How to LangChain: LangFlow VS Flowise!! WHICH IS BEST?!

hello everybody in today's video we're going to check out and compare uh the two similar but the very awesome tools that a lot of people have been using recently which is flow wise and Lang flow I will show you both of them and we'll go through some of the ways that we can install and get things running so let's just run through the setup of both of them very quick and then we'll compare some of the features and see what we can go from there so hopefully stick around because we'll we'll check out both of them and just a quick preview here is what flowwise looks like and here's what Lang flow looks like and let's get through it so starting with flow wise the the uh um the one that I think uh I'm more familiar with is in order to get it installed I use render okay so we go to render and the of course as a my previous video has already described this one we go to render we start a new web service we select we connect ourselves first of all to GitHub here and there's my GitHub that testsavon and uh we connect the GitHub and then we uh connect here for example uh flowwise we give ourselves a uh a name that we wanna and then here for runtime we select we select node go down here and the only thing that we we add is yarn install yarn and install um forget that okay and everything else stays the same and you can check that out actually in the repository itself it will show you that the the build commands will be yarn install your build and yarn start and that's where we include them we select the starter as opposed to the free which will erase itself every 15 minutes if you don't use it scroll down here and create web service it's going to run run run run for a couple of minutes and then it's going to create something that looks like this here which is our flow wise link to an actual website that allows us to play around with flowwise and create flows and and then and then chat and and basically create the API that we can then include in our code so that's how we install flow wise you can also do it locally but I was having some issues with connecting into language models after installing it locally and this was the better version and then in order to install length flow in in my case Lang flow over here I was having issues installing in render because the runtime is in Python and for some reason I was having issues and if you guys anybody knows like if we're going to Lane flow and we take a look at the at the logs when I run a deploy the latest commit it gives me so it'll give me an error and you guys can check it out it might be that's because I'm using an older python version but the build fails because it's not able to find the matching distribution for Lang flow and so uh my settings if you want to see are over here and that's using pip install link flow and then python M link flow and you can get all this from the actual link flow repository so on GitHub you go to this Ling gig Hub log space Dash AI length flow you clone or Fork the repository into your own account let's check it out and this is link flow so I'm sorry this is uh yeah uh this is uh yeah Lang flow get the names confused and so you can see the instructions to pip install and this is how you get this working locally I was attempting to do so on render couldn't get it to work but I did get it to work locally uh and luckily we got it to go in the actual vs code terminal so what I do is I'll start from scratch is that what I do is I create a virtual environment right so but before that what we run is We Run The git clone command for Lane flow so if we were to go to the length flow and in this case if I when I fork over to my account I can then go to code and let's do that my repos so we're gonna go to Lang flow and Link flow we go to code uh okay stand by okay go to code go to local we get this link here we copy go back into vs do git clone there's our link it's going to download after download we change directory CD into Lang flow okay once we're in link flow we're going to run command python-mv NV virtual environment and then we're going to call that virtual environment link flow you can call it anything you want but that's the command structure is Dash and python-em virtual environment and then the name of the virtual environment that will create it then the next thing what we do since I already have it created let's run it I don't think there's any harm in that it does that it doesn't seem like it does very much but it will do it in the background okay so then we go virtual environment Lang flow okay so virtual environment Lang flow we're now not yet that actually activates the virtual environment and then we can do the dot slash Lang flow scripts activate okay and now take a look you can see that we changed the way that the uh the command line Looks Like It Starts Now with link flow and now we're in a virtual environment where we're isolated from your regular development environment and anything with screw up here is only going to be screwed up in this in this isolated place and then from here we do pip install link flow okay which is uh something that I've already done and then we launch Lang flow okay once you hit launch just you just do Lang flow just type in Lang flow and you hear of here you got UV corn uvi corn running on local 127 so the Loop pack with a this port and then if you click control and then click boom you now have Lane flow up and running now let's take a look let's take a quick look as to what's going on here now first of all Lang flow Lang flow as you will see if you start looking through the documentation is a GUI 4 Lang chain designed with react flow what's react flow well react flow is the base application that both flow wise and land flow are built on and you can come down and you can build your own which is actually incredible thing to just to get a real good grip on because I I can see a lot of potential I can see actually in the future a lot of potential of having whole uh templates here that we design in Lang flow and flow wise have the separate application or separate way where we can just have a a um a tile for the entire template right so create whole templates and then isolate them somehow in an individual tile and I wouldn't be surprised if either Lang flow or flow wise actually come up with some of this in in the very near future so yes uh that link flow and flow wise are both built under react flow okay and react flow is a pretty awesome thing to check out and then when you go to Lang flow so when you start looking at installations and how you can install it not only can you install it locally like I just showed you you can also install it in the Google Cloud it's a little bit more complex simply because Google Cloud platform or the is well it's like if you're a drone pilot and you are asked to fly in the NASA space shuttle I mean you go into these uh these shells and uh you're just basically it it's just a lot of buttons and the a lot of stuff to uh to to know and figure out all possible all awesome stuff to to learn uh just not right now for this session so we're gonna get the heck out of there but what I wanted to show you guys is let's leave is that we can also deploy um Gina AI Cloud so that's another option okay and Gina AI cloud is actually pretty interesting it's there's Lane chain apps on production with Gina and fast API so you can Lang chain serve under Gene AI is actually a way for us to launch Lang chain template based applications into production and so super cool very powerful check this out as well and now Gina itself if you take a look is a really awesome streamline Ai and machine learning product delivery so we create prototypes we wrap them in a service and then we deploy those servers to the cloud I think it uses the Amazon I think it's AWS in the background but this is uh open source and also very very interesting got the collab notebooks here show you how to do all the stuff all the code samples very cool very nerdy definitely worth checking out so Lang flow wise who is the winner well let's see what's going on so when we start out in the initial application as you can see we start out in uh flow wise we start out in this menu where we got the chat flows we got the marketplace and we got the API keys in order for us to uh to play around with the all of these things you can open close and you can uh come in here and uh change everything to dark mode after dark mode we also have a loaded load database and Export database functionality here in Lang flow a slightly differently as we come in we see this as the very first thing now it's pretty cool we also have the ability to turn into dark and then dark you have both an ability like I think I like this better in certain ways and I like it worse in other ways because you can see certain things better and certain things worse so we start out with a completely new tab so uh unlike flow wise where we actually go into the flow itself into the template after selecting template right so we either pick one from the marketplace here which is rebuilt once or we pick the ones that we've already worked with and already saved so if we pick the one that we'll talk with a new website here it looks kind of like that and here what we have to do is we would have to import and then we either import from local files that we saved earlier or these examples over here now we'll go through them real quick uh they actually have some pretty cool examples that are a little bit different if we want to compare the the two well let's go in here when the marketplace so check it out Auto GPT baby AGI chat GPT plugin conversational retrieval metadata filter load so these are the examples in flow wise and here are the examples in Lang flow and you can see some interesting things that are are kind of different is the pal math tool here multiple Vector stores that one was in these one we'll look into as you can see it's a fairly fairly intricate one not too complex but it allows us to separate into several local Vector stores so that's the uh that's the difference between flow wise and Lang flow on that case and you see how every time we pick a new tab gets created then we can create new tabs up on here which I kind of like that's that's pretty neat an easy way to navigate also what I do like is this components bar unlike in flow wise if it were to go and open up our tab so in flow wise in order to get to the component bar we have this bar here and we have to push plus minus and also goes away after we pick something and drag it back in here it goes away and it forces me to that to go in and start clicking again so I don't really