CrewAI Documentation

Introduction

CrewAI is a robust multi-agent orchestration framework that enables developers to build complex AI agent systems by defining roles, goals, and tasks for each agent.

Installation

pip install crewai

Quick Start

Define Your Agents

from crewai import Agent

researcher = Agent(
    role='Senior Research Analyst',
    goal='Uncover cutting-edge developments in AI and data science',
    backstory="""You are an expert at understanding the nuances of AI and data science.
    You are skilled at turning complex data into actionable insights.""",
    verbose=True,
    allow_delegation=False,
)

Define Your Tasks

from crewai import Task

task = Task(
    description='''Analyze the latest developments in AI and data science.''',
    expected_output='A detailed report on the latest AI developments.',
    agent=researcher,
)

Create Your Crew

from crewai import Crew

crew = Crew(
    agents=[researcher],
    tasks=[task],
    verbose=2,
)

result = crew.kickoff()

Process Types

Process	Description
sequential	Tasks are executed in the order they are defined.
hierarchical	A manager agent assigns tasks to other agents.
consensual	All agents must agree on task assignments.

Features

Role-Based Agent Definition

Each agent has a specific role, goal, and backstory that guides its behavior.

Task Delegation

Agents can delegate tasks to other agents based on their expertise.

Process Management

Choose from sequential, hierarchical, or consensual process models.

Built-in Tool Integration

Integrate with tools like Google Search, Serper, and more.

Agent Memory

Agents can remember past interactions and context.

Use Cases

CrewAI excels in scenarios requiring role-based multi-agent collaboration:

Use Case	Why CrewAI
Content Production Pipeline	Assign roles: Researcher → Writer → Editor → Publisher
Market Research	Multiple analysts with different focus areas collaborate
Customer Support Automation	Triage → Research → Response → Follow-up agents
Data Analysis Workflows	Data extractor → Analyst → Visualizer → Reporter
Code Review Pipeline	Linter → Security auditor → Performance reviewer

Pros & Cons

✅ Pros

• Intuitive role-based design — Natural fit for real-world team structures
• Built-in task delegation — Agents can automatically delegate to specialists
• Strong documentation — Well-maintained docs with many examples
• Active community — Large user base, frequent updates, good support
• Tool ecosystem — Extensive built-in tools and easy custom tool creation
• Memory support — Agents can retain context across conversations
• Hierarchical process — Manager agent for complex task coordination

❌ Cons

• Python-only — No native support for JavaScript/TypeScript
• Learning curve for advanced features — Hierarchical and consensual processes require deeper understanding
• Limited streaming support — Not as mature as LangGraph for real-time streaming
• State management — Less explicit than LangGraph's TypedDict approach
• No built-in persistence — Requires external checkpointing for long-running workflows

Comparison with Alternatives

Feature	CrewAI	LangGraph	AutoGen
Paradigm	Role-based teams	Graph-based workflows	Conversation-centric
Best for	Content pipelines, research teams	Complex stateful workflows	Multi-agent conversations
Learning curve	Low-Medium	Medium	Low
Streaming	Limited	Native	Basic
Persistence	External	Built-in	Conversation history
Human-in-the-loop	Basic	Advanced	Built-in

Real-World Examples

Example 1: AI Content Creation Team

from crewai import Agent, Task, Crew

# Define specialized agents
researcher = Agent(
    role='Market Research Analyst',
    goal='Find the latest trends and data in the AI industry',
    backstory='Expert analyst with 10+ years of experience in tech trends',
    verbose=True,
)

writer = Agent(
    role='Technical Content Writer',
    goal='Transform research into engaging blog posts',
    backstory='Senior writer specializing in AI and developer tools',
    verbose=True,
)

editor = Agent(
    role='Content Editor',
    goal='Review and polish content for publication',
    backstory='Detail-oriented editor with expertise in technical content',
    verbose=True,
)

# Define tasks
research_task = Task(
    description='Research the top 5 AI agent frameworks and their latest updates',
    expected_output='A comprehensive research report with citations',
    agent=researcher,
)

write_task = Task(
    description='Write a blog post comparing the top AI agent frameworks',
    expected_output='A well-structured blog post ready for review',
    agent=writer,
    context=[research_task],
)

edit_task = Task(
    description='Review and edit the blog post for clarity and accuracy',
    expected_output='A polished, publication-ready blog post',
    agent=editor,
    context=[write_task],
)

# Create and run the crew
crew = Crew(
    agents=[researcher, writer, editor],
    tasks=[research_task, write_task, edit_task],
    process='sequential',
)

result = crew.kickoff()

Example 2: Autonomous Research Agent

from crewai import Agent, Task, Crew, Tool
from crewai_tools import SerperDevTool

# Create a search tool
search_tool = SerperDevTool()

researcher = Agent(
    role='AI Research Specialist',
    goal='Conduct deep research on any AI topic',
    backstory='PhD-level researcher with expertise in AI systems',
    tools=[search_tool],
    verbose=True,
    allow_delegation=True,
)

research_task = Task(
    description='Research the current state of MCP (Model Context Protocol)',
    expected_output='A comprehensive research report with sources',
    agent=researcher,
)

crew = Crew(agents=[researcher], tasks=[research_task])
result = crew.kickoff()

Best Practices

1. Define clear roles and goals - Each agent should have a well-defined purpose.
2. Write detailed backstories - Helps guide agent behavior and improves output quality.
3. Use task context - Pass outputs between tasks for coherent workflows.
4. Enable delegation - Let agents collaborate on complex problems.
5. Use verbose mode - For debugging and understanding agent decision-making.
6. Start with sequential - Easier to debug, then move to hierarchical for complex workflows.
7. Set proper expected_output - Clear output specifications improve task quality.

Troubleshooting

Issue	Solution
Agents not delegating	Check allow_delegation=True on the agent
Poor output quality	Improve backstories and task descriptions
Infinite loops	Set max_iter on agents, reduce delegation
Slow execution	Use smaller models for simple tasks, enable caching
Memory issues	Clear agent memory between runs or use smaller context

Resources

• CrewAI Documentation
• GitHub Repository
• Examples Gallery
• CrewAI Tools
• Discord Community

---

Last updated: June 2026