Auto-GPT Tutorial - Create Your Personal AI Assistant 🦾

Key Takeaways

welcome back would you like me to return to being Dave's AI assistant okay so chat GPT is old news Auto GPT is the new thing and if you want to know how to use it how to get started then this is the video for you I will give you a full tutorial showing you how to install it get started and then create your own personal AI assistance and you can even use voice clones to let it talk as well make sure to like this video And subscribe to the channel okay so Auto GPT is an open source project and this is the GitHub page so what is it well in their own words it's an experimental open source application showcasing the capabilities of the gpt4 language model and basically the power is in chaining together large language model thoughts to autonomously achieve whatever goal you set so you can give it a set of goals and then it will start to use the gpt4 language model using the open AI API to iteratively prompt the API get results back and then use that again to evaluate it and and then come up with the next prompt to reach all the goals you set for it so what can you do with it and why should you experiment with it well first of all I think this is the future of AI we will all have a personal AI assistant that can basically do tasks for us so instead of right now we have to go to chatgpt fill it in and then it will help us what you can do in the future and auto GPT isn't already nice example of what it will be like in the future if you give it some a set of goals and then it will go out on its own to achieve to reach those goals that you specify so here in the demo they give the example of Jeff GPT which is an AI designed to browse the web to discover the next upcoming event and event a unique and original recipe that would suit it so again this is not possible with chatgpt because it does not have access to current or Live Events so here are the goals event and original out of the box recipe save the resulting recipe to a file so then it goes on with the thoughts the reasoning and then it can execute commands so here you can see it's working in the background to basically go out on its own explore the web save the results and this is really really cool but also quite scary right so it is an autonomous agent and in between you have to basically say yes you can continue but other than that it is doing things in the background without you actually knowing what's really going on okay so what do you need to actually get started well first of all you need a python installation on your system so python 3.8 or later and I will link this document this YouTube page in the description and basically everything that you need is in this document but I will walk you through it in this video so you also have a graphical representation of what it looks like and just watch me do all the steps basically but make sure you have a python installation on my system I have an anaconda installation so I will use that also to create a kunda in environment and if you are totally new to python or anaconda then I would recommend installing Anaconda because that is what I will show in this video and then you can follow along if you already have a python installation you can just use a virtual environment alright so first we need python then you need an open API API key so if you follow the link over here I will go to my account I already have an open AI account so you need that and you also need an API key and I believe that right now you have to get on the wait list to request an API so if you follow this link go to API Keys then there you will find the next step to request your API Keys get on the waitlist alright It also says here that you need a Pinecone API key but I believe this is actually not true because you can also use local storage so Pinecone is a long-term memory for AI it's basically a vector database and I won't get into the specifics right now but you can just skip it because we will use local storage for this example and then optional is an 11 laps key if you want the AI to speak so this is where we can actually give it a voice and get it to talk so 11 laps is a speech synthesis platform where you can basically clone your own voice or use pre-trained voices and I will show you how to use this later in this video because it's not required and I first want to get you up and running alright so let's get started so if you have get installed on your system and you are familiar with Git then you should clone this repository using the URL over here if you are not familiar with Git then download this ZIP to download basically all the files that are here so I will show you how to set this up on Mac OS but the principles basically are the same if you are on windows so I created a new folder called Auto GPT experiments and this is where we want to clone the repository to or if you have downloaded the zip this is where we want to place the files so I'm going to open up a terminal and I'm going to basically say git clone and paste the URL so this will clone into the repository creating a folder Auto GPT and in here will all the files be basically which you also see on negative repository so if you follow along using the zip download make sure you also have a folder Auto GPT with everything in here okay next we're going to create a python environment because there are some requirements in here that we have to install using pip and we want to do that in a new environment so like I've said I'll be doing this in a Kola environment but if you just use a regular python installation you can also use a python virtual environment again if you're not sure what that is or how to use it look it up use chat GPT to ask you how to do that so I'm going to create a conda environment called Auto GPT and I'm going to use Python version 3.8 so let's run this and wait for the environment to install alright so the installation is complete and now to validate it I'm going to activate Auto GPT and run python in here to see if everything is working correctly python 3.8 alright looking good okay so next I'm going to set up a visual studio code workspace and that is because PS gold is my IDE of choice but you can use any python editor basically to do this so I'm first going to CD into the auto GPT folder then make sure that we are in this folder within our terminal and then I'm going