Multi Agent System Crew

Posted on Mar 30, 2026

CrewAI

$ uv tool install crewai --upgrade

Available providers:

  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

I will use gemini provider:

$ uv add "crewai[google-genai]"

File: .env

MODEL=gemini/gemini-2.5-flash
GEMINI_API_KEY=AIz...

Get the API key here: https://aistudio.google.com/api-keys

Crew

Docs: https://docs.crewai.com/en/concepts/crews

A crew in crewAI represents a collaborative group of agents working together to achieve a set of tasks. Each crew defines the strategy for task execution, agent collaboration, and the overall workflow.

crew = Example().crew()
crew.kickoff(inputs={'topic': 'AI Agents'})

Agents

Docs: https://docs.crewai.com/en/concepts/agents

An agent is an autonomous unit that can:

  • Perform specific tasks
  • Make decisions based on its role and goal
  • Use tools to accomplish objectives
  • Communicate and collaborate with other agents
  • Maintain memory of interactions
  • Delegate tasks when allowed

There are two ways to create agents in CrewAI: using YAML configuration (recommended) or defining them directly in code.

researcher:
  role: >
    {topic} Senior Data Researcher    
  goal: >
    Uncover cutting-edge developments in {topic}    
  backstory: >
    You're a seasoned researcher with a knack for uncovering the latest
    developments in {topic}. Known for your ability to find the most relevant
    information and present it in a clear and concise manner.

Variables in your YAML files (e.g.: {topic}) will be replaced with values then inputs dict when running the crew:

@CrewBase
class Example():
    """Example crew"""
  
    @agent
    def researcher(self) -> Agent:
        return Agent(
            config=self.agents_config['researcher'], # type: ignore[index]
            verbose=True,
            tools=[SerperDevTool()]
        )
    
    @crew
    def crew(self) -> Crew:
        """Creates the Example crew"""

        return Crew(
            agents=self.agents, # Automatically created by the @agent decorator
            tasks=self.tasks, # Automatically created by the @task decorator
            process=Process.sequential,
            verbose=True,
        )

Tasks

Docs: https://docs.crewai.com/en/concepts/tasks

from crewai import Task

task = Task(
    description="Evaluate the quality of the given text and provide recommendations.",
    expected_output="A score and a list of recommendations.",
    agent=agent,
    output_pydantic=AgentOutput
)

Flows

Docs: https://docs.crewai.com/en/concepts/flows

Human-in-the-Loop

The @human_feedback decorator enables human-in-the-loop workflows by pausing flow execution to collect feedback from a human. This is useful for approval gates, quality review, and decision points that require human judgment.

Skills

Docs: https://docs.crewai.com/en/concepts/skills

Skills are self-contained directories that provide agents with domain-specific instructions, references, and assets. Each skill is defined by a SKILL.md file with YAML frontmatter and a markdown body.

---
name: my-skill
description: Short description of what this skill does and when to use it.
license: Apache-2.0                    # optional
compatibility: crewai>=0.1.0           # optional
metadata:                              # optional
  author: your-name
  version: "1.0"
allowed-tools: web-search file-read    # optional, space-delimited
---

Instructions for the agent go here. This markdown body is injected
into the agent's prompt when the skill is activated.

Tools

Docs: https://docs.crewai.com/en/concepts/tools

$ pip install 'crewai[tools]'

from crewai_tools import (
    DirectoryReadTool,
    FileReadTool,
    SerperDevTool,
    WebsiteSearchTool
)

Custom tools

from crewai.tools import BaseTool
from typing import Type
from pydantic import BaseModel, Field


class MyCustomToolInput(BaseModel):
    """Input schema for MyCustomTool."""
    argument: str = Field(..., description="Description of the argument.")

class MyCustomTool(BaseTool):
    name: str = "Name of my tool"
    description: str = (
        "Clear description for what this tool is useful for, your agent will need this information to use it."
    )
    args_schema: Type[BaseModel] = MyCustomToolInput

    def _run(self, argument: str) -> str:
        # Implementation goes here
        return "this is an example of a tool output, ignore it and move along."

References