CrewAI creates intelligent agents.

CrewAI is an open-source framework for multi-agent collaboration, specifically designed to orchestrate and coordinate multiple AI agents to work together.

CrewAI can break down a complex task into multiple roles, each responsible for a part, and complete it through process collaboration.

Comparative understanding:

• Single Agent: A large model, from start to finish.
• CrewAIProduct Manager + Engineer + Analyst + Editor, each doing their own job.

CrewAI is a tool for coordinating, managing, and framing AI Agents, built on LangChain and Pydantic, to facilitate role-playing, autonomous, and collaborative Agent teams.

• Crew: One consisting of multipleAgentThe project team formed.
• Agent (Member): : Individuals in the team, each has a clearrole(Role),goal(Target) andbackstory(background story).
• Task: The specific work that needs to be completed by the team. OneCrewContains multiple ordered or parallelTask, and assign to appropriateAgent。
• Process: Defines the team's workflow, e.g., whether tasks are executed sequentially or concurrently.

In simple terms, crewAI provides a structured way to define who (Agent) does what (Task) under which process (Process), ultimately achieving team goals.

The flowchart below clearly shows how the core components in the crewAI framework work together:

The process begins with defining agents with specific roles and goals, then creating specific tasks for them.

Next, these agents and their tasks are organized into a crew, and a collaboration process is selected for the crew, such as sequential or parallel execution. Finally, the crew executes all tasks according to the defined process and produces the final result.

Environment setup and installation

Before we start building, we need to prepare the development environment.

Precondition

Python version requirements:

• Must: Python ≥ 3.10 and < 3.14
• You can type in the terminalpython --versionto check. If the version range is not met, there will be many problems later, and it is not recommended to tough it out.

CrewAI usesUVfor dependency and package management, with only one goal:
To make multi-agent projects more stable and not be brought down by environment issues.

UV introductory tutorial reference:UV - Python package and environment management tool。

API key:

crewAI itself does not provide AI models; it needs to connect to large language models such as OpenAI's GPT, Anthropic's Claude, and others.

Install crewAI

Open your terminal or command-line tool and use the pip command to install the crewAI package.

# 正常安装
pip install crewai
# 其他依赖
pip install langchain
pip install openai


# 如果安装慢，可以使用国内镜像安装
pip install crewai -i https://mirrors.aliyun.com/pypi/simple/
pip install langchain -i https://mirrors.aliyun.com/pypi/simple/
pip install openai -i https://mirrors.aliyun.com/pypi/simple/

If you wish to use some of crewAI's built-in advanced tools (such as web search), you can install additional dependencies:

# 正常安装
pip install 'crewai[tools]'

# 如果安装慢，可以使用国内镜像安装
pip install 'crewai[tools]' -i https://mirrors.aliyun.com/pypi/simple/

In China, we can use the DeepSeek large model for testing. If you don't have one yet, you need to first go tohttps://platform.deepseek.com/api_keysCreate an API key.

DeepSeek API documentation reference:https://api-docs.deepseek.com/zh-cn/。

After installation, create a new Python file, for examplemy_first_crew.py, and import the necessary libraries.

For third-party models (including DeepSeek), crewAI's LLM must go through LiteLLM at the bottom layer. Before use, we need to install:

pip install -U litellm

Next, the task will start executing and output relevant information:

After completion, the output content will be written to the python_data_cleaning_blog.md file.

The following is an explanation of the related attributes in the code.

LLM (Model Layer) Key Attributes

Agent (Intelligent Agent) Key Attributes

researcher

writer (writing)

Conclusion
In a sequential pipeline, the writing Agentshould disable delegation, otherwise it is easy to fail on re-delegation.

Key attributes of Task

Task common fields:

research_task:

write_task:

Crew (Actuator) Key Attributes

kickoff() and output structure

crewai command

We can usecrewaicommand to generate the complete project structure:

crewai create crew <项目名>

For example, we create a project example-agent-test and run the following command:

crewai create crew example-agent-test

Next, the following content appears. You can first select the large model provider:

Creating folder example_agent_test...
Cache expired or not found. Fetching provider data from the web...
Downloading  [####################################]  1102019/56339
Select a provider to set up:
1. openai
2. anthropic
3. gemini
4. nvidia_nim
5. groq
6. huggingface
7. ollama
8. watson
9. bedrock
10. azure
11. cerebras
12. sambanova
13. other
q. Quit

In this chapter, we will choose DeepSeek. First enter 13 to select other, then you can scroll with the mouse to see DeepSeek is at 23. Just enter the number 23:

If you have API keys for other models, just select according to the number.

After completion, you can see the generated directory as follows:

Project structure description:

my_project/
├── .env                # 环境变量（API Key）
├── pyproject.toml      # 项目依赖声明
├── README.md
└── src/
    └── my_project/
        ├── main.py     # 程序入口
        ├── crew.py     # Crew 与 Agent 的核心逻辑
        ├── tools/      # 自定义工具
        └── config/
            ├── agents.yaml  # Agent 定义
            └── tasks.yaml   # Task 定义

Run your Crew.

Enter the project root directory:

cd my_project

Lock and install dependencies:

crewai install

Run:

crewai run

Or directly:

python src/my_project/main.py

Configure API key

For security reasons, it is not recommended to write the API Key directly in the code; we can set it as an environment variable.

Set in code (for testing only)：

import os
os.environ["OPENAI_API_KEY"] = "你的-openai-api-key-here"

Recommended way: use.envfile

In the project root directory, create a file named.envfile.

Write the following in the file:OPENAI_API_KEY=你的-openai-api-key-here

In Python code, usepython-dotenvpackage to load.

Installation command:

pip install python-dotenv

Next, load it in the code:

from dotenv import load_dotenv
load_dotenv() # 这会自动加载 .env 文件中的变量
# 现在 os.environ["OPENAI_API_KEY"] 就已经有了值

Core Components: Detailed Explanation and Practice

Now, let's create our AI team step by step, just like building a company. We will simulate a technical blog creation team.

Step 1: Define the Agent

Agentis your team member. When creating it, you need to define several key attributes:

Let's create two agents: oneresearcherand onewriter。

Step 2: Create Task

Taskis a specific work item that needs to be assigned toAgentto complete. Key attributes include:

We create one task for each of the researcher and the writer.

Note write_taskin thecontext=[research_task], which establishes dependencies between tasks, meaning the writer needs to wait until the researcher completes the task before starting work.

Step 3: Build the Team and Set Up Processes (Crew & Process)

Now, put theAgentandTaskassemble intoCrew, and define their workProcess。

crewAI mainly supports two processes:

• Process.sequential: Tasks are executed in order according to the list. Suitable for pipeline operations with strict dependencies.
• Process.hierarchical: Combined with a manager (Manager) Agent, which coordinates and assigns tasks. Suitable for more complex, dynamic collaboration scenarios.

Step 4: Execute the task and obtain results.

Everything is ready, launch your AI team!

Run your Python script (python my_first_crew.py). Since we setverbose=Trueandverbose=2, you will see each Agent's thinking process and task execution logs in the terminal, and finally see the generated blog article.

Advanced Techniques: Equipping Agents with Tools

crewAI allowsAgentto call externalTool, such as searching the web, querying databases, performing calculations, etc.

The following example equips the researcher withweb searchandcalculationtools: