Tutorial 1-Getting Started With LangGraph- Building Stateful Multi AI Agents

Key Takeaways

hello guys so one of the most requested video uh by many of you was to start a series on developing multi agents using langra so finally I've started this particular Series in this video we are going to discuss about what is Lang graph why do you require langra along with this we'll also be getting started with one practical example how to probably go ahead and create your multi agents so if you're new to this Channel please make sure that you subscribe the channel and watch this video till the end because this entire video will be for 40 minutes you know somewhere around 35 to 40 minutes where I will be explaining each and everything so that you can also get started with Lang langra not only this uh I have planned lot of videos in this series so that where you where we discuss lot of end to-end projects how to could probably create chat Bots how you can use it in different different domains like customer support you know um let's say if you want to book your tickets if you want to probably do multiple task itself so all those things will be included in this particular series one more additional thing guys uh I've been uh bringing up some amazing courses in Udi so if you are interested to take up all those kind of courses it is an affordable courses just in 399 rupees you can probably check out in the link in the description of this particular video so yes let's go ahead and enjoy this particular series thank you hello guys in this video and in the upcoming series of video we are going to discuss about a new module in Lang chin which is called as Lang graph now Lang graph is called quite amazing because it actually helps you to build stateful multiactor application with llms it is used to create agent and multi-agent workflow so this is really important now there are lot of libraries a lot of Open Source libraries that are available which actually helps you to create multi-agent workflow but if I probably consider Lang graph it's quite amazing uh so in this entire video and in the upcoming series of video I will be discussing many things about Lang graph first of all the agenda of this particular video will be that we'll try to discuss what exactly is Lang graph why Lang graph and uh we'll see or we'll try to create a chatbot wherein we will get just started with langra where I will be showing you with respect to all the coding things right so all these things we will be covering and when we'll discuss about why Lang graph at that point of time you'll understand the importance of Lang graph and uh through this you know you'll also be able to cover or you'll be able to just compare with all the libraries open source libraries that are available to create multi- aai agents workflows right and I feel right now langra is quite amazing and as you all know Lang uh what Lang chain is also doing is that Lang chain is coming up with this own cloud platform through which you will be able to build your own uh agent multi-agent workflow uh using langra and it you can do it completely in a visualization way right and right now it it just supports for the Mac machine so so uh we'll be discussing all about that as we go ahead but right now let's go ahead and focus on first of all we'll go ahead and see the definition of Lang graph so if I talk about what is langra langra is a library for building stateful multi-actor application with llms used to create agent and multi- agent workflows compared to other llm Frameworks it offers the core benefit Cycles so this is really important controllability and persistence okay Lang graphs allows you to Define the flows that involves cycle essential for most agentic architecture differentiating it from dag based Solutions now if you have seen already my previous video I hope everybody knows about graph DB okay so what exactly is graph DB it probably helps you to create you know if I probably just talk about creating rag application using graph knowledge right so here specifically you have nodes you have relationship between the nodes uh like this you know and uh with the help of this kind of relationships you can actually build this graph knowledge okay and graph knowledge is specifically built right now even in Google search engine you'll be able to see graph knowledge and with the help of this graph knowledge you'll be able to develop some amazing rag application because the efficiency uh with respect to the graph knowledge search is quite amazing you know so graph knowledge has the combination of all the kind of search the hybrid search the graph knowledge search you know so whenever you have any information which is uh displayed or which is probably converted into a graph knowledge uh information then your conversation with this kind of data store right can be quite amazing so already I have discussed about this in our previous video but in this video we will try to talk about Lang graph and the main thing will be that how we have actually created this graph DB structure right in the form of graph knowledge similarly Lang graph is also offers those kind of functionality we can create the entire multi- aai agent workflow using this graph knowledge itself right so like how we have nodes relationship and all similarly in the Lang graph also we'll be able to do that okay so uh till now we have just discussed about the definition don't worry if you're not able to understand because once we start the practical application everything you'll be able to understand okay now let's talk about the second question the second most important question y langra graph y Lang graph okay now here I will be talking about some amazing benefits of using Lang graph in pointwise okay so let's go and talk about this amazing benefits of using Lang graph so number one okay number one is that it really simplifies the development simplifies development now when I say simplifies development what does this