Mastering LangChain Deep Agents: 3 Ways to Initialize & Switch LLMs Dynamically

Hi everyone. Welcome back. Nagesh here. In the last couple of videos, we discussed about deep agent architecture and we have done a sample program, kind of a hello world program using LangChain deep agents. And in this video and coming couple of videos, we'll be discussing on some core components of LangChain deep agent. Okay, so understanding of this core components is very important. So whenever we discuss about the the real real project, then this understanding will help us to go very smoothly. So let's start. So first, we can start about the models. So as you know, the deep agent is um actually is a model agnostic. So that means we can plug different LLM depending on our requirement. For example, uh lightweight models for the fast response and reasoning models for the complex task. And we also use the different providers. We can use OpenAI, Anthropic, Grok, and the Google, right? So we can see different model initialization in deep agents. So let me execute this uh load environment. Um Okay, let me do this. Okay, so let's do this. Okay. So here we can see uh so uh the installation is in progress, that is fine. Okay, it's done. Okay. Now we can see in a simple terms, we can import the create react agent from the deep agents. And here in a simple way, I can initialize a model something like this. I can mention the provider and I can specify the the type of the model. So here the provider is the OpenAI and the model that we are using the GPT-4o mini. And let's execute this. Okay, so this is the simplest way of initialization. Okay. Now, consider in a scenario where we need to handle the model configurations. Like, for example, I need to control the reasoning level, temperature, time what I retrace, etc. Then, in in that case, what we can do we can make use of the init chat model property or the method. Okay. So, here this is we import from the LangChain dot chat model. This we commonly using this init chat model. Guys, here I can specifically say about the model. And I can tell that, okay, thinking level is equal to medium here. So, in that way I can control each and every parameter of this model. So, let's execute it. This also will work without any problem. Okay. So, the next scenario is that, okay, imagine a situation like we don't want to repeatedly pass the same configuration everywhere. Okay. For example, I'm saying that, okay, whenever I'm using I'm using a provider is a OpenAI, then its temperature is always zero. All right, it's kind of a global settings I want to do. Okay. In that scenario, we can use the provider profile and the register provider profile from Deep Agents. Okay. So, here what I'm what what this means is that here I'm registering here registering the provider profile saying that, okay, if my provider is OpenAI, then its temperature should be zero. Okay. So, this is what I mean by this. So, if I'm using any different provider, then let me execute this. Then, it may be its own default temperature. Okay, it's not zero. Okay, this is only for the OpenAI provider. Now, I can simply execute something like this. So, in this scenario, this OpenAI GPT-4 or mini, its temperature is zero. Now, I'm thinking that, okay, for GPT 5.4, I want to specifically set some configuration. Like, I want to make it like a reasoning effort as a medium. So, in that case, what I can do, I can specifically say that the provider and the model and I can set it like a reasoning effort as medium. Okay. So, in that scenario, if I use any other model other than GPT 5.4, it's reasoning efforts should not be medium. So, this is only applicable to GPT 5.4. So, here, for this GPT 5.4, uh the temperature is zero and the reasoning effort is medium. Okay. So, let's execute it. Okay. So, this is a a second method that we can use. Okay. Now, in another situation, okay, this is the last one. Situation, consider that we have to dynamically switch the model. Okay. Consider that I have a UI is there and based on each and every use cases, I want to switch the model from the drop down. So, in that case, building the deep agents is every time is really costly and that is not a practical way of doing that, right? So, in that case, we can use a middleware that is um called the uh wrap model. We can use this um configuration like a wrap model called and this can be used as a middleware. Okay. So, here we can see a configurable model middleware is there. And here, what we are doing, we are actually checking. Okay. We are We are actually checking the model from the context. Okay. So, this will be get understand if I scroll it down, but for the time being, just understand that in a request, in a context, I'm getting the model name and based on this model name, I am initializing the model. Then, here I am updating the entire model. Okay. So, uh based on this, I'm going to create a deep agent. Here, we can see a open AI mini is my model and middleware is the configurable middleware and I'm passing the context. So here context means only the model. Okay, now let execute it. So let execute it. So you may be thinking that okay, what is this contact from where we are getting the context all those things. So that can that I can answer by executing this thing. Okay, so here I'm invoking this model and I'm dynamically passing the context here model context here. Here I'm saying that okay, the model for this context will be GPT 5.4. So let me execute it. So the execution is in progress. So let me do a result here. So if you scroll down little bit scroll down here little bit here we can see the model provider is open AI but the model name is GPT 5.4 because we have given the GPT 5.4 here. So it is dynamically changing. Okay, now we can go back again to the same code by understanding this particular the code segment. We can go and see this, right? So here I'm passing the context with the model name and using this middleware configurable model, I will get the model information from the runtime context. I am initializing the model then I am updating the configuration back. Okay, so this is how it works. Okay, so in this way that these are the like a three different ways that we can initialize the model and the last one is very important when if you want to change the model dynamically. Okay, so that is all about in this video. If you have any questions, please ask in the comment box. I will try to answer that. So that's it. Please do like and subscribe. Have a nice day. Bye-bye.