like that part and in Lang flow all of the components are very easily accessible here from from the sidebar now in the sidebar you can see we have pretty much the same components and let's take a look at some of the differences that we can see so we have agents chains chat models document loaders embeddings LMS memory prompts text Splitters tools and Vector stores pretty much almost exactly the same thing agents change loaders embeddings LMS memories prompts text splitter toolkits tools so it's a little bit of a difference there utilities that's an uh I think we got that as well Vector stores and then there's this wrappers and the wrappers has a text request wrapper which 100 sure how to to apply but you can see it kind of LightWave wrapper around request Library but interesting so let's also dive in a little bit deeper into the actual tiles all right well let's continue apologies let's continue on here with this menu if we were to take a look at the some of the differences if we're let's say for example agents right if we go into agents the agents have pretty much similar thing we got CSV agent Json SQL Vector store Vector store router zero shot and initialize agent and here we have a more specified as Auto GPT baby AGI conversational uh uses react framework to decide what actions to take and [Music] mrkl agent for llm so a little bit different right so you a little bit different you can see how these agents are um let's take a look these are more wrapped already like Auto GPT and baby AGI that's a pretty much complete package when it comes to these agents and here if we have a conversational agent it lets you to have allowed tools language models and memory and is there any equivalent here uh Vector store um no it doesn't look like it's it would be here so interesting let's go on uh the chains let's see how the chains are different uh let's pop this open see what I mean like every time about needing to do this so chains we got a conversational uh conversation chain conversational retrieval QA chain so this is basically a a one shot you know like what's in there give me a summary and this one is a question answer where it grabs your response or the response from the AI to your original query and wraps it up into your the next query uh llm chain retrieval QA chain SQL database chain and the vector DB chain so uh in this one conversation chain llm llm Checker chain math chain let's take a look at what the Checker chain is and as you can see the interface is fairly similar uh the only difference is that this is the maximum zoom out on Lang flow where flow wise allows us to zoom out really really far so I guess in theory creating these huge canvases and Lang flow not so much which you can see actually when we start to look at like importing examples and I think a multiple store one because we're right now at maximum zoom out there there's this need to have to overlay tiles over one another so it doesn't look as good okay so let's see let's carry on we got loader so this is what I see that Lang float has a wider selection and and somewhat of a different selection more um I'd say more specific and more in-depth selection like certainly when it comes to loaders they got a lot of loaders here all right if you go to uh flow wise and close up the chains um they call yeah document loaders so the document loaders right here we only have uh what do we got like five eight nine yeah we got web Square I mean we got the important ones is like a web scraper the uh Excel files the document Google Docs uh folder the entire you know folder GitHub speak to code kind of stuff uh Json the notion the database now platform and just regular PDF text file but in Lang flow as you can see we've got a whole bunch of interesting one including like CSV there's a College Confidential one not sure where that goes Facebook Chat loader Gutenberg loader so interesting they have the pi PDF loader so it's it's the python more native python one uh that's cool I like that when we get to embeddings though so you see here are where we start to notice like differences that are actually in favor of uh a flow wise start to open up embeddings and we have a couple of more options here we have outside of open the AI like we do in Lane flow we have the hugging face cohere and and Azure open AI whereas link flow only has the open AI embeddings so but when it comes to llm Lang flow has those four and those four hugging face llama and open Ai and here flowwise has hugging face open Ai cohere and Azure so same as the embeddings memory we've got just like a buffer memory and memory we have a couple of actual memories here conversation buffer memory conversation kgm memory not sure what that is Knowledge Graph memory for storing conversation interesting we'll uh we'll check that out by the way just to look at the the actual tile difference here if we were to look at the tiles in flow wise so what we have is a duplicate we got a way to delete this whole thing and like for example the chat open the AI one right we have the ability to put in the API key here select our models and our model here is actually pretty good pretty good model selection GPT look it's all of them gpt4 uh the the the lock 314 version and the 32k version as well which is really cool because uh yeah it's pretty much similar to here we have the 3.5 