basically mean see whatever complexities usually whenever we develop any multi- AI agents right there are multiple things that is involved the State Management State Management State Management of the agent like at which stage at what work it needs to really do and agent coordination let's say one agent needs to communicate with any other agents right when we say agents we are basically talking about multi- AI agents right so this two things are very important you know here if I probably seen most of the libraries that I've actually used most of the open source libraries that I've used to probably develop multi- a agents this two information that we have that is State Management and agent coordination I found out a lot of difficulties with respect to managing this things right but with the help of Lang graph it really simplifies it now when we talk about State Management and a agent coordination what does this basically mean let's say that I'm using three to four multi- aai agents in my generative AI application One agent is probably doing some Wikipedia search so let's say I probably need I require agent one I require agent two I require agent three right so let's say all these specific agents are there right each and every agent probably needs to it does some specific task right let's say this will be doing a Google search this will be doing a Wikipedia search and this will probably do a vector search Vector DB search right and here is my chatbot which is specifically using all this multi- a agents now whenever we develop this kind of application it is very important that we have to do the state management efficiently for all the agents and also the agent coordination what does agent coordination basically means the communication between the communication between this kind of Agents let's say whenever a query is asked first of all we go Ahad and ask agent one hey whether you really need to do the Google search if you doing it you really need to provide that information to agent two then agent three will probably come into picture so this kind of communication is basically a kind of agent coordination okay so whenever we are developing this kind of things you know and there may be also scenario that when you're developing a use case the agent one response may be dependent on agent two uh response you know so agent two will only be able to work when they get the response from agent one I'm just giving some hypothetical scenarios okay now when we are developing this kind of chat Bots uh as a developer we we really need to Define workflows we need to Define workflows we need to Define Logics right without worrying right without worrying about so many different things right so that is what it actually does langra actually provides you this entire simply it simplifies this entire development process and it will actually help you to create an efficient multi- aai agents right so we'll be discussing more about it then we go ahead and discuss about the Practical implementation second thing is that uh why Lang graph you should use it second thing a very important point is related to flexibility okay so whenever we talk about flexibility I will just uh make sure to provide some description information over here so that you'll be able to refer this okay so with respect to flexibility you can see with langra develop developers have the flexibility to Define their own agent logic communication protocols this allows for highly customized application tailored to specific use cases whether you need a chatboard that can handle various types of user request or a multi-agent system that performs complex task langra provides the tools to build exactly what you need okay so it's all about giving you the power to create so uh with respect to flexibility as I said the kind of Logics that you can basically write will be quite amazing you'll be seeing that how Grand Man level Logics you can actually write with respect to each and every agents that you specifically create I'll be showing you as we go ahead with one very good practical example okay then the third thing is that if I talk about the third important Point why specifically we should use Lang graph uh it is related to scalability okay now whenever we talk about scalability here when we are building this multi-agent application we can actually build large scale multi-agent application large scale multi- agent applications now what what do we mean by large scale multi-agent application see the kind of robust architecture we basically use this large scale multi- agent we basically say this as lar scale multi- agent application because it can it can handle it can handle high volume of interaction high volume of interaction right the interaction can be between agents and it also has a really amazing or complex workflows and you'll be seeing when we Define about complex workflows it it it will be very simple to develop this entire workflow itself right and even Lang chain is coming up with a feature and probably many people many companies are going to use specifically in Enterprise right and there is also an Enterprise version we can specifically develop this as an Enterprise level application right with all the features and it is also coming up with its own cloud where you can just do drag and drop and you can actually create this right uh final uh important feature uh if I talk about is something called as fault tolerance fault tolerance so it also has this amazing feature Lang graph okay so the kind of libraries you'll be seeing Lang graph libraries will include right uh it will be able to handle errors you'll be seeing that it'll be able to handle errors uh it'll be ensuring uh even though your application fails it should be continuously working and um let's say that one individual agent encounters an error right then also there will be some fault tolerance mechanic M it will basically have wherein it will not just stop the application but instead it will keep