's working successfully so congratulations okay you can also switch to dark mode if you want I guess dark mode is fine for now okay all right so here are the few things that you should do to like improve your experience Okay so first of all it says like autogen requires python version greater than or equals to 38 and less than 313 okay so now if you just scroll down so you can see right over here that it says like even if you install autogen locally we highly recommend using Docker for code execution okay so you need to actually install Docker as well to like run it smoothly and efficiently okay so now what I can do I can simply like shut down this server from here by hitting contrl C okay then I could simply say like pip install Docker okay as it's recommended by Microsoft and it's installed you can see right over here so now I simply need to run this once again okay yeah it's live uh let me just try to refresh it yeah it's here you can see right over here okay so now let me just give you the feature overview like what it can actually do okay so it's basically kind of a web page in which you can do all sorts of things okay you can make teams that could work together and make your work like a lot easier okay so there are three tabs here like build playground and gallery okay we could create skills models agents and workflows inside of like build tab okay like skills skills basically these are the python function that agents can use to solve like different tasks if you go to models okay you could create model configurations that that can be reused in your agents and workflows okay like there are Azure AI model configuration then open AI model configuration and then you can also use like a local model example with vlm server endpoint you can see right over here that you can also use like zire 7B from BL okay then we have agents you could configure an agent that can reused in your agent workflow then there are workflows you could configure an agent workflow that can be used to handle like multiple or different tasks so these are the four things in the build tab okay so you can see that they have like by default three workflows like travel agent group chat workflow then there's like General agent workflow which is like used award and general purpose tasks okay then you have like visualization agent workflow like it's used for visualization tasks okay so like they're really great you could also create new workflows as well okay so all right now let's talk about like skills so skills are basically functions that can do anything okay so if you like see this skill which says like find papers uncore archive so archive basically is like a resource which has like multiple research papers so you can find papers like from there okay so if you just open this up so it would have like a script okay so it's basically like precoded for you okay so if you just move on to like fetch profile what does it do it basically like fetches the content from a personal website so that's really great then what you can do you can just create a skill for like anything like suppose if you just want to create a text to speech capability so what you need to do you just need to like have an API for instance like for 11 Labs or some other tool like whisper and what it will do it will just simply try to convert that text into speech then you need to code like uh a skill for that particular use case okay so now then we have like a playground where we can like create different like sections okay and to interact with the application okay so that's pretty much uh it for the features overview okay so all right so now let's try to see how to use this particular application suppose I just click on like new okay then you need to actually select an agent workflow to begin okay you have all workflows here or you could simply scroll down so for now I could simply like pick General agent workflow that would be fine and then click all like create okay so here is my session you can see right over here it's pretty much uh similar to like what you see inside of like chat gbt UI so it's looking like really cool so you can see right over here that they have given you like several examples like stock price in which you could like compare Apple versus Microsoft something like that then we have sine wave marck down and paint as well where you could like simply paint different different images okay all right so if we just try to give you an example of like paint if I just click on like paint all right so you can see right over here that I'm getting an error like error occurred while processing message Api key is not present in llm config or open AI key environment variable for agent primary assistant okay update your workflow to provide an API key to use the llm let me just grab my API key once again so I would just go to like build tab okay so instead of models if I just click on this thing I need to provide in my API key right here then I need to Simply Save this okay to resolve this error what you need to do you just simply need to go to your workflows and inside of like General agent workflow where you have like the primary assistant okay this one is the user proxy we don't need to change it instead of primary assistant and inside of like system message what you need to do you just need to write in this l at the end set open aior API uncore key that would be equals to your API key Okay click on okay and if you just go back to like playground and now if you just try to click on like paint now you can see that I'm not getting that error okay let's see whether it will generate the stuff or not okay all right so you can see here that I just like ran into an error of like you know the r limit of Dolly 3 okay because I am on the free tier account if I just show you in the settings and if I just go to limit so I am on the free tier account and it has the limit for like Dolly 3 image generation so that's why it wasn't painting the images so then I just gave it a prompt like write a blog on effects of generative AI so it just provided me with the response okay user proxy agent just like said to the primary assistant like right up log generative Ai and primary assistant like came back with a response in this particular way okay all right so if you want to like Implement like more use cases so what you can do there is a GPD for that purpose okay so if we just show you this is like autogen Builder it's like great it says like I will transform your ideas into as many agents as required okay all right so I set it like I want an application in which which I can provide a YouTube link and it can transcribe and then can write an article on that and I'm saying that help me create skills models agents and workflows for that let's see what it will respond in this particular scenario so all right it just gave me an overview like what I need to do in this particular scenario okay so let's just try to like implement this thing all right so first of all I need to create a skill okay so here is a skill which I know is working like fetch Prof profile so what I'll do I'll simply can grab this code okay then I can