From 32cb6f0cd91adfb5554be251490b9b5d5ea60a19 Mon Sep 17 00:00:00 2001
From: Jonathan Barazany <jry157@gmail.com>
Date: Fri, 20 Mar 2026 01:07:20 +0200
Subject: [PATCH] fix: guard short-circuit against providers with native
 agentic loop
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Skip short-circuit for providers that have a BaseAnthropicMessagesConfig
  (bedrock, vertex_ai, azure_ai, anthropic) — they use the agentic loop
  which includes a follow-up LLM synthesis step. Short-circuiting would
  return raw search text instead of an LLM-synthesized answer.
- Add fallback to litellm.get_llm_provider() for custom_llm_provider
  derivation when litellm_params is overwritten by kwargs.
- Add test for bedrock guard.

Addresses Greptile review comments #3 and #4.
---
 .../websearch_interception/handler.py         | 23 +++++++++++++++++++
 .../messages/handler.py                       |  9 +++++++-
 .../test_websearch_short_circuit.py           | 23 +++++++++++++++++++
 3 files changed, 54 insertions(+), 1 deletion(-)

diff --git a/litellm/integrations/websearch_interception/handler.py b/litellm/integrations/websearch_interception/handler.py
index b557d8c0e77..34396849e24 100644
--- a/litellm/integrations/websearch_interception/handler.py
+++ b/litellm/integrations/websearch_interception/handler.py
@@ -29,6 +29,7 @@ from litellm.types.integrations.websearch_interception import (
     WebSearchInterceptionConfig,
 )
 from litellm.types.utils import LlmProviders
+from litellm.utils import ProviderConfigManager
 
 
 class WebSearchInterceptionLogger(CustomLogger):
@@ -106,6 +107,28 @@ class WebSearchInterceptionLogger(CustomLogger):
         ):
             return None
 
+        # Only short-circuit for providers without native Anthropic Messages
+        # support.  Providers that have a BaseAnthropicMessagesConfig (bedrock,
+        # vertex_ai, azure_ai, anthropic) already use the agentic loop, which
+        # includes a follow-up LLM call to synthesize the answer from search
+        # results.  Short-circuiting those would skip that synthesis step and
+        # return raw search text — a regression for existing users.
+        try:
+            provider_enum = LlmProviders(provider_str)
+            anthropic_config = (
+                ProviderConfigManager.get_provider_anthropic_messages_config(
+                    model=model, provider=provider_enum
+                )
+            )
+            if anthropic_config is not None:
+                verbose_logger.debug(
+                    f"WebSearchInterception: Skipping short-circuit for {provider_str} "
+                    "(provider has native Anthropic Messages support, using agentic loop)"
+                )
+                return None
+        except (ValueError, Exception):
+            pass  # unknown provider enum → safe to short-circuit
+
         # All tools must be web search tools
         if not all(is_web_search_tool(t) for t in tools):
             return None
diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/handler.py b/litellm/llms/anthropic/experimental_pass_through/messages/handler.py
index dcd9214cf8c..dae2b5da1f8 100644
--- a/litellm/llms/anthropic/experimental_pass_through/messages/handler.py
+++ b/litellm/llms/anthropic/experimental_pass_through/messages/handler.py
@@ -205,11 +205,18 @@ async def anthropic_messages(
     # Extract modified parameters
     tools = request_kwargs.pop("tools", tools)
     stream = request_kwargs.pop("stream", stream)
-    # Propagate the provider derived inside pre-request hooks, if not already set
+    # Propagate the provider derived inside pre-request hooks, if not already set.
+    # The litellm_params dict may have been overwritten by **kwargs in
+    # _execute_pre_request_hooks, so fall back to get_llm_provider() if needed.
     if not custom_llm_provider:
         custom_llm_provider = request_kwargs.get("litellm_params", {}).get(
             "custom_llm_provider"
         )
+        if not custom_llm_provider:
+            try:
+                _, custom_llm_provider, _, _ = litellm.get_llm_provider(model=model)
+            except Exception:
+                pass
     # Remove litellm_params from kwargs (only needed for hooks)
     request_kwargs.pop("litellm_params", None)
     # Merge back any other modifications
diff --git a/tests/test_litellm/integrations/websearch_interception/test_websearch_short_circuit.py b/tests/test_litellm/integrations/websearch_interception/test_websearch_short_circuit.py
index cb90b254e40..82c1c9839e7 100644
--- a/tests/test_litellm/integrations/websearch_interception/test_websearch_short_circuit.py
+++ b/tests/test_litellm/integrations/websearch_interception/test_websearch_short_circuit.py
@@ -114,6 +114,29 @@ class TestTryShortCircuitSearch:
 
         assert result is None
 
+    @pytest.mark.asyncio
+    async def test_does_not_short_circuit_bedrock(self):
+        """Bedrock has native agentic loop support → NOT short-circuited.
+
+        Providers with a BaseAnthropicMessagesConfig (bedrock, vertex_ai, etc.)
+        use the agentic loop which includes a follow-up LLM synthesis step.
+        The short-circuit must not fire for them.
+        """
+        logger = WebSearchInterceptionLogger(
+            enabled_providers=["bedrock", "github_copilot"]
+        )
+
+        result = await logger.try_short_circuit_search(
+            model="bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0",
+            messages=[{"role": "user", "content": "Search for something"}],
+            tools=[
+                {"type": "web_search_20250305", "name": "web_search", "max_uses": 8}
+            ],
+            custom_llm_provider="bedrock",
+        )
+
+        assert result is None
+
     @pytest.mark.asyncio
     async def test_does_not_short_circuit_no_messages(self):
         """Empty messages → NOT short-circuited"""