CrewAI Tutorial For Beginners 2026 | How To Create Multi Agent Workflow With CrewAI | Simplilearn

Key Takeaways

[music] >> Hello and welcome to the CrewAI tutorial by Simply Learn. In this video, we are going to learn how to build AI agent system using CrewAI. Now, this entire tutorial is divided into three modules so that we can move step by step starting from the very fundamentals and eventually building a complete working system. But before we start installing CrewAI or writing any code, it is very important for us to understand what CrewAI actually is and why are so many developers talking about it right now. So, let me explain this with a very simple story. Imagine you're working in a company and your manager gives you a task to prepare a detailed report on a topic. Normally, one person doesn't handle everything alone. Instead, the work is divided among different people. One person might start by researching the topic, someone else analyzes the information, another person turns those insights into a structured report, and finally someone else reviews the output to make sure everything looks good. So, basically, a group of people with different skills collaborate to complete one larger task. Now, imagine doing the same thing, but instead of humans, you have AI agents. One agent focuses on research, another analyzes the information, another generates the report, and the next one reviews or improves the output. Each agent has its own roles and responsibilities, and together they collaborate to complete the task. This idea of multiple AI agents working together is exactly what CrewAI enables. Now, many people today uses AI tools like ChatGPT by simply giving prompts and getting responses. That works really well for smaller tasks, but when you start building real application, things usually involve multiple steps. For example, you need to gather information, analyze it, generate content, and then refine the result. Trying to manage all of that with just one prompt can quickly become very messy. Now, CrewAI solves the problem by allowing you to build team of AI agents that collaborate automatically inside a structured workflow. Instead of one AI doing everything, you create specialized AI agents that focuses on specific responsibilities and work together towards a shared goal. Now, if we look at how CrewAI structured, there are two important concepts you will often see, flows and crews. You can think of a flow as a manager of the entire system. It controls the workflow, decide what steps should run first and what should happen next, and how information moves through the process. A crew, on the other hand, is a team of AI agents that actually performs the work. Each agent inside the crew has a defined role, a goal, and sometimes even access to different tools to complete the task. So, in simple term, you can say the flow organizes the process, while the crew carries out the intelligent work. This combination allows developer to build AI system that are much more structured, scalable, and capable of solving complex problems. Now, because of this very approach, CrewAI is being used in many real-world application today. Developers are building automated research assistant, AI content generation pipelines, customer support system, and even AI-powered development assistant using it. Now, instead of interacting with AI through individual prompts, CrewAI allows you to build complete AI workflow where multiple agents collaborate. Now that you have a clear understanding of what CrewAI is and the idea behind AI agent collaboration, in the very next part of this tutorial, we will start exploring the core component in more details and see how the system actually work. Before moving on to the installation and hands-on implement. All right. Now that we understand what CrewAI is and how AI agent can collaborate like a team, the very next step is to actually install CrewAI and set up your development environment. I will walk you through this step by step, and I will also clearly mention when you should use command prompt or PowerShell, and when you should use the terminal inside your VS code. Now, the first thing we need to do is to check our Python version because CrewAI requires Python 3.10 or higher, but lower than 3.14. To do this, open your command prompt or PowerShell on your system. If you're on Windows, you can simply press the Windows key, type command prompt or PowerShell, and open it. Once it opens, type the following command and press enter, python3 version. This command will display the version of Python installed onto your system. If your version falls within the required range, you're good to go. And if not, you can download a compatible version from the official website and install it before continuing. And don't worry, I will be giving all the download links in the description below, so you can follow along step by step. Now, moving ahead, once Python is ready, now open your Visual Studio Code and open any folder where you want your project to live. After that, open the terminal inside the VS Code. Going to the terminal. Now, inside this terminal, we will first create a virtual environment using python -m venv venv. Because the Basically, the reason we are doing this is to keep all our project dependency isolated. Meaning, whatever libraries we will install for this project won't affect other project onto your system. After creating it, activate the environment and you will see venv in your terminal. Now, inside the same terminal, install CrewAI by running pip install CrewAI. This installs everything required to start building your AI agent. Now comes a very important