to type code followed by a DOT now depending on what you use on your system this might not work so again you just open up this directory within a software within a app that can edit python files so in here I'm going to save this workspace and I'm going to save it as Auto GPT so that is all good and now I'm basically in here and here I can browse through all the files and I can also I can basically manage my project over here alright so that was the vs code project and now we're going to install the requirements and we can can do that by running pip install Dash R requirements Dot txt and before running this command really make sure that the environment is active so here you can see auto GPT meaning that I'm in the auto GPT environment and whatever I will install in here is installed within this environment so I'm going to run this this will basically pip install everything that is in here alright and it's finished and now it's time to update the dot and file which there is a template file in the main directory that we've just downloaded and here you can see all the placeholders basically so it's the dot end dot template file and if we come back to the GitHub over here you can see what it is that you have to do so you have to rename it to dot NF so first step is basically to come over here rename it and remove dot template so this is the first step and now we have to fill in all the API keys that you want to use so again like Pinecone is not necessary this we can all leave default we should update the open API API key you need that one then over here you have the 11 laps API keys and voice IDs if you want to use voice and then here make sure to check for example if you're on Mac which I am set this to true and here you can also specify so the memory backend is currently set to local and we can also set that to Pine Cone sorry pine cone but for now we leave that at local so for your open API API key what it is that you have to do and I'm not going to show you my key that I'm using but you are going to create a new key then you can see the key over here so I will delete this in a bit and then so you can copy this then come over here and then here you basically paste it and then make sure to put it between quotes So This is how you save the API key and now it's in here and now the other python files can load this key from your environment variable and it will use that to connect to the API so fill in your own API over here I will now make sure to delete this one so you cannot use open AI on my credits and then basically fill this in set everything up all right and that's it for the setup you are now ready to start using your own personal AI assistant but please keep in mind that the open API to use gpt4 and the other other models is a paid API so you have to set up a billing method and open API will charge you for this now don't worry it's pretty cheap to start experimenting with this so you can look at open api's pricing but it's a couple of cents per request usually so I've been using it this whole morning and for me it came down to 1.84 cents so please keep that in mind and you can also set up restrictions so you can set the hard limit or a soft limit so you can basically limit it and be in control of how much money you spend but please keep that in mind it is a paid API alright so now let's get started I'm going to clear up this terminal over here make sure that we are in the main directory of this whole project that you are in the environment and it's activated so the auto GPT if you followed along and now we can actually start to run and interact with the agent by running Python scripts slash main Dot buy which will basically execute the main.bi file within the scripts folder so coming back to the terminal I'm going to type Python scripts slash main dot Pi enter this will fire up the agent if everything is correct and installed within your environment and there we go welcome to Auto GPT enter the name of your AI in its role below okay so I've been experimenting with this this morning and I got varying results and right now I'm going to try something completely new so I have no idea what the results are going to be like so normally on this YouTube channel I make videos about data science so about python about machine learning basically videos to help you become a better data scientist to help you better work with data and that requires me doing some research creating tutorials finding data sets and I would like some help with that so I'm going to create a personal assistant that can create python coding tutorials for me and I'm going to call this AI Dave AI assistant all right at your surface and now we have to describe the ai's role so Dave's AI assistant is an autonomous agent that can create python coding tutorials to help Dave with his YouTube videos alright and now we can specify up to five goals for the agent and I'm going to start with a tutorial for the most used python library for data science which is pandas so first go create a tutorial for the pandas Library start with the basics and work up to Advanced Techniques alright goal number two save the tutorial to a markdown file and then three shut down alright and then if you don't specify five goals you can just hit enter again alright and it's giving me a quick error that the file Auto gpd.json does not exist so it cannot store the file so I'm going to quickly fix this so this might be a bug so they are constantly updating this repository so I'm first going to exit out of the program by typing in n and then I'm going to quickly create a auto what was again we need a auto gpt.json so rename this file Auto gpt.json and this should be an empty Json file which the script will append the information to so coming back to the terminal I'm going to re-run the script and then it will ask me to continue with Dave's AI assistant and I will say yes alright and here it goes so it starts off by saying I think a good place to start would be to perform a Google search for a brief introduction to the pandas library then it provides the reasoning and a plan so it will look on Google introduction to pandas and then I basically have to confirm it by typing a y and then hitting enter so that way you authorize the script to continue and now it will perform this Google Search and now we wait for the results alright so now it looks like it will read through