gpt4 and GPT for 32k so overall you have uh just a bit more here not really anything significant like this one and this one isn't really needed so it's really just a turbo gpt4 and the 432k so those are the same um the tiles themselves I kind of like the tiles here so there's this uh this question where this exclamation point where it'll actually give you the code that is being reference um so if we were to look at the some of the some of the examples here so let's see getting started basic chatbot you take a look at and hover over the green check mark here it actually gives you the full code or at least the the the the the full syntax of all of the different variables that can be converted here so it's actually really really neat and then when um so yeah when you when you're wearing you need to actually dive in a little bit more into the the technical things or on the technical side of things this is this is a very neat feature here that I doesn't exist in flow wise so the tiles themselves are a little bit more intuitive and they have a bit more of an explanation here then you can click on the um uh the settings icon there and now we can do verbose output key input key so interesting options here that do not exist in flowwise however in flowwise in certain cases let's see if we uh chat open AI here chat open AI uh we can come in here and click on additional parameters and specify more things in flow wise and then link flow looks like most of that is already here and then when we go to the settings that's when we can kind of start to maybe play around with the keyword arguments here so very interesting uh I think we're probably going to just wrap it up there against just open these things up for you take a look and see what it looks like in flow wise flowwise does have a good amount of tools here that can be used and it does have Vector stores there's another difference that you can see is that chroma these are the local Vector stores this is a straight up in memory it has pine cone and Super Bass and weaviate now in Lang flow if we were to go down to Vector stores all we would see would be the chroma local Vector store so still I mean a good option but not as good as pine cone in my case I I like Pinecone a lot and use it a lot so that would be really awesome addition tools got a a big selection of tools here there's actually another tool kit as well so open AI toolkit toolkit for interacting with openai API um it looks like with Lang flow there are a lot more possibilities which is very interesting so I'm going to play around more with that and and if you guys like this video please like And subscribe and let me know what you think and what you'd like to see a bit more when it comes to Lang flow or flow wise just to finish off I'll pop open these uh these other tabs here so you guys can you know take a look pause on the video and see if you want to see anything um then the last one is this wrapper if anybody can let me know what this means uh if you know what exactly the function of this one is leave a comment that would be awesome so text request wrapper uh then after uh we can save the stuff I like to save export as is a little bit better here in Lang flow getting started basic chatbot we can get the name the description uh save with my API Keys download flow so the interface is a little bit cleaner and nicer and then there is the actual same thing like the the Lang flow chat and it's a pretty cool little chat bot here that's kind of neat and then the best part of course is the code so just similar to flow wise here we can grab the code but flowwise as you see has more options when it comes to embedding the website or uses API okay we can embed we can use python JavaScript curl we can add the the authorization as well so basically the key which I will now have to delete because you internet boys love to rack that up um there's this other option here for show input config and actually kind of detail what is going to be going in this in this model and then a little bit more detail as far as the actual what's being submitted so the code part is a little bit more detailed a little bit more intuitive but how it differs here is that we have the python API but also a tab for python code so you can you can then use this actual this this diagram that you set up as a lang flow just to import Lang flow from Json and reference this particular template if you export the template and if the template is sitting somewhere in the uh in the local local file or a local drive for it to be using so check it out uh play around with it let me know what you guys like to see I'm going to make uh probably a bit more videos about how to actually use some of this new functionality here and see what we can do that we can do with low wise earlier but um I hope this was informative and if it was check out some of the other videos where I go into more detail on how to install flow wise and and how to play around with these these tiles so thanks very much and I appreciate you all

Channel: TestSavant AI

Views: 9,548

Keywords: ArtificialIntelligence, AIethics, AGIrisks, FutureofAI, ThoughtExperiment, TechnologyDebate, EthicalAI, AIfuture, langchain, flowise, openai, chatbot, pinecone, langflow, machine learning

Id: OLuqTPofJ9g

Length: 30min 10sec (1810 seconds)

Published: Fri May 26 2023