Your First Crew: The Quickstart Guide

Your First Crew: The Quickstart Guide

Theory is essential, but nothing beats running code. In this chapter, we will build a simple crew that researches a technology topic and writes a short blog post about it. We will use the CrewAI CLI to generate the project structure.

Scaffolding the Project

Open your terminal and run the create command. We will name our project latest_ai_news.

crewai create crew latest_ai_news

The CLI will ask you to select a provider (like OpenAI) and a model. Once completed, navigate into your new folder:

cd latest_ai_news

Understanding the Directory Structure

CrewAI encourages a clean separation of concerns. Here is the layout of the generated project:

latest_ai_news/
├── src/
│   └── latest_ai_news/
│       ├── config/
│       │   ├── agents.yaml    <-- Define your Agents' personas here
│       │   └── tasks.yaml     <-- Define the Tasks here
│       ├── crew.py            <-- The logic binding everything together
│       └── main.py            <-- Entry point to run the crew
├── .env                       <-- API Keys (DON'T commit this)
└── pyproject.toml             <-- Dependencies

Step 1: Configuring Agents

Open src/latest_ai_news/config/agents.yaml. This file allows you to define agents using natural language, keeping your code clean. Let’s define a Researcher and a Writer.

src/latest_ai_news/config/agents.yaml

researcher:
  role: >
    Senior Data Researcher
  goal: >
    Uncover cutting-edge developments in {topic}
  backstory: >
    You are a veteran tech journalist with an instinct for
    finding the most important news. You prioritize facts over hype.

writer:
  role: >
    Tech Content Strategist
  goal: >
    Craft compelling content on tech advancements
  backstory: >
    You are a famous writer, known for simplifying complex topics.
    You base your writing on the work of the Data Researcher.

Notice the {topic} placeholder? We will pass this dynamic input when we run the crew.

Step 2: Configuring Tasks

Next, open src/latest_ai_news/config/tasks.yaml. We define what the agents need to do.

src/latest_ai_news/config/tasks.yaml

research_task:
  description: >
    Conduct a thorough research about {topic}.
    Identify key trends and breakthrough technologies.
  expected_output: >
    A list of top 5 trends with brief descriptions.
  agent: researcher

writing_task:
  description: >
    Using the research findings, write a blog post about {topic}.
  expected_output: >
    A 4-paragraph blog post formatted in markdown.
  agent: writer

Step 3: The Crew Logic

Look at src/latest_ai_news/crew.py. The CLI has already generated the Python code to load these YAML files. It uses decorators (@agent, @task, @crew) to map the YAML configurations to actual Python class instances.

Step 4: Running the Crew

Before running, ensure your .env file contains your API key (e.g., OPENAI_API_KEY=sk-...).

Now, execute the crew using the command below. This invokes the logic in main.py.

crewai run

You will see the agents interacting in the console: the Researcher will generate the trends, and the Writer will pick that up and draft the post. Congratulations! You’ve just built your first multi-agent AI system.