CrewAI

Use Lightcone as a CrewAI tool to give your agents the ability to browse the web, fill forms, take screenshots, and extract data.

Install

pip install crewai tzafon

Example

from crewai import Agent, Task, Crew
from crewai.tools import BaseTool
from tzafon import Lightcone


class LightconeBrowserTool(BaseTool):
    name: str = "browser"
    description: str = (
        "Browse a website and return a screenshot URL. "
        "Input should be a URL to visit."
    )

    def _run(self, url: str) -> str:
        client = Lightcone()
        with client.computer.create(kind="browser") as computer:
            computer.navigate(url)
            computer.wait(2)
            result = computer.screenshot()
            return computer.get_screenshot_url(result)


# Create an agent with the browser tool
researcher = Agent(
    role="Web Researcher",
    goal="Find information on websites",
    backstory="You are a skilled web researcher.",
    tools=[LightconeBrowserTool()],
)

task = Task(
    description="Visit https://news.ycombinator.com and describe the top 3 stories",
    expected_output="A summary of the top 3 stories on Hacker News",
    agent=researcher,
)

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

How it works

The integration wraps Lightcone’s computer actions into a CrewAI BaseTool. When a CrewAI agent decides it needs to visit a website, it calls the tool, which:

1. Creates a Lightcone computer session
2. Navigates to the URL
3. Takes a screenshot and returns the URL
4. Terminates the session

You can extend the tool to support more actions (clicking, typing, form filling) based on your workflow needs.

Resources

• CrewAI custom tools documentation
• CrewAI GitHub

See also

• LangChain — document loader and computer tool for LangChain
• Vercel AI SDK — Lightcone tools for TypeScript workflows
• Run a task — use Lightcone’s built-in Northstar instead