crewai_tools_browser_use

Overview

With the rise of Large Language Models (LLMs), AI agents are transforming productivity across various fields. Typically seen as standalone tools, these agents can achieve even more when working together. This repository explores the collaboration between two distinct agents: the CrewAI , which excels in generation and planning, and the Browser Use, specialized in GUI-based automation tasks. By connecting their strengths, we can create a powerful synergy where one agent's capabilities complement the other's limitations. This initial trial demonstrates how heterogeneous agents can work together seamlessly, unlocking new possibilities for complex tasks that require both creativity and precision.

1. Browser Use Tool

BrowserUseTool (browser_use_tool.py) is a standard CrewAI toolk implementation. It can be placed in the src/tools. This tool receives an instruction and sends it to the browser-use agent. Since GUI automation tasks can take a fairly long time, it submits the task and performs rolling polling to check if the task has finished and produced results.

2. Browser Use Service

BrowserUseService (browser_use_service.py) is the backend component responsible for executing web automation task. The service can only work on one specific task at a time. If other tasks are submitted, they will be pending until the current task is finished.

sequenceDiagram
    participant CrewAI_Agent as CrewAI Agent
    participant BrowserUseService as Browser Use Service
    participant BrowserUseAgent as Browser Use Agent

    CrewAI_Agent->>BrowserUseService: Submit Task with Instructions
    BrowserUseService-->>CrewAI_Agent: Return task_id

    BrowserUseService->>BrowserUseAgent: Submit Task for Automation
    loop Every 2 seconds
        CrewAI_Agent->>BrowserUseService: Query Status with task_id
        alt Task is not completed
            BrowserUseService-->>CrewAI_Agent: Status: In Progress
        else Task is completed
            BrowserUseAgent-->>BrowserUseService: Return Results
            BrowserUseService-->>CrewAI_Agent: Status: Completed
            BrowserUseService-->>CrewAI_Agent: Return Results
        end
    end

Loading

Customization

You may need do some customization by yourself, to set up .env file, the browser configuration etc. The service default port is 4999. Following is an example .env file of browser_use_service.py. Place them in the same folder.

OPENAI_API_KEY=<YOUR_OPENAI_API_KEY>
MODEL_NAME=gpt-4o-mini
USERNAME=user
PASSWORD=password

Here, USERNAME and PASSWORD are sensitive_data will be passed to BrowserUse Agent.

sensitive_data = {
    'x_name': os.environ['USERNAME'],
    'x_password': os.environ['PASSWORD']
}

BrowserUse Installation

Please refer to the Browser Use page.

With pip (Python>=3.11):

pip install browser-use

install playwright:

playwright install