Building a ReAct AI Agent (Tutorial)

Key Takeaways

in this demonstration we're going to use flse to create our first basic react agent inside of flowes you will choose chop flows then you will choose add new then you will start off with a blank canvas okay so now what we can do here is we can start off with our language model so we need a language model this is going to help power this agent capabilities such as planning and reasoning and we're going to use chat open AI so here we have chat open Ai and you should have already set up your credentials so we have a video about that as well in the previous section where we introduced LCI and you will take that credential and you're going to set it up here so I'm going to choose my credential here because I already have it stored and then for this I'm going to use the GPD 40 mini which is the one of the latest models from openi and I need to use one of the latest models because I want to use the model that has really good reasoning capabilities but you can experiment with any of these models or different providers as well so we have this model here and then what we need is we need our react agent so instead of flow wise because it's built on top of Lang chain we have the ability to use Lang Chain's react agent functionality so if we go under agents right here you will see that there are several types of agents that have been built in into the FL I tool and so the one that we're going to use is this react agent for chat models so agent that uses the react logic to decide what action to take optimize to be used with chat models so I'm just going to drag this here and so now we have two of the core components that makes this react agent so what I need to do is I need to connect this now and we have our chat model right here and then we're missing two components here we discuss the ability of the react agent to use these external tools so we can connect an external tool right here so we can use for instance something like a calculator as a very basic example here so I'm going to go here and you will see that there's also a tool section so here's a tool section and under Tools we have this calculator tool so I can go and drag that here and now we have a calculator tool it's not asking me to do anything special it's just a calculator but I can now connect that to allow tools so I can continue to add more tools if I want but for this demonstration I'm going to keep it very basic in future use cases and demonstrations we're going to be using all kinds of very complex tools such as search engines and so forth then the last component here is memory so I'm going to take a memory component from here so if we go to the list here we have memory and under memory we have different kinds of memory components one type of memory component that we can use here is this buffer memory and what does it do it retrieve chat messages stored in a database so basically it's going to allow the react agent to use its chat history if it needs to access whatever information it had pulled already from the external tool so I'm just going to drag it here and then I'm just going to connect it right here we have our memory component we have our chat model and we have our calculator so that's it we have developed a very basic react agent and now we can save this so I'm going to go here to this button I'm going to save and I'm going to give it a name I'm going to call this basic react I'm going to save this and the cool thing about flowise is that I can very quickly test this idea so we have this chat feature here that I can go and open and then I can start to interact with my agent so I can type a question like how much is 98,000 times I'm giving it a m problem here I want agent to actually use the calculator because that is what it has access to I'm just going to ask you this question for problems like this you really don't want to trust a language model to do Tas like this so it's going to be very useful for the agent to access this calculator and it's going to see it it's going to see oh I have a sort of mat problem here and it's going to decide okay I need to take these two different numbers and I need to give and provide that to the calculator as some type of input and then the calculator is going to respond back with the result and then it's going to have the result and then the language mod will compose a final answer for us that's sort of what the process is here so you can see here it says 98,000 * 150,000 is this huge number here so again for calculations like this we really want to use like a calculator tool to do this so this is the first example of a basic react agent with OP the different components that we discussed in this module and in upcoming modules we are going to build on top of this idea and build more complex agentic systems that are going to leverage complex tools and leverage this idea of the ability of an agent to use memory to use different tools and use a very powerful large language model to complete very complex tasks