part, setting up the AI model. So, basically, CrewAI needs a language model to actually think and generate output. Most people use APIs like OpenAI, but that comes with a cost and also requires internet access every time. So, instead, we are going to use Ollama. Now, why we are using Ollama? This is very important to understand. Basically, Ollama allows to run AI models locally onto your system. Meaning, you don't have to pay per API call. Your data stays onto your machine, which in which means better privacy. No dependency on the internet once the model is downloaded. And it gives faster responses in many of the cases. So, for beginner and even for many real-world use cases, this is a very powerful and cost-effective solution. Now, what you need to do is to download Ollama from the link in the description and install it on your system. Once installed, go back to your command prompt or VS Code terminal and type Ollama version just to verify that that is working. Now, moving ahead, Ollama itself is just a runtime. It doesn't comes with model pre-installed. So, the next step is to download a model. In your terminal, run Ollama pull llama3. Basically, what this command does is to download the llama3 model onto your system so that you can use it locally. Now, this step might take some time depending on your internet speed because models are usually large in size. Once it is done, you can test it by running Ollama run llama3. If you're able to chat with the model in your terminal, that means everything is set up correctly. Now, after this, come back to your VS Code project. This is where we actually start building. So, basically, your project has multiple files, agent.py, task.py, tools.py, and main.py. Now, let us understand this properly in detail. Now, starting with agents.py, this is where you define your agent. Now, agents are basically AI robots. Instead of using one AI to do everything, we divide the work into multiple agents, each having a specific responsibility. Now, each agent has a role, a goal, and sometimes a backstory. The role define what the agent is, the goal define what it needs to achieve, and the backstory give context so the responses are more accurate and relevant. For example, you can have a research agent whose job is to gather information, and a writer agent whose job is to convert that information into structured content. Now, the reason we do this is specialization. Just like in real life, when different people handle different tasks, the output is much more better. Moving to task.py, this is where you define that what works need to be done. Basically, tasks are instructions that you assign to agent. Each task clearly describes the objective and is linked to a specific agent. For example, one task could be find the latest trends in AI, which will be assigned to the research agent. Another task could be write a summary based on the research, which will be assigned to the writer agent. So, what happens is one agent complete its task and the output can be passed to another agent. This creates a complete workflow instead of just a single response. Now, this is the main reason why CrewAI is powerful. You're not just prompting an AI, you're building a multi-step intelligent system where agent collaborate with each other. Now, moving ahead, we have tools.py. Tools are basically additional capabilities that you can give to your agents. For example, if you want your agent to search the web, call an API, or run a function, you can define it here. Then, the agent can use these tools while performing task. This makes your system much more dynamic and powerful. And finally, we have main.py, which is the entry point for your project. This is where everything comes together. You import your agents, import your task, create a crew, which is basically your team, and then run it. So, when you open your VS Code terminal and run python main.py, this file starts the entire workflow and your agent begin executing their task step by step. So, overall, the reason behind each step is very clear. We install Python to run everything, we use a virtual environment to manage dependencies, we install CrewAI to build agent workflows, we use Ollama to run AI model locally without cost, and then we define agent and task to create a structured intelligent system. So, now that we are clear with the theory, let us move to the demonstration part, creating our agents. Now, coming to this VS Code, which you can see here, I have already installed all the dependencies because it will again going to take some times. And if you're going to follow the steps which I mentioned earlier, it will be easier for you to install all the things. Now, you can see this This is the CrewAI folder which I have made on the desktop, and it has got basically four main files which we earlier discussed, agent.py, main.py, task.py, and tool.py. Fine. Now, coming to this agent.py, you can see that what we have done here is from CrewAI, we have imported our agent. Okay? And then, we have created two agents here. One is senior researcher, which we have said agent, okay, now your role is a senior YouTube researcher. Basically, you will research about the given topic. Now, what is the goal here is explain topic simply. Whatever topics [clears throat] will be given to you, you have to explain the topic simply. We are defining our senior agent these roles. Now, what is this verbose is true, which is meaning will show everything in the terminal. Now, we have set the memory as false because if we will set the memory as true, it will remember the history. And false works fresh everything. Now, backstory is what is