everything and I'm just going to continue accepting the commands that it wants to execute right seems like like it wants to browse another website I'm going to press y again alright so it looks like it's doing some research and what you can also do instead of typing why every time you can also type in y and then Dash and then a number to basically authorize the next and prompt now be careful with this because it is going to execute on those prompts without any human intervention but I've been playing with this tool this morning and I am quite confident with with the setup like this I'm okay with giving it some steps ahead so I'm going to say Y and then dash 10 which basically means I'm going to authorize the next 10 steps so now it's going to run basically on autopilot for a moment okay so the script was kind of stuck in a loop so I slightly change the AI goals so I made them a little easier and be overall a bit more specific so I just had created tutorial for the panels library in markdown format save the results to markdown file shutdown so you can set this up in the AI Ai settings.yaml and then basically re-run the script so now you can see it's using the command right to file upon this tutorial.md and it's still going on over here you can also see it's using a read file command so let's see what we we have a file bundest tutorials.md okay what's in here so we have some basic markdown on those Library tutorial introduction we have a series then we talk about the data frame and a conclusion okay this is a very brief very basic but hey it's it's a start so let's see what this looks like if we render it in a markdown viewer copy and paste all of this into a craft document over here and then take the shared link and see what that looks like in the browser all right and we have a pandas tutorial alright so it covers a series and it covers data frames and we have a conclusion so hey it's like I said it's very basic but it's a start these are literally the documents that I use to prepare my YouTube videos so if an AI assistant can help me with this and this gets better then that will make my life so much easier but now let's actually come back to our script because what it's doing over here okay so we have bonus tutorial tests okay it's it's still going oh we have a new version pandas tutorials refined so serious data frames and we have some tips let's see what what's the difference over here so come back over here just gonna replace everything now we have some tips here in as well so let's see if I refresh this document we have some bonus tips okay so that is the refined one now let's see what we have in tests oh wow we have some examples right so let's copy paste this come back to the document paste the tests in here let's see what we got all right let's reload this alright and we have some examples in here test case example two test case oh wait it's trying to actually push this to GitHub so this is actually getting quite scary so as you can see it has the command execute shell and it can input an argument over here in the command line get init and then add and then it will just push it to a GitHub repository but as you can see it says username repository over here so it's it's not doing that correctly it's not yet linked correctly to my GitHub repository but just imagine once this is all hooked up you can basically just give it a few commands and it can basically code up a whole app or a whole tutorial for you push it to your GitHub repository manage it like what the this is really exciting but for now I'm going to cancel the script over here I think this was a nice experiment so coming back to artificial Studio code project we have a bonus tutorials upon those tutorials refined and some tests and as you can see this is actual markdown that we can instantly copy paste into a document and can you for a tutorial and now of course this was a very specific example for my channel and an AI assistant that would be helpful to me but of course the possibilities here are endless so try to come up with ideas what can you use an AI assistant for and start to experiment with it and now like I've said you can make this even cooler by enabling speech mode to basically read out all the text that is generated and you can activate this using the speak argument but in order to do this you first have to set up an 11 laps key if you want the AI to speak so go to 11labs.io link will be in the description set up a free account if you come over to the subscriptions you can start with a free plan then you come over here to your profile and here you can view your API key so make it visible then copy and paste it into your EnV file just like you did with the open AI API key and then the next thing you have to do is you also have to set up the voice ID so if I come over to the GitHub page and look in the template just for illustration so you have your 11 live ski this is where you put the key and then you have the voices and you can get the ideas of the voices going to resources API and then you do the get voices you paste in your API key here basically by trying it out pasting the API key and then executing it and then you will get a result back with all the voice IDs that are available to you and then you just copy and paste them over here and now once that is ready you can start using the speak parameter so if I now come back to the terminal and then use the parameter speak it will enable voice mode and it will say everything out loud and the funny thing is I actually created a clone of my own voice using 11 laps by uploading some MP3 samples and I now have a voice clone of my own voice and it's actually pretty scary it's not 100 accurate but it's getting there and it's it's weird alright so now let's run this and listen to The Voice okay on Mac I got an error no module named appkit so I had to do a quick install of this Library over here so by opgc so after running the PIP install command I'm going to run the script again welcome back would you like me to return to being Dave's AI assistant I will now create a tutorial for the pandas Library it will include an introduction to the library that is so cool all right and that's it for this tutorial that's how you set up and use Auto gbt I'm really curious to know what you guys can come up with like this video And subscribe to Dave's Channel or else this AI will hunt you down