From c9f0fa5af7cba8ceeaac3d6f7b032c40a31de292 Mon Sep 17 00:00:00 2001
From: Ishaan Jaffer <ishaanjaffer0324@gmail.com>
Date: Mon, 23 Feb 2026 20:44:03 -0800
Subject: [PATCH] refactor(realtime): call apply_guardrail directly, remove
 dedicated hook method
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The async_realtime_input_transcription_hook in CustomGuardrail and
ContentFilterGuardrail was just a thin wrapper that called apply_guardrail —
the same interface used by /chat and /messages. Remove the wrapper and call
apply_guardrail directly from run_realtime_guardrails, keeping the pattern
consistent across all endpoints.
---
 litellm/integrations/custom_guardrail.py         | 15 ---------------
 litellm/litellm_core_utils/realtime_streaming.py |  8 ++++----
 .../litellm_content_filter/content_filter.py     | 16 ----------------
 3 files changed, 4 insertions(+), 35 deletions(-)

diff --git a/litellm/integrations/custom_guardrail.py b/litellm/integrations/custom_guardrail.py
index 18590a6b73b..bf330944ef8 100644
--- a/litellm/integrations/custom_guardrail.py
+++ b/litellm/integrations/custom_guardrail.py
@@ -334,21 +334,6 @@ class CustomGuardrail(CustomLogger):
 
         return kwargs
 
-    async def async_realtime_input_transcription_hook(
-        self,
-        transcription: str,
-        user_api_key_dict: Optional[Any],
-        session_id: Optional[str] = None,
-    ) -> None:
-        """
-        Called when a user's voice transcription completes in a Realtime API session,
-        before the LLM generates a response.
-
-        Raise an exception to block the response.
-        Return None to allow the LLM to respond.
-        """
-        return None
-
     async def async_post_call_success_deployment_hook(
         self,
         request_data: dict,
diff --git a/litellm/litellm_core_utils/realtime_streaming.py b/litellm/litellm_core_utils/realtime_streaming.py
index e8f04901f7c..1599420f20a 100644
--- a/litellm/litellm_core_utils/realtime_streaming.py
+++ b/litellm/litellm_core_utils/realtime_streaming.py
@@ -155,10 +155,10 @@ class RealTimeStreaming:
             ):
                 continue
             try:
-                await callback.async_realtime_input_transcription_hook(
-                    transcription=transcript,
-                    user_api_key_dict=self.user_api_key_dict,
-                    session_id=item_id,
+                await callback.apply_guardrail(
+                    inputs={"texts": [transcript], "images": []},
+                    request_data={"user_api_key_dict": self.user_api_key_dict},
+                    input_type="request",
                 )
             except Exception as e:
                 # Extract the human-readable error from HTTPException detail dict,
diff --git a/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py b/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py
index abdc1482c43..f653d4602d9 100644
--- a/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py
+++ b/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py
@@ -1921,22 +1921,6 @@ class ContentFilterGuardrail(CustomGuardrail):
             # We already reached the end of the generator
             pass
 
-    async def async_realtime_input_transcription_hook(
-        self,
-        transcription: str,
-        user_api_key_dict,
-        session_id=None,
-    ) -> None:
-        """
-        Run content filter checks on a Realtime API speech transcription.
-        Raises ValueError if the transcription contains blocked content.
-        """
-        await self.apply_guardrail(
-            inputs={"texts": [transcription], "images": []},
-            request_data={},
-            input_type="request",
-        )
-
     @staticmethod
     def get_config_model():
         from litellm.types.proxy.guardrails.guardrail_hooks.litellm_content_filter import (