Automate browser based workflows with AI
Python aiapiautomationbrowserbrowser-automationcomputergptllmplaywrightpowerautomatepuppeteerpythonrpaseleniumvisionworkflow
Use gemini 2_5 flash lite for create_extract_action (#3429)
LawyZheng66b2004b7022
diff --git a/skyvern/forge/agent.py b/skyvern/forge/agent.pyindex c78351a8f..3185f4d33 100644--- a/skyvern/forge/agent.py+++ b/skyvern/forge/agent.py@@ -3020,7 +3020,7 @@ class ForgeAgent:local_datetime=datetime.now(context.tz_info).isoformat(),)- data_extraction_summary_resp = await app.SECONDARY_LLM_API_HANDLER(+ data_extraction_summary_resp = await app.EXTRACTION_LLM_API_HANDLER(prompt=prompt, step=step, prompt_name="data-extraction-summary")return ExtractAction(diff --git a/skyvern/forge/sdk/api/llm/config_registry.py b/skyvern/forge/sdk/api/llm/config_registry.pyindex 44d155f69..cbf8bb978 100644--- a/skyvern/forge/sdk/api/llm/config_registry.py+++ b/skyvern/forge/sdk/api/llm/config_registry.py@@ -1097,6 +1097,25 @@ if settings.ENABLE_VERTEX_AI and settings.VERTEX_CREDENTIALS:),),)+ LLMConfigRegistry.register_config(+ "VERTEX_GEMINI_2.5_FLASH_LITE",+ LLMConfig(+ "vertex_ai/gemini-2.5-flash-lite",+ ["VERTEX_CREDENTIALS"],+ supports_vision=True,+ add_assistant_prefix=False,+ max_completion_tokens=65535,+ litellm_params=LiteLLMParams(+ vertex_credentials=settings.VERTEX_CREDENTIALS,+ api_base=f"{api_base}/gemini-2.5-flash-lite" if api_base else None,+ vertex_location=settings.VERTEX_LOCATION,+ thinking={+ "budget_tokens": settings.GEMINI_THINKING_BUDGET,+ "type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,+ },+ ),+ ),+ )LLMConfigRegistry.register_config("VERTEX_GEMINI_2.5_FLASH_PREVIEW",LLMConfig(diff --git a/skyvern/webeye/actions/handler.py b/skyvern/webeye/actions/handler.pyindex 4cdea6fc4..017680ea1 100644--- a/skyvern/webeye/actions/handler.py+++ b/skyvern/webeye/actions/handler.py@@ -3603,7 +3603,10 @@ async def extract_information_for_navigation_goal(# CUA tasks should use the default data extraction llm keyllm_key_override = None- llm_api_handler = LLMAPIHandlerFactory.get_override_llm_api_handler(llm_key_override, default=app.LLM_API_HANDLER)+ # Use the appropriate LLM handler based on the feature flag+ llm_api_handler = LLMAPIHandlerFactory.get_override_llm_api_handler(+ llm_key_override, default=app.EXTRACTION_LLM_API_HANDLER+ )json_response = await llm_api_handler(prompt=extract_information_prompt,step=step,