come back to like chat GPT I could like give it this code okay so I could say like here is an example script make something similar but instead make it accept a YouTube url as a parameter it should go and get the transcript from the URL via YouTube API and then return the transcript okay I said like make sure to get the YouTube API directly I don't want to use Google Cloud method okay so let's see how it will perform in this particular scenario it provided me with the same okay it says like your YouTube API key should go in here okay so I don't want that so I could say like I don't want to enter API key use the method that uses YouTube API directly okay let's see yeah I actually want this method which actually installs like YouTube transcript API so it just gave me the code okay so I'll simply copy this code and come back to my autogen okay and I will simply create a new skill and I will call it as YouTube transcription and I'll paste my code in right here okay and then I'll simply click on like save okay here you can see that my skill is here okay so now we already have our model set up with an API key we also have like skill ready now we need to go to like agents okay so now I need actually two agents okay first agent what it will go it will take in like the URL of a YouTube video and bring back the transcript okay other one will take in that transcript and convert into an article for me okay so I'll create a new agent which would be like transcript extractor okay the description would be like take the YouTube URL and like returns back the transcript okay Max consecutive UT reply will be like default human input mode would be like never that's fine so in inside of system message what we can say let me just uh click on okay and see what I call my skill that's actually YouTube unor transcription okay so I'll come back to my agent and I will say like you an AI agent that uses YouTube unor transcription skill to get YouTube transcript for further processing okay so you will be using this model if you want to add another you can do that as well okay so here we we need to add our skill which is like YouTube transcription okay then click on like add skill and my skill is added here okay so now we have like a transcript extractor agent ready okay so now we actually need another agent so I will call it as like article writer okay so the description would be like take YouTube transcript and convert it into article okay Max consecutive auto reply it's fine human input mode would be fine as will never so for system message what I could say like you are an intelligent and skillful article writer who can convert raw YouTube transcripts into an amazing article okay and we do not need skill for this particular scenario and the model will be the same as before that I can simply click on okay now we have two agents transcript extractor and article writer okay so now if we go to like workflows now we can simply create a new workflow now we need to like choose group chat because we will have three agents okay first of all we'll have like the user proxy then the transp extractor and then the article writer okay so if we just click on like workflow then group chat so here what we need to do so I could say like YouTube url to article okay so the description would be like take a YouTube url extract the transcript of the video and then transform that transcript into an article okay a summary method cost is fine sender would be user proxy that's pretty much okay receiver would be like group chat assistant but we need to fill in the details of that okay so inside of like group chat agents so I'll simply get rid of both of these like primary assistants okay and I'll add like transcript extractor and like the article writer okay name could be the same that's fine so now inside of like system message what I can say I can simply make like it more disc descriptive okay because here you want to like tell your uh like workflow like what to do so I would say like you are a helpful assistant at coordinating a group of other assistants to solve a problem the problem to solve is taking a YouTube url okay make an assistant to use transcript extractor skill to extract the transcript from the YouTube video pass the transcript to an article writer have an article writer create an article based on that transcript okay then I guess we also need to like uh add the open aior API uncore key okay that we saw that it is actually necessary to do that okay I don't think we need to give it a skill because the agent already had the skill okay so I'll simply click like okay uh just make sure that the group chat agents are correctly set up okay so again I'll simply click on okay so now our workflow is successfully being created all right so I think that should be it okay that should work let me just uh close this session okay and I'll simply click on like new session and I'll take it like YouTube url to article Okay click on create now my session has successfully been created so all right I'm saying like take this YouTube url okay this is like one of the videos of our Channel use YouTube transcription skill to get the transcript of the video and then create an article based on that transcript okay let me just try to send this thing all right so now it's working you can see right over here it's installing like YouTube transcript API uh everything is like automated okay so you don't need to worry much about that so all right you can see here that it just came back with a response okay so this was my YouTube url if I just tried to show you the uh like video which is here like this was the YouTube video like create an AI s song and music video for free Sono Ai and dolly3 and paa okay I just gave this like prompt like take this YouTube URL and YouTube transcription skill to get the transcript of the video okay and then create an article based on that transcript first of all you can see that it says like we have successfully retrieved the transcript now for step two I'll convert the transcript into an article format so creating an AI music step by step first of all it gives like genesis of AI song okay and then crafting the visuals with AI that animating the imagery that editing and finalizing the video and then this is the result okay so it's like really great so here is our complete article now with this tool like autogen the possibilities are endless so you just need to create the skills models agents and workflows to make this application like work all like endless possibilities okay so I hope you like this video and that Ms the nend of our this autogen studio tutorial so I will catch you up in some other great amazing tutorial so subscribe to our Channel and make sure to click the Bell icon so you never miss any update from our Channel all right take care good day bye

Channel: SkillCurb

Views: 3,348

Keywords: autogen, autogen studio, chatgpt, gpt, gpt 4, gpt4, microsoft, coding, tutorial, open source, llm, ai, prompt engineering, Prompt Engineer, AutoGen, Autogen studio, autogen 2.0 tutorial, ai agents

Id: MUhRP8QCb9A

Length: 19min 49sec (1189 seconds)

Published: Fri Feb 02 2024