Fix: support file_search + vector_store with sync completion() and Anthropic

- Pass tools into should_run_prompt_management_hooks and get_chat_completion_prompt
  in the sync completion path (main.py) so the vector store pre-call hook runs.
- Add tools parameter to get_chat_completion_prompt in Logging and CustomLogger.
- Implement sync get_chat_completion_prompt in VectorStorePreCallHook using
  pop_vector_stores_to_run and vector_stores.search() so file_search tools
  are stripped and context is injected before the request reaches Anthropic.
- After the hook, set tools to None when the list is empty (match async behavior).

Fixes: Unsupported tool type: file_search when using completion() with
vector_store_ids and Anthropic models (e.g. https://github.com/BerriAI/litellm/issues/22124)

Made-with: Cursor

litellm/integrations/custom_logger.py

@@ -216,6 +216,7 @@ class CustomLogger:  # https://docs.litellm.ai/docs/observability/custom_callbac
         prompt_version: Optional[int] = None,
         ignore_prompt_manager_model: Optional[bool] = False,
         ignore_prompt_manager_optional_params: Optional[bool] = False,
+        tools: Optional[List[Dict]] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Returns:

litellm/integrations/vector_store_integrations/vector_store_pre_call_hook.py

@@ -41,6 +41,72 @@ class VectorStorePreCallHook(CustomLogger):
     def __init__(self):
         super().__init__()
 
+    def get_chat_completion_prompt(
+        self,
+        model: str,
+        messages: List[AllMessageValues],
+        non_default_params: dict,
+        prompt_id: Optional[str],
+        prompt_variables: Optional[dict],
+        dynamic_callback_params: StandardCallbackDynamicParams,
+        prompt_spec: Optional[PromptSpec] = None,
+        prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
+        ignore_prompt_manager_model: Optional[bool] = False,
+        ignore_prompt_manager_optional_params: Optional[bool] = False,
+        tools: Optional[List[Dict]] = None,
+    ) -> Tuple[str, List[AllMessageValues], dict]:
+        """
+        Perform vector store search and append results as context to messages (sync).
+        Used when completion() is called so file_search tools are handled before the request.
+        """
+        try:
+            if litellm.vector_store_registry is None:
+                return model, messages, non_default_params
+
+            vector_stores_to_run: List[LiteLLM_ManagedVectorStore] = (
+                litellm.vector_store_registry.pop_vector_stores_to_run(
+                    non_default_params=non_default_params,
+                    tools=tools,
+                )
+            )
+
+            if not vector_stores_to_run:
+                return model, messages, non_default_params
+
+            query = self._extract_query_from_messages(messages)
+            if not query:
+                verbose_logger.debug(
+                    "No query found in messages for vector store search"
+                )
+                return model, messages, non_default_params
+
+            modified_messages: List[AllMessageValues] = messages.copy()
+
+            for vector_store_to_run in vector_stores_to_run:
+                vector_store_id = vector_store_to_run.get("vector_store_id", "")
+                custom_llm_provider = vector_store_to_run.get("custom_llm_provider")
+                litellm_params_for_vector_store = (
+                    vector_store_to_run.get("litellm_params", {}) or {}
+                )
+                search_response = litellm.vector_stores.search(
+                    **{
+                        "vector_store_id": vector_store_id,
+                        "query": query,
+                        "custom_llm_provider": custom_llm_provider,
+                        **litellm_params_for_vector_store,
+                    },
+                )
+                modified_messages = self._append_search_results_to_messages(
+                    messages=modified_messages, search_response=search_response
+                )
+
+            return model, modified_messages, non_default_params
+
+        except Exception as e:
+            verbose_logger.exception(f"Error in VectorStorePreCallHook: {str(e)}")
+            return model, messages, non_default_params
+
     async def async_get_chat_completion_prompt(
         self,
         model: str,

litellm/litellm_core_utils/litellm_logging.py

@@ -636,6 +636,7 @@ class Logging(LiteLLMLoggingBaseClass):
         prompt_management_logger: Optional[CustomLogger] = None,
         prompt_label: Optional[str] = None,
         prompt_version: Optional[int] = None,
+        tools: Optional[List[Dict]] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         custom_logger = (
             prompt_management_logger
@@ -645,6 +646,7 @@ class Logging(LiteLLMLoggingBaseClass):
                 prompt_id=prompt_id,
                 prompt_spec=prompt_spec,
                 dynamic_callback_params=self.standard_callback_dynamic_params,
+                tools=tools,
             )
         )
 
@@ -663,6 +665,7 @@ class Logging(LiteLLMLoggingBaseClass):
                 dynamic_callback_params=self.standard_callback_dynamic_params,
                 prompt_label=prompt_label,
                 prompt_version=prompt_version,
+                tools=tools,
             )
         self.messages = messages
         return model, messages, non_default_params

litellm/main.py

@@ -1262,7 +1262,9 @@ def completion(  # type: ignore # noqa: PLR0915
 
     if isinstance(litellm_logging_obj, LiteLLMLoggingObj) and (
         litellm_logging_obj.should_run_prompt_management_hooks(
-            prompt_id=prompt_id, non_default_params=non_default_params
+            prompt_id=prompt_id,
+            non_default_params=non_default_params,
+            tools=tools,
         )
     ):
         (
@@ -1277,7 +1279,17 @@ def completion(  # type: ignore # noqa: PLR0915
             prompt_variables=prompt_variables,
             prompt_label=kwargs.get("prompt_label", None),
             prompt_version=kwargs.get("prompt_version", None),
+            tools=tools,
         )
+        #########################################################
+        # if the chat completion logging hook removed all tools,
+        # set tools to None
+        # eg. in certain cases when users send vector stores as tools
+        # we don't want the tools to go to the upstream llm
+        # relevant issue: https://github.com/BerriAI/litellm/issues/11404
+        #########################################################
+        if tools is not None and len(tools) == 0:
+            tools = None
 
     ### LITELLM SYSTEM PROMPT ###
     if litellm_system_prompt: