Chaindesk: Setting Up No-Code Custom ChatGPT Agents

--- type: transcript date: 2023-04-27 youtube_id: QsjFZe2Tk-I --- # Transcript: Chaindesk: Setting Up No-Code Custom ChatGPT Agents all right welcome back so in this video I'm going to be showing you databerry so databerry is a no code Solution on how you can connect your data to large language models so what do I mean by that so databerry will take your data and it offers the ability to embed data in a variety of formats from websites to PDFs to text files and once you've embedded the data that you want to have within your data store you'll be able to have a conversation with it so think chat GPT but it has that guided and potentially more relevant and guided information that you've given it based on the data set that you feed it so let's just say a hypothetical scenario is say you're a company and you want to have a chat bot that is integrated into your website now if you were to let's just say implement the chat GPT uh API or one of the llms from open AI now those llms are trained on data prior to 2021. some earlier and at the time of recording there might be some that's you know within a range of that but the the big thing to note with them is they're trained with a lag on today's date so let's say you ask Chachi PT about your company it might say it only has information up to that date so if you want to have more relevant information and up-to-date information or maybe even information that Chad GPT didn't catch because maybe it wasn't publicly available or within their data set that they scraped to make the model this is how you can integrate that so I'm going to be going in and showing you how to actually set up and use databerry a little bit I'm going to come up with a pretty a simple example what I'm going to be doing here is I'm going to be showing you an example with a recent blog post so I want to be able to interact with this blog post ask questions of the blog post and just sort of show you a simple example on how you can set this up and and the ease of use on how easy it is to set up so if I go into agent let's just say I want to let's call it example data store for this one I'll go in and I'll click save for now and then I'll head over to my data store now once I have the data store I'm going to name it the same thing just for continuity I'm not going to make it public right now but as soon as I get here so I have the options so we have the option to choose text web page we can crawl the web page but this is a premium plug-in that they have I think it's 25 U.S a month if you want to use this one and then they have the files that you can upload PDF CSV Json Etc and then coming soon you have notion so in this case I'm going to use a web page let's say and I'm going to head over to the openai website so if I just let's just use this tab here open AI and then let's go to their blog so I'm actually going to use this blog post here um or actually let's say I believe their blog post might be at the bottom of this article gpt4 let's say we want to interact with this blog post so gpd4 came out before GPD 3.5 was it was trained and it goes without saying that the llm wouldn't actually have a context about this particular article so if I go ahead and name it I'll say GPT for info and then I paste in the URL here you'll see that it takes just a moment here and once I have it set up I'll be able to link the data store here so once I have it linked I'll just quickly link it here I'll say choose save and then from there I'll go into chat now let's actually look at something within this article that we could ask it that's something like chat GPD might not get maybe we go in and we say What are the percentiles for simulated exams in chai gpt4 let's say foreign so what we've done so with the data store they use the text embeddings to API from openai and that will embed the data that you provided so once you embed the data it will be able to quickly provide you a result based on that information so if I look at the answer here so it says the simulated scores for gpt4 are all listed in a table on the open AI website according to the table gpd4 achieved high scores and multiple exams including etc etc I find more information on the technical report can you be more specific let's say so obviously you can use this on a host of different things I'll go into a couple of their features within databerry but you can get a sense right off the bat on how simple this was to set up and you can sort of get your gears turning in terms of how you might apply this in your your own application or use case so while this is running I'm just going to hop over to a different tab here for a moment so if I go quickly back into the interface I just wanted to quickly touch on the apps that they have available so this is the way that you can train your model recursively on a website so instead of just providing it a link say you want it to essentially crawl your website it will go in and collect all the data sources that are available then there is also the slack bot and this is similarly a premium version where you'll you'll pay a little bit more to integrate something like this so say you're onboarding a new employee and there's a particular Channel and you want to have a bot that responds to questions they might have this could be a potential use case for something like that but I just wanted to point out the apps that they have available right now so if I go back here we say okay I'm sorry I'm not sure what you're asking could you be provide more context or clarify your question so let's say okay let's say what is the SAT or okay so we'll again just give this a moment and you see the score for the SAT evidence uh based reading and writing is 710 out of 800 which ranks 93rd percentile so if I go back here we'll see 710 out of 800 and it is showing the 93rd percentile so pretty easy to set up right so you go and create your agent create your data store and you're sort of Off to the Races now if I dig into both of these a little bit more and just go into the configuration between these So within our data store we have a handful of things so we have the AI plug-in Json we have this yaml file there's not as much within the data store that you're going to be tweaking once you have it set up but if you head over to the agents this is where it sort of get gets a bit more interesting in my opinion so if we head down here and you scrolled down you'll see that there's an agent ID where you can query the databerry API so say you want to use your own solution or maybe you have a chatbot interface that you've built and you just want to Leverage The databerry API they have that available integrated into the service additionally if you just want to go down here and integrate the chat bot directly like you see here in your website it gives you the ability to do some you know brand styling the positioning and then also it will give you the embed code to put this on your website so I'm not going to go into too much more in this video I just more wanted to introduce databerry I'm very impressed with what I've seen already I'm going to be diving into actually setting this up locally and if there's a need or want for it I'll create a video in depth on how you can also do this and deploy it yourself but off the bat it's great work to the contributors here on what they've built I think given what it can do for individuals or companies it is uh pretty competitively priced as well if I go back to the pricing page here um and yeah it's interesting to see all these different AI apps that are coming out um and yeah so if you found this video useful please like comment share subscribe and until the next one