How to use crewAI: install, build agents and run a crew

What does CrewAI do?

CrewAI is described in its README as an open-source Python framework with high-level abstractions and low-level APIs for building multi-agent workflows. It has two main parts: Crews, where role-based AI agents collaborate autonomously, and Flows, which are event-driven workflows that give precise control over execution paths and state.

The project is aimed at developers building multi-step automations with LLMs. It is not a no-code tool.

• Crews: teams of agents with defined roles, goals and expertise that collaborate on a task.
• Flows: production-ready, event-driven workflows with fine-grained control, state management between tasks, and native support for Crews.

What you need before you install CrewAI

The README lists Python as the prerequisite and installs the package from PyPI. Because CrewAI is a Python framework, you need to be comfortable writing and running Python code. The README does not list other specific requirements, so check the project documentation for the Python version it supports.

Install CrewAI with pip

Install the package from PyPI using pip, as shown in the project's README.

Create your first crew: agents, tasks and a run

The README's Getting Started section walks through three stages: installation, setting up your crew, and running your crew. Setting up a crew means defining the agents, each with a role and goal, and the work they will do together. The README points to its Quick Tutorial and the official docs at docs.crewai.com for the exact project scaffolding and code.

Agent roles and goals are the core of the design. Describe each agent's specialty clearly so the crew can divide work sensibly.

• Install CrewAI and confirm it imports in your Python environment.
• Define one or more agents, each with a role and goal.
• Define the tasks the crew should complete.
• Run the crew and review the output.

Choose between Crews and Flows

Use Crews when you want agents to decide how to divide and carry out work. Use Flows when you need tighter control over execution paths and state. The README says the two approaches work together, so a Flow can also make use of Crews.

The README's best-fit list says CrewAI suits developers who want both autonomous agent teams and step-by-step control via Flows, and builders who want low-level control over prompts and execution logic.

• Pick Crews for autonomous, role-based collaboration.
• Pick Flows for precise, event-driven control and consistent state.
• Combine both when a workflow needs structure around some autonomous steps.

Connect your crew to a model and use the build tools

The README includes a section titled Connecting Your Crew to a Model, but the source provided here does not include its contents, so consult the docs for model configuration details.

If you use an AI coding agent, the README says CrewAI Skills can be installed to teach it CrewAI patterns. For Claude Code the project lists a plugin marketplace, and for Cursor, Codex, Windsurf and others it lists the npx skills command shown in its README.