the backstory of this agent? It is expert in simplifying AI topics. Now, tools is list of the external tools such as YouTube search, etc. If you'll come to this documentation, which is CrewAI official documentation, you will find what is the tool and what are the tools which are present which you can use. So, for example, you have server depth tool, file read tool, websites host tool, and directory read tool. Along with these, you'll find a you will find a multiple option of the tool. You can go and select whatever you want. Okay. Now, coming back to this, then similarly, what is the LLM we are using is Ollama Tinylama. You can download this version. You can download this LLM Tinylama from Ollama pull Tinylama. Okay, fine. Now, similarly, in a similar manner, just as this senior researcher, we have a script writer agent. Now, what is the role we are assigning his that your role is a YouTube script writer. Our goal is to write short scripts and verbose true, similarly, this thing only, show everything in terminal, and then memory is false. And what's the backstory is it is expert in writing engaging scripts. Okay, for tools is also same. And what the a large language model we are using is the same. Fine. So, in this file, what we are doing is creating two agents. Now, coming to the task.py. Now, in task.py, starting from here, uh from CrewAI, we are importing task, and from agents, the agent file, we are importing senior researcher and script write agent which we created. >> [snorts] >> Now, what is the research task we are giving as the description? Explain topic, whatever topic will be asked, it will give in three simple bullet point. And what will be the expected output? Expected output is I want three clear bullet point explaining the topic. Fine, we [snorts] have given the description, and we have given what is the expected output which we want. Now, agent is we are naming the agent as senior researcher. Now, in the similar manner, we have script writing task. Now, script writing task is we are defining the task, and we are giving the description. Using the research, create a short YouTube script on the topic, the topic what whatever we will give, with a hook line and two to three line explanation, and with a outro. Now, what is the expected output? Expected output is a short and engaging YouTube script of whatever topic which we will assign. Then, agent here is script write. And here the agent was senior researcher. So, fine. Now that we have understood the two files, which is agent.py, and now coming to this tools.py. So, as of now, we are not using any of the tools, so we have left it blank. But in future, if you want to add more tools, if you want to add more functionality, you can use these tools. And again, you can go to this documentation, wherein you can find all the tools and required things. Now, here the main.py, where everything will be collected, and this main.py will be the file which you will be running. So, what we have is from CrewAI, we are importing crew and the process. From agents, so from this agent file, we are importing senior research, senior researcher, and the script writer agent, the two agents which we created. Then, from task, we are importing research task and script task. Okay. Now, here is what we are creating the crew. Basically, we are creating the agents and the workflow and everything. So, what we are doing is agent, I am giving this agents our two agents, which is senior researcher and script writer. Then, what is the task? Task is research task and script task. So, one agent will be researching the task, and another agent will be scripting that task. Then, what [snorts] is the process? Process is sequential. We are putting here sequential meaning task will be executed one by one. Okay. So, for example, firstly, the research task will be happening, then the script task will happen. Now, verbose, similarly, we are setting it to two. Now, this is what we are running the crew. So, in the result, we are starting crew.kickoff. Input is topic CrewAI tutorial for beginners. So, you remember in curly braces here, we are putting the topic, and here in main, we are giving the topic as CrewAI tutorial for beginners. So, this starts the entire workflow, and then in the end, we are printing the result. So, this were basically your four of the files. Now, if we run this main.py, let us see what are we getting. It's again going to take some time. Now, as you can see here is the task has started, senior YouTube researcher, explain CrewAI tutorial for beginner in three simple bullet points. Now, agent is senior YouTube researcher, firstly do and research. Fine, it has researched, and it has given the final answer in three points. Acknowledging your expertise, this what is CrewAI, how to access the result, and then it is summarizing also. When task is completed, then it is uh our the next task is of YouTube script writer. This will again start the task. Using the research, create a short uh CrewAI tutorial for beginner. Okay, fine. We have seen that one agent has already completed the task, and second agent has started. Okay, fine. See, the final output is thank you for your feedback. In this tutorial, we will walk through it has given all the uh YouTube script using this CrewAI tutorial. Regards, and then your name. So, you can see that it has given the final output also without any of your intervention. So, it has researched also, it has written the script also. So, you see this is so much easier. Now, if you want the entire code file, just comment the code in the comment section, and I'll be giving you all the entire code file for the same.