on running right so this is one of the very important core core feature that is reliability right whenever we talk about reliability so uh this was two important things we have discussed about the definition of Lang graph we have discussed why Lang graph now it's time that I show you some of the amazing uh applications how we can actually go ahead with the development here uh what we are basically going to do is that we are going to create a chatbot okay and uh we'll try to create an agent uh let's say in the form of chatbot and again in the back back side right or in the back end you'll be able to see that we'll also be able to create a graph like how that entire agent specifically works so quickly I will go ahead and this I'm executing in the Google collab you can also go ahead and use it okay so let's say that I will just use any of the GPU that I have okay I'll just go ahead and connect it anyhow you don't require GPU in this uh but uh definitely we will be seeing that if you have any kind of uh if you if you require any kind of keys over here and when I say that I'm going to create a chatbot let me just go ahead and discuss about this architecture of the chatbot so what I'm going to create over here right so let's say that uh there's a chatbot over here this chatbot I will try to create some kind of node let's say this will be my start node okay whenever any user query comes you should be able to see that I will create a chatbot over here this chatbot can basically interact with any llm models okay and once it probably gives you the output we should end this entire flow okay so the reason why I'm drawing this in the form of graph because Lang graph also does the same thing okay so Lang graph also creates the entire flow in the form of graph itself where you'll have nodes and all right now there are two main important things when we see this these are specifically nodes right start and end node this chatbot is one node and when I write chatbot that basically means it is having some kind of definition right if a chatbot needs to move from one step let's say from one state to the other state we basically call that as a state management mement right so probably after executing this the chart bot is moving towards the end state so something is basically happening so let's go ahead and create this and then you'll be able to understand more about it okay so first of all we will go ahead and install pip install Lang graph so we are going to specifically use Lang graph Lang graph okay and we will also use Lang Smith so let's go ahead and execute this lsmith I hope everybody knows what exactly is so till then I will also go ahead and show you the documentation page of Lang graph so this is what langra is all about right it has python it has it supports JavaScript and all but uh uh if I just go ahead and show you Lang graph tutorials right so here you can see this is the tutorial right and uh not this sorry uh let's see yes yes here you go right in the Lang graph page it should be somewhere or let me just go to Lang chain Lang chain.com and uh here here is your Lang graph yes so here you can actually see we will try to create this kind of nodes right and uh I hope everybody has seen this kind of nodes specifically right so what does langra actually supports you to it provides controllable cognitive architecture for any task it is designed for human agent collaborations all right this kind of graphs also you'll be able to see all these things I will try to show you in this video so it has this fault tolerant scalability optimized for real world interaction and integrated developer experience amazing features and I feel right if you probably uh get the access of the beta server right and you'll just be able to see the power of this but at the end of the day we will be creating some amazing agents you know so here you can see I have installed both the libraries langra and lsmith now let's go ahead and start more things over here now as we go ahead we are also going to install Lang chain because we definitely require Lang chain and then we are going to use Lang chenore Gro along with this we are also going to use linore Community right so these all libraries will be specifically required in order to build our entire chatbot right so here you can see the installation is basically taking place okay perfect now let's go ahead and import Os or what I will do is that I will write from google. collab okay import user data okay uh now see as you all know when we worked specifically with vs code you know we used to create environment variables to ex ACC any Keys it can be grock API it can be lsmith Keys now similarly with respect to Google collab whenever you are working with Google collab here you can see one option here you can actually go ahead and upload all your secret Keys let's say if you want to use grock API if you want to use open API if you want to use lsmith API if you want to use hugging face API you can just go ahead and enter it over here add a new secret can you can add the key and value pair okay here is your password here is your key name now in order to access this key you can actually see that I've already uploaded it over here right so let's say since I'm going to use a chatbot which will be interacting with my Gro API so if you don't know about grock API you can just go over here and search for gro.com uh when you go to gro.com you can sign it with your email ID let's say I'm going to sign it because it provides you free API to access the open source models you know so I will go to grock API over here and here you you will be able to see that lot of different different models are basically available right Lama 3.1 you also have llama 3 you have gamma 29b you have distal whisper large V3 encoding uh so all these kind of models are there you can specifically use this particular model I've already covered this in my previous videos okay now in this scenario what I'm actually going to do I will be using some apis for this so if I go and click on start building here you can see that I will go to my grock Cloud console then here you can go to API Keys once you go to API Keys you can probably create your own API key I've created so many different API Keys uh you can go ahead and create your own API key right I've already shown that also in my previous videos once you create an API key just go ahead and click over here add a new secret key just go ahead and write groor API key and put it over here right all the API Keys itself now once you're able to add in the secret the next thing is that I will try to read that particular API key in my Google collab so in order to read I will go ahead and write from google. collab UT user data and uh let's say that I want my grock uncore API uncore key I can just use this user data do get grock API key so if I go ahead and just print this you'll be able to see that I'm actually able to read it you know groc API key this is just to to show you don't use this anyhow uh so here you can see it is Grant it is asking should we access this particular secret name Gro API key The Notebook titled untitled6 Point IP does not have access to secret name Grant access so if I click on Grant access but right now what happened timeout happened right so I'll go ahead and execute it once again and now you'll be able to access this right so this is my grock API key that you are able to see perfect right so now this grock API key I can actually use it I can use it for uh creating my L application anything that I want okay now once this is done uh what I'm actually going to do I'm also going to use lsmith okay and I hope everybody knows what exactly is lsmith right so here I will go ahead and write lsmith uh Langs Smith actually helps you to so if I go ahead and just do sign up right it'll be able to help you to track each and every information with respect to the interaction that you have created in your chat B right so if you go to lsmith over here here also you'll be able to get the dashboard right so I'll go to settings here is my API key if you don't have an API key go ahead and create an API key but if you go ahead and see in my secret key I've already added that Langs smithore API _ key okay now let's go ahead and read the lsmith key also so I will go ahead and write Lang Smith is equal to user data. getet and here also I'm going to use BL Smith blank Smith let's see what is the key name okay okay so I will just go ahead and print my L Smith okay so here you'll be able to see that I will also be able to get it so it will ask for Grant access I will give the grant access over here and it should be able to display it and it is displaying AF is the last term AF is also over here so I have read both this particular key and both this key will be specifically used for creating my chatbot application lsmith will be specifically used for tracking everything that is basically happening over here now the next thing you do that I will go ahead and import OS then let's go ahead and import os. enviren I will just go ahead and keep save all these things right so first of all I need Lang chain uh sorry lsmith API key okay I can also go ahead and write lsmith or I can also go ahead and write langin API key so let's go ahead and do this Lang chain API key I will set this to my lsmith okay we need to set up this specific Keys itself okay um then my endpoint endpoint is specifically not required I will just go ahead and copy this two things and this is basically just used to set up the basic requirements right lch and tracing V2 and langin project okay so let's say this will basically be my course Lang graph okay so this is what is my project name that I'm actually writing I'm saying that hey you need to make sure that you need to keep the tracing _ V2 as true and L langin API key is nothing but it is assigned to lsmith API key now you may be thinking Chris why did you use l lore API key no you can also use lsmith but according to your documentation when I saw both of them were satisfied so I just used linore API key so let's go ahead and execute this so once this is executed you'll be able to see all the information over here now quickly let's go ahead and write from linore grock import Chad Gro because I need to use the Chad gr open source models okay now this is done now I will go and create my llm model so which is nothing but chat grock and inside this I will be using my groor API uncope key which is equal to grock API key and my model underscore name will be nothing but you can use multiple models any models that you want so in the gro Cloud I can see there is Lama 3.1 anything that you want I will just go ahead and use this gamma 29b it okay so this specific model I will go ahead and use it it is up to to you now nothing compulsory which model you really need to use you can use any model that is provided by Gro right so this is done this is my llm model okay now the next step will be that now here you can see yeah I'm able to access all the models I'm able to access the model gamma 29b it okay now it's time we start building chatbot using Lang graph right so here is what we are going to basically start and this is the most important thing right so if I probably go ahead and say this okay so first of all uh understand there are multiple things which we really need to start with the importing part okay so first of all I will be importing something called as annotated right so from typing I will be importing annotated from typing extension I'll be importing type dict okay uh along with this I will be importing two important libraries from Lang graph so I will go ahead and write from Lang graph uh you'll understand why this importing is done just in some time okay from Lang graph. graph okay I'm going to just go ahead and import two things one is State graph State graph actually helps you to manage the entire State Management then your start node start node and your end node right