Skip to content

Browser Use agent opens blank Chrome windows and does not navigate #3

@00hello

Description

@00hello

Description:
When attempting to run tasks using the browser-use library (version 0.1.40), the agent launches Chrome (or Chromium) windows that remain blank (displaying about:blank or an empty page) and does not proceed with the instructed navigation or actions. This occurs even with very simple tasks like "Go to example.com". The agent may appear to hang or get stuck in a loop, repeatedly opening new blank windows if not constrained by max_steps.

This issue has been observed on macOS Sequioa 15.4.1 (24E263) with Python 3.13.3 and browser-use==0.1.40. The problem persists even when running the library's own example.py.

Current Behavior:

  • Chrome windows are launched by Browser Use.
  • These windows remain blank and do not navigate to the target URL.
  • The agent script does not progress with the task.
  • If max_steps is not set or is high, multiple blank windows may be opened sequentially.
  • The test script using fetch_round_guesses (which internally uses Browser Use) times out or is manually interrupted due to this behavior.

Expected Behavior:

  • Chrome should launch and navigate to the URL specified or implied by the task.
  • The agent should interact with the web page as instructed.
  • The task should complete, and the agent should return the requested information.

Troubleshooting Steps Taken So Far:

  • Verified Chrome installation and path (/Applications/Google Chrome.app/Contents/MacOS/Google Chrome).
  • Simplified the task to basic navigation (e.g., "Go to example.com").
  • Set max_steps for the agent to prevent infinite loops.
  • Attempted to connect to a manually launched Chrome instance with --remote-debugging-port=9222 (failed due to BrowserConfig not supporting remote_debugging_port argument directly).
  • Currently in the process of reinstalling Playwright and its browser drivers (python -m playwright install --with-deps chromium) and enabling debug logging (BROWSER_USE_LOGGING_LEVEL=debug) as per suggestions in GitHub Issue #1020, which describes similar symptoms.

Environment:

  • browser-use version: 0.1.40
  • Python version: 3.13.3
  • Operating System: macOS Sequioa 15.4.1 (24E263)
  • Playwright version: [Run python -m pip show playwright to get this]
  • LLM Model used: GPT-4o (though the issue seems to occur before significant LLM interaction for navigation).

Next Steps:

  • Complete Playwright reinstallation and driver setup.
  • Run tests with BROWSER_USE_LOGGING_LEVEL=debug enabled to gather more detailed logs.
  • Investigate potential conflicts with other installed browser automation tools or Chrome extensions (if any).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions