Understanding AutoGen Message Flow

Key Takeaways

Hey everyone, welcome back. So, now let's move forward and try to understand a little bit deep about how our agent exactly is working. Uh we saw that how we can define an assistant agent, provided with a brain and a tool, and can run the same. For that, we used a weather agent overall. We defined our own weather agent. If I can show you here by going up, yeah. So, this was the weather agent which we have defined. One thing to note, and I will show you that from the official documentation, that we can have different different kinds of agent. Though assistant agent is going to be the one which we majorly use, because it is the agent which can use AI and answer your question. If we go on the API reference, and then go on agents, so autogen.agentchat.agents, then we can see that how we have the base chat agent class, where we can extend to define other agents. And we have assistant agent, which, as the name suggests, can assist us in something. We have the code executor agent, right? Then we can have a user proxy agent as well. So, agent is just a type, and we can have different different types of agents which we can define. Along with that, if you go on autogen.extensions, there as well we have multiple agents. So, let me show you that as well. And I think that will help you to understand dot agents. And we have present here, yeah. So, agents for file surfer, for open AI, web surfer, like multi-model web surfer agent that act as a web surfer and can search the web for us, right? So, these are the different different agents which are defined, everyone. Now, let's go back and try to understand a little bit more about how it's working internally. Though the steps are exactly the same as we have saw on previously, where how it is receiving a task, then understanding the context, everything. But I would like to show you this with respect to the result. Because understanding this result which we are getting is very, very important, right? One thing to note before is that whenever and I will have a icon here, so it is easily understand. Just a second. So, 1 minute. I think this can help us to understand better. We provide our agent with some task everyone, so I can just say that this is our task, right? Which we give to our agent. Now, when agent is doing a particular task, there are multiple messages which it omits, right? So, I will show you these messages as well. Just a second. And we can have the icon. Great. Now, if you pay a lot of attention, what exactly is happening is that we have different, different types of messages which we omit. Now, the very first thing is that we also send a text message. If you see this result, we are also sending a text message, right? To our agent. So, we as a user, the task is being sent as a message here, right? So, this is what goes to the agent, to the digital human. After that, let me send it backwards. After that, what happens everyone is that this message is something which we also receive finally. So, in the end if you remember, we also receive the text message, right? In between, there are multiple things which are happening. Now, if we have multiple agents, something which we will see when we see multi-agent teams, right? They can also converse in terms of messages, right? They can have text message. If you're calling a tool, then tool call event is getting uh placed here. So, I will show you this in a little bit of understanding. So, pay attention. This is exactly what is happening here. We have the text message which we have sent. So, this is sent by us user. All right? This is we as a user. When the agent gets this particular message. When the agent Okay, I hope this drawing for an agent is pretty good. So, when our agent get this particular message, I can make it a little bit better. Not to make it mail. Right? It has a tool which it calls. So, this is why it is tool called request event. The tool then give us an answer. It is tool called execution event. And then finally, agent give us the response in a text message. So, this is how internally it is working. In chat GPT, this thing is hidden. But here, we can exactly see what is being sent to my function city daily. What we are returning here. And then finally, what are the number of tokens getting used etc. Okay? So, a better way to see this result and to explain you this. Always and always So, if say say result.messages. This is where all my messages are. Text message, tool called request. Tool called execution and finally, the text message which is a response. If I get the last one by minus one and do content. I will be able to just see the final answer which is easily readable. And I think this is what even chat GPT sends us. Right? So, here this is where this content is mentioned everyone. See? And this is the last message. Okay? Now, to show you one more thing. If we go on agent here. And yeah, these messages. Right? So, this is what I want to show you. In AutoGen, agent chat message facilitates communication information communication and information exchange with other agent orchestrator and application. So, again, just like how we in real life, we humans, we talk on messages. A message can be of text type, of image, of voice. In a similar sense, in a very, very similar sense, our agent are also uh in AutoGen are working on these messages. So, understanding this analogy is very, very important so that you can make sense of what is happening inside. Now, just like in real life, we can have different different types of messages. Uh we can define them either by type of where they are going. So, agent to agent messages, right? Or let's say agent message going to some internal types. Now, we can see uh like I will discuss on this how we can have these different different types. But, the idea is that we can have a text message, we can have a multimodal message where we have images, etc. And they all are based on base chat message. Something which will be defined in your uh overall core. In the library, we can see that how in the messages we have the base message class. Okay? So, this is what is to be understood. Then, internally, everything is happening with these messages, and that is what we kind of sent with the different different agents, applications, and things, right? So, with that, I hope that this thing is clear. Also, go through this particular Uh let me show you this one, which is defined here, and it is having the similar kind of explanation with all the things. So, it is easy to follow with headings, right? But, I will highly suggest to do the practicals with me so you understand what exactly is happening. Let me close these one, and that will make a lot of sense that how you can use AutoGen to create any appli- agentic application for yourself. So, with that, everyone, now let's meet in the next video where we will learn about how we can create better agents, what are all the other things we have with respect to our agents? How can we control the models, etc.