your you remember that I actually created a uh a diagram right where we had a startat node then we had an end node which actually uh shows the flow of the entire chart bot itself okay along with this uh I will also be showing let this see the state graph is basically used for the State Management purpose now the state graph needs to keep on changing based on some parameters okay so for that what I will be doing I will go ahead and write from Lang graph langra do graph. messages message I'm just going to go ahead and import addore messages okay now this is one of the function what it does is that we have this something called as messages right as we keep on adding the messages over here we will be able to see that our state management our state of the agent will keep on changing now this messages is just like you can understand like when a user is giving a query the llm model is giving a response when it is giving a response the message will get added over here and then my state of the chat bot will also get changed right so this messages is basically keeping the track of all those things okay so now let's go ahead and execute this so this has got executed successfully you can understand each and everything I've have actually explained we will talk about this too right just in some time now now we will go ahead and create our own class let's say this is my class state which will be um making sure to uh probably control the entire State Management here I will go ahead and write typed dict okay so I will go ahead and write typed dict typed dict okay typed dict okay so this class is basically inheriting this particular type D uh okay now what we will do we will go ahead and write some comments over here okay so that you'll be able to understand messages that have the type list the add message function in The annotation defines how this state key should be updated it appends the message to the list rather than overwriting so inside this class I've created a variable called as messages okay and this messages is of type annotated and this annotated is basically saying hey it is a list type and here I'm using this function called add messages this add messages will be responsible on adding or appending the messages to this list okay which is basically created in the form of list in this messages variable and it will not override them the entire State Management will be controlled as we keep on adding this particular messages because on every ad we will be changing our state okay now the next thing that we will go ahead and Define since I've defined my class State I will go ahead and write my graph Builder okay I will go ahead and Define My Graph Builder so this is what we start our graph building process so here I'm going to use State graph the state graph which I have actually imported it over here and this state graph will have this particular class which is called as state because this state knows which messages are there what messages are basically getting appended and this graph Builder should be responsible in managing the enti State Management right so once I execute this here is my graph Builder so let me just go ahead and execute this so This Is My Graph Builder over here right so here you can see Lang graph. graph. state state graph at this specific location so guys now after we have created the graph Builder uh what we are basically going to do is that uh already in this diagram you saw that right we are going to create this chatbot now this chatbot should be interacting with our llm so for this we will go ahead and write some definition so let's quickly go ahead and uh create a function definition chatbot and remember whenever we creating any node right in this entire graph Builder it needs to take the parameter as state okay whatever state we are giving over here and this will be important because the reason is very simple because based on this state message right whatever messages whatever interaction that we are doing the state management will keep on changing so here we go ahead and Define this chatbot where we take the parameter as state and then we return you know that after any response that we get from the chat bot we need to update this messages variable so here I will go ahead and create this messages and here I will just go ahead and write llm do invoke okay and this invoke will be specifically invoking from the state message which will have the user queries right so whenever a user query you'll be able to see that this messages will get appended in this particular message variable and once it does that we just going to invoke that and whatever response we getting it we again appending it over here right or we are putting in this particular value we are returning this entire in this particular format when we return this when we are inheriting the state class right it is going to understand Hey whenever we return this kind of message it is going to trigger this add message and it is going to add this in this messages variable okay so just understand in this way is that here I have a chatbot function which is invoking the previous messages based given by the user and we are returning some kind of messages over here okay so once we execute this I will be getting my chatbot now this chatbot needs to get added in this graph Builder so here I will go ahead and write graphcore Builder dot addore node and here I'm going to specifically use chatbot as my node name name which will have the functionality of this particular chatbot okay so once I execute this you'll be able to see that this entire node is basically getting added so now if I go ahead and just display this you'll not be able to see much things over here but we will be seeing the entire diagram so step by step you can see that what we have done first of all we have created this class State then we have created our graph builder then we have defined our chatbot what functionality is basically doing and what is returning it is