Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.
Python ai-agentsai-toolsbrowser-agentbrowser-automationbrowser-usecdpcloud-browsercomputer-useheadless-browserllmmcpplaywrightpythonremote-browserstealth-browserweb-automationweb-scraping
Dont cache the first step
Magnus Müllerfbba5ba035e6
diff --git a/browser_use/agent/service.py b/browser_use/agent/service.pyindex 31a6f53bb..ec4038f15 100644--- a/browser_use/agent/service.py+++ b/browser_use/agent/service.py@@ -644,14 +644,10 @@ class Agent(Generic[Context, AgentStructuredOutput]):self.logger.debug(f'🌐 Step {self.state.n_steps}: Getting browser state...')# Always take screenshots for all steps- # Use caching based on directly_open_url setting - if directly_open_url is False, don't use cached state- is_first_step = self.state.n_steps in (0, 1)- use_cache = is_first_step and self.directly_open_url- self.logger.debug(f'📸 Requesting browser state with include_screenshot=True, cached={use_cache}')+ self.logger.debug('📸 Requesting browser state with include_screenshot=True')browser_state_summary = await self.browser_session.get_browser_state_summary(cache_clickable_elements_hashes=True,include_screenshot=True, # always capture even if use_vision=False so that cloud sync is useful (it's fast now anyway)- cached=use_cache,include_recent_events=self.include_recent_events,)if browser_state_summary.screenshot: