browser use Integration

browser-use Integration

The browser-use package can be wrapped as a LightAgent tool for browser-based data collection.

Use browser automation when no direct API is available, page interaction is required, or data is only visible after navigation. Prefer direct API tools when possible.

Install

pip install browser-use langchain-openai

Compatibility Helper

browser-use 0.11 and newer may expect the LangChain LLM object to expose a provider attribute. Use this helper when needed:

def ensure_browser_use_provider(llm, provider: str = "openai"):
    if hasattr(llm, "provider"):
        return llm
    try:
        setattr(llm, "provider", provider)
    except Exception:
        object.__setattr__(llm, "provider", provider)
    return llm

Tool Shape

import os
from browser_use import Agent
from langchain_openai import ChatOpenAI

async def fetch_data_with_browser(task_description: str) -> str:
    llm = ensure_browser_use_provider(ChatOpenAI(
        model=os.getenv("BROWSER_USE_MODEL", "gpt-4.1-mini"),
        api_key=os.getenv("OPENAI_API_KEY"),
        base_url=os.getenv("OPENAI_BASE_URL", "https://api.openai.com/v1"),
    ))

    browser_agent = Agent(
        task=task_description,
        llm=llm,
        use_vision=False,
    )
    result = await browser_agent.run()
    return result.final_result()

Attach tool_info and pass it as a normal LightAgent tool.

Operational Notes

• Keep browser tasks narrow and explicit.
• Use environment variables for keys.
• Expect browser tasks to be slower than API calls.
• Avoid letting arbitrary users describe unrestricted browser actions.
• Log task metadata, not secrets.

Reference example: example/08.browser_use.py.