Web search

from browser_use import Agent

agent = Agent('Find the current version of the browser-use package on PyPI', web_search=True)
result = agent.run_sync()
print(result.final_result(), result.cost_usd)  # cost_usd includes the searches

search(query, num_results=None, category=None) -> {'query', 'count', 'results'}

• results holds one text entry per hit: title, URL and matching excerpts. The agent opens pages in the browser for their full content.
• num_results: default 8, max 20.
• category: 'company', 'people', 'news' or 'publication'.

​

Billing and keys

• Billed per search to your Browser Use key. Each charge is added to result.cost_usd.
• It uses the gateway model’s key, else your Browser Use key. Without one, run() raises before any browser starts.
• Before the browser starts, a free check confirms the search endpoint answers.
• The key never enters the agent’s Python worker: the parent process makes the request.

​

Notes

• Off by default. When off, the system prompt is unchanged.
• It can’t be combined with Browser(allowed_domains=...): run() raises a ValueError before any browser starts, because search would reach sites the browser may not open.