feat(proxy): add per-deployment force_websearch_model setting

When a request contains only web_search tools (e.g. Claude Code's
step-2 contextless search request), redirect it to the model group
specified in the deployment's `force_websearch_model` litellm_param.

litellm/proxy/common_request_processing.py

@@ -32,6 +32,7 @@ from litellm.constants import (
     STREAM_SSE_DATA_PREFIX,
 )
 from litellm.integrations.custom_guardrail import CustomGuardrail
+from litellm.integrations.websearch_interception.tools import is_web_search_tool
 from litellm.litellm_core_utils.dd_tracing import NullTracer, tracer
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
 from litellm.litellm_core_utils.llm_response_utils.get_headers import (
@@ -922,6 +923,23 @@ class ProxyBaseLLMRequestProcessing:
         ):
             self.data["model"] = user_api_key_dict.aliases[self.data["model"]]
 
+        ### WEB SEARCH REDIRECT (per-deployment force) ###
+        # if request only contains a web_search tool and has any deployment that has
+        # `force_websearch_model`, redirect to that model.
+        if (
+            isinstance(self.data.get("model"), str)
+            and llm_router is not None
+            and self.data.get("tools")
+            and all(
+                isinstance(t, dict) and is_web_search_tool(t)
+                for t in self.data["tools"]
+            )
+        ):
+            for dep in llm_router.get_model_list(model_name=self.data["model"]) or []:
+                if model := dep.get("litellm_params", {}).get("force_websearch_model"):
+                    self.data["model"] = model
+                    break
+
         self.data["litellm_call_id"] = request.headers.get(
             "x-litellm-call-id", str(uuid.uuid4())
         )

litellm/types/router.py

@@ -222,6 +222,7 @@ class GenericLiteLLMParams(CredentialLiteLLMParams, CustomPricingLiteLLMParams):
     use_chat_completions_api: Optional[bool] = None
     model_config = ConfigDict(extra="allow", arbitrary_types_allowed=True)
     merge_reasoning_content_in_choices: Optional[bool] = False
+    force_websearch_model: Optional[str] = None
     model_info: Optional[Dict] = None
     mock_response: Optional[Union[str, ModelResponse, Exception, Any]] = None
 
@@ -351,6 +352,8 @@ class LiteLLMParamsTypedDict(TypedDict, total=False):
     drop_params: Optional[bool]
     ## RESPONSES API → CHAT COMPLETIONS BRIDGE ##
     use_chat_completions_api: Optional[bool]
+    ## WEB SEARCH REDIRECT ##
+    force_websearch_model: Optional[str]
     ## UNIFIED PROJECT/REGION ##
     region_name: Optional[str]
     ## VERTEX AI ##