returning just message colon some some output that we are getting over here right and remember this chart bot is inheriting the state class and the state class has one variable messages in the form of list and there is a function called as addore message which will keep on appending whatever message we are specifically getting over here right so that is what it is basically doing now since we have created this graph Builder now it's time that we connect this chatbot to the start and the end node okay so here I'm going to go ahead and write graph Builder dot add edge okay so add edge start chatbot okay so we are basically going to add the start and the chatbot and similarly we are going to add chatbot till the end right so then only we will be able to find out uh we'll be able to get this kind of flow so start is basically appended with chatbot and chatbot is basically getting appended with end start basically means once ever we get get a user query it will just start the process then chatbot will do the interaction with llm it'll give us the value and it will go to the end State and how this state managing is basically happening because of this state class which we have inher in graph Builder Okay so once we execute this this is done so finally what we do after creating this after adding all these nodes we go ahead and write graph Builder we need to compile it so we'll go ahead and write do compile so once we compile it we'll be getting our entire graph ready so here is my graph itself so once I execute this my graph is ready now let me just go ahead and display this so in order to display this uh in the documentation I saw this code where they are importing dis uh image and display and there is something called as graph. get graph so here instead of graph I'll write graph Builder okay or sorry it should be graph only because we have compiled it get graph draw mermaid PNG okay okay so once I execute this you'll be able to see that I'll be getting this entire flow so start and end are just like one private uh functions that are present inside the uh State uh State uh I probably say this particular class the state graph class right uh which is present inside this lang. graph Lang Lang graph. gra right and now this will basically as soon as the user queries we will go to the start State then it goes to the chat chatbot here the interaction will basically happen and once the interaction happens the state should change right once we get the response it will go to the end State now let's go ahead and execute this and then only you'll be able to see the output okay uh and we are just going to start this okay so here I will just go ahead and write while true user uncore input is equal to input user input user and here I'm going to just go ahead and write if user uncore input do lower in let's say I will go ahead and say hey if you just going and typing in quit or Q right we just going to stop the conversation with the flow right so here I'm just going to write print goodbye okay and then we just going to break from here okay but if the user input something else so here I will goad and write for event in graph. stream okay we just going to stream and here we are going to use this messages as my key variable messages is a colon and here I'm going to give based on user see initially whenever user gives the input message I need to set this as user right as a parameter and here will basically be my user input okay then I will go ahead and write print event. values we are just going to print the values then here more we can go ahead and probably convert this or explore this values so here I will write for Value in event do values print value and inside that value there will be a key which is called as message messages okay and then we will go ahead and print our assistant message assistant basically means the message that is basically coming from our uh llms so here you'll be able to see value messages do content so here let's explore this code first of all a user is giving an input if it is not giving quit and Q so it will go to graph. stream with respect to all the messages that users have already given then we are going to get that event. values for Value we are going to get two important things one is value of messages and one is uh the value of messages of content so this value of messages I think it should have the user message and this should have the llm model respon uh response so let me go ahead and execute it so I will go ahead and write hello let's see so here assistant you can see right hello how can I help you see all these messages are basically coming in the form of Dick value right so here uh hello how can I help you today is the response and I'm able to get so let me go ahead and ask what is J AI right so here you'll be able to see that all the response is there right geni is a type of artificial intelligence think of it so it's amazing right you're able to get the entire response and here also you can see response metadata content is over here so if you just go ahead and explore this some of the other keys will be available over here right and uh uh if you go ahead and see down so this is my entire generative message so if I just go ahead and write Q it should be exiting over here right so here you can see it says goodbye so guys I hope you like this specific video uh and you were able to understand how to basically create a simple chatbot in the form of Lang with the help of Lang graph and here we have basically used the graph itself right from start till the end node we have understood that whenever we create any node it also needs to have a definition and uh how we can connect each and every node now in the upcoming videos we are going to develop more complex projects on Lang graph where we'll be using external tools uh there will just not use one node but there will be multiple nodes as we go ahead so I hope you like this particular video I will see you all in the next video thank you