From 6544beae3408a1bd67229b4ba34bde369b1ca91b Mon Sep 17 00:00:00 2001
From: Vigilans <vigilans@foxmail.com>
Date: Sun, 17 May 2026 12:16:58 +0800
Subject: [PATCH] feat(proxy): add global websearch_fallback_model setting
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

When a web-search-only request targets a model that does not support
web search, redirect it to the model group specified in
litellm_settings.websearch_fallback_model. Uses
litellm.supports_web_search() for capability detection, which
correctly strips provider prefixes (e.g. github_copilot/gpt-5.5
→ gpt-5.5) when looking up the model registry.

Also add github_copilot to _check_provider_match exemption list,
aligning it with the existing github exemption.
---
 litellm/proxy/common_request_processing.py | 15 ++++++++++++---
 litellm/utils.py                           |  2 +-
 2 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py
index 9d69f4c97ba..0b985c6b785 100644
--- a/litellm/proxy/common_request_processing.py
+++ b/litellm/proxy/common_request_processing.py
@@ -923,9 +923,11 @@ 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.
+        ### WEB SEARCH REDIRECT ###
+        # if request only contains web_search tools, check for redirect:
+        # 1. per-deployment force: always redirect to `force_websearch_model`
+        # 2. global fallback: redirect to `websearch_fallback_model` only when
+        #    the current model group does not support web search
         if (
             isinstance(self.data.get("model"), str)
             and llm_router is not None
@@ -939,6 +941,13 @@ class ProxyBaseLLMRequestProcessing:
                 if model := dep.get("litellm_params", {}).get("force_websearch_model"):
                     self.data["model"] = model
                     break
+                if (
+                    (dep_model := dep.get("litellm_params", {}).get("model", ""))
+                    and not litellm.supports_web_search(dep_model)
+                    and (model := getattr(litellm, "websearch_fallback_model", None))
+                ):
+                    self.data["model"] = model
+                    break
 
         self.data["litellm_call_id"] = request.headers.get(
             "x-litellm-call-id", str(uuid.uuid4())
diff --git a/litellm/utils.py b/litellm/utils.py
index 2ba6ef9cae8..9121473918c 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -5579,7 +5579,7 @@ def _check_provider_match(model_info: dict, custom_llm_provider: Optional[str])
             # as a last attempt if the model is not on Azure AI, Azure then fallback to OpenAI cost
             # tracking the cost is better than attributing 0 cost to it.
             return True
-        elif custom_llm_provider == "github":
+        elif custom_llm_provider in ("github", "github_copilot"):
             # Allow github/<model> aliases to reuse existing provider metadata.
             return True
         else: