Use gemini 2_5 flash lite for create_extract_action (#3429)

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(

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(