browser-use stuck at 'Step 1' with blank Chromium page, what causes it and how to fix it?

6 answers

• Best answerAgentArmor teamTeam3d ago

  The thread was closed without a single confirmed root cause. Commenters reported three different causes: a mismatched Playwright/Chromium install, an OpenAI quota error that was not shown in the logs, and a model without tool-calling support.

  1. Playwright install mismatch. The README's playwright install line installs browser binaries, not the Playwright package. One user had installed Playwright with npm, which led to a Chromium version-not-found error. They ran npm uninstall playwright and npm uninstall playwright -g, then pip install playwright and playwright install. The thread does not confirm whether this resolved the loop.

  2. Hidden API errors. Several users hit RateLimitError: 429 insufficient_quota from OpenAI. browser-use kept retrying the steps without printing the error. A maintainer asked reporters to add BROWSER_USE_LOGGING_LEVEL=debug to .env and look for Agent paused in the logs. No commenter confirmed seeing that message.

  3. Model without tool calling. A user whose DeepSeek-V3 vendor did not support tool calls, and another using llama3, saw the loop. Switching to a vendor that supports tool calls, or to llama3.1, fixed it for them.

  General advice: check your API quota and confirm the model supports tool calling before reinstalling Playwright.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @illtellyoulater tried on GitHub (Outcome unclear)

  Removed the npm-installed Playwright with npm uninstall playwright and npm uninstall playwright -g, then installed it with pip install playwright and ran playwright install.

  Source: https://github.com/browser-use/browser-use/issues/1020#issuecomment-2722695671 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @Buzimu tried on GitHub (Worked)

  Switched from llama3, which does not support tool calling, to llama3.1, and the agent started producing steps.

  Source: https://github.com/browser-use/browser-use/issues/1020#issuecomment-2731379921 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @MaoMao0350 tried on GitHub (Outcome unclear)

  Suggested setting a breakpoint in browser_use/agent/service.py near line 373 to inspect the exception. The commenter reported the RateLimitError (429, insufficient_quota) but did not say the breakpoint revealed it.

  Source: https://github.com/browser-use/browser-use/issues/1020#issuecomment-2723344034 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @pppp606 tried on GitHub (Outcome unclear)

  Maintainer asked the reporter to add BROWSER_USE_LOGGING_LEVEL=debug to .env and look for Agent paused in the logs. No commenter confirmed seeing that message.

  Source: https://github.com/browser-use/browser-use/issues/1020#issuecomment-2723380273 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What a GitHub user tried on GitHub (Didn't work)

  Installed browser-use per the README in a conda env with Python 3.11, saved OPENAI_API_KEY in .env, and ran example.py and examples/simple.py. Both opened a blank Chromium page and looped on Step 1.

  Source: https://github.com/browser-use/browser-use/issues/1020 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport