How to Create a Multi-Agent Team with CrewAI

What you need

Python 3.10+ installed.

CrewAI installed: `pip install crewai`.

An API key from your model provider.

Step 1: Define your agents with roles

Each CrewAI agent has a role, goal, backstory, and toolset. Think of them as employees with job descriptions.

Example: Researcher agent — role is 'Senior Research Analyst', goal is 'find and summarize authoritative sources'.

Define 3-5 agents for a typical crew. Keep roles distinct to avoid overlap.

Step 2: Assign tasks to agents

Tasks have descriptions, expected outputs, and assigned agents.

Researcher task: 'Research the topic and provide a brief with 5-7 cited sources.'

Writer task: 'Write a 1000-word article based on the research brief.'

Editor task: 'Review the article for quality, accuracy, and brand alignment.'

Step 3: Configure the crew workflow

Tasks can run sequentially or in parallel. Sequential is best for content production where each step depends on the previous.

Set the process type: sequential for linear workflows, hierarchical for manager-led teams.

Define task dependencies so the writer waits for the researcher, and the editor waits for the writer.

Step 4: Set tool permissions per agent

Researcher gets web search and content retrieval tools.

Writer gets document generation and formatting tools.

Editor gets review and quality check tools.

No agent should have tools it does not need for its specific role.

Step 5: Run the crew and review output

Initialize the crew with `Crew(agents=agents, tasks=tasks, process=Process.sequential)`.

Run with `crew.kickoff()` and watch the agents work through their tasks.

Review each agent's output. CrewAI provides detailed logs of agent reasoning and actions.

FAQ

Create a CrewAI Team

1. Install CrewAI: pip install crewai
2. Define agent roles, goals, and backstories
3. Assign tasks to specific agents
4. Configure sequential or hierarchical workflow
5. Set tool permissions per agent role
6. Run the crew and review outputs