fix(guardrails): stop the Javelin api_version default leaking into Azure Content Safety

LitellmParams mixes every provider config model into one class, so the
Javelin api_version default of "v1" reached the Azure Content Safety
guardrails whenever config.yaml omitted api_version and Azure answered 404.
The shared field now defaults to None, Javelin keeps filling in "v1" itself,
and the Azure guardrails fall back to the documented 2024-09-01 at request
time so a DB update that omits api_version stays on the default too.

litellm/proxy/_lazy_openapi_snapshot.json

@@ -11155,7 +11155,6 @@
                   "type": "null"
                 }
               ],
-              "default": "v1",
               "description": "API version for Javelin service",
               "title": "Api Version"
             },
@@ -19622,7 +19621,7 @@
                   }
                 }
               },
-              "description": "\n    Unified rate-limit error.\n\n    Every rate-limit condition surfaced by litellm \u2014 whether it originated from\n    an upstream LLM provider, a vendor batch endpoint, or one of litellm's own\n    proxy-side limiters (parallel-requests, dynamic-rate, batch-rate, budget,\n    max-iterations, etc.) \u2014 is raised as an instance of this class.\n\n    The :attr:`category` attribute lets callers distinguish the source. See\n    :class:`RateLimitErrorCategory` for the available values.\n    "
+              "description": "\nUnified rate-limit error.\n\nEvery rate-limit condition surfaced by litellm \u2014 whether it originated from\nan upstream LLM provider, a vendor batch endpoint, or one of litellm's own\nproxy-side limiters (parallel-requests, dynamic-rate, batch-rate, budget,\nmax-iterations, etc.) \u2014 is raised as an instance of this class.\n\nThe :attr:`category` attribute lets callers distinguish the source. See\n:class:`RateLimitErrorCategory` for the available values.\n"
             },
             "500": {
               "content": {

litellm/proxy/guardrails/guardrail_hooks/azure/base.py

@@ -20,6 +20,8 @@ AZURE_CONTENT_SAFETY_MAX_TEXT_LENGTH: Final = 10000
 # chunk of N characters consumes ceil(N / 1000) text records.
 AZURE_CONTENT_SAFETY_TEXT_RECORD_LENGTH: Final = 1000
 
+AZURE_CONTENT_SAFETY_DEFAULT_API_VERSION: Final = "2024-09-01"
+
 
 class AzureGuardrailBase:
     """
@@ -43,7 +45,7 @@ class AzureGuardrailBase:
         self.async_handler = get_async_httpx_client(llm_provider=httpxSpecialProvider.GuardrailCallback)
         self.api_key = api_key
         self.api_base = api_base
-        self.api_version: str = kwargs.get("api_version") or "2024-09-01"
+        self.api_version: str | None = kwargs.get("api_version")
 
     async def _post_to_content_safety(self, endpoint_path: str, request_body: dict[str, object]) -> dict[str, Any]:
         """POST to an Azure Content Safety endpoint with standard auth headers.
@@ -56,7 +58,8 @@ class AzureGuardrailBase:
         Returns:
             Parsed JSON response dict.
         """
-        url: Final = f"{self.api_base}/contentsafety/{endpoint_path}?api-version={self.api_version}"
+        api_version: Final = self.api_version or AZURE_CONTENT_SAFETY_DEFAULT_API_VERSION
+        url: Final = f"{self.api_base}/contentsafety/{endpoint_path}?api-version={api_version}"
         headers: Final = {
             "Ocp-Apim-Subscription-Key": self.api_key,
             "Content-Type": "application/json",

litellm/types/guardrails.py

@@ -819,7 +819,7 @@ class JavelinGuardrailConfigModel(BaseModel):
     """Configuration parameters for the Javelin guardrail"""
 
     guard_name: str | None = Field(default=None, description="Name of the Javelin guard to use")
-    api_version: str | None = Field(default="v1", description="API version for Javelin service")
+    api_version: str | None = Field(default=None, description="API version for Javelin service")
     metadata: dict | None = Field(default=None, description="Additional metadata to send with requests")
     application: str | None = Field(default=None, description="Application name for Javelin service")
     config: dict | None = Field(default=None, description="Additional configuration for the guardrail")

tests/test_litellm/proxy/guardrails/guardrail_hooks/azure/test_azure_prompt_shield.py

@@ -7,6 +7,7 @@ from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.guardrails.guardrail_hooks.azure.prompt_shield import (
     AzureContentSafetyPromptShieldGuardrail,
 )
+from litellm.proxy.guardrails.guardrail_registry import InMemoryGuardrailHandler
 from litellm.types.guardrails import LitellmParams
 
 
@@ -635,3 +636,57 @@ def test_update_in_memory_litellm_params_dead_env_credential_rejected_untouched(
 
     assert guardrail.api_key == "azure_prompt_shield_api_key"
     assert guardrail.price_per_1000_text_records == 0.38
+
+
+@pytest.mark.asyncio
+async def test_config_without_api_version_calls_documented_azure_api_version():
+    """A config.yaml entry that omits api_version must reach Azure at the documented
+    default. LitellmParams inherits every provider's config model, so a sibling
+    provider's api_version default used to leak into the Azure URL and 404."""
+    handler = InMemoryGuardrailHandler()
+    registered = handler.initialize_guardrail(
+        guardrail={
+            "guardrail_name": "azure-prompt-shield-no-api-version",
+            "litellm_params": {
+                "guardrail": "azure/prompt_shield",
+                "mode": "pre_call",
+                "api_key": "azure_prompt_shield_api_key",
+                "api_base": "https://example.cognitiveservices.azure.com",
+            },
+        }
+    )
+    assert registered is not None
+    guardrail = handler.guardrail_id_to_custom_guardrail[registered["guardrail_id"]]
+    assert isinstance(guardrail, AzureContentSafetyPromptShieldGuardrail)
+
+    with patch.object(guardrail.async_handler, "post", return_value=_shield_response(False)) as mock_post:
+        result = await guardrail.apply_guardrail(inputs={"texts": ["hello"]}, request_data={}, input_type="request")
+
+    assert result == {"texts": ["hello"]}
+    assert mock_post.call_args.kwargs["url"] == (
+        "https://example.cognitiveservices.azure.com/contentsafety/text:shieldPrompt?api-version=2024-09-01"
+    )
+
+
+@pytest.mark.asyncio
+async def test_update_without_api_version_keeps_documented_azure_api_version():
+    """The DB update path copies every LitellmParams attribute onto the live
+    instance, api_version included, so an update that omits it must still leave
+    the request on the documented default rather than a None or leaked value."""
+    guardrail = _shield_guardrail()
+    guardrail.update_in_memory_litellm_params(
+        LitellmParams(
+            guardrail="azure/prompt_shield",
+            mode="pre_call",
+            api_key="azure_prompt_shield_api_key",
+            api_base="https://example.cognitiveservices.azure.com",
+        )
+    )
+
+    with patch.object(guardrail.async_handler, "post", return_value=_shield_response(False)) as mock_post:
+        result = await guardrail.apply_guardrail(inputs={"texts": ["hello"]}, request_data={}, input_type="request")
+
+    assert result == {"texts": ["hello"]}
+    assert mock_post.call_args.kwargs["url"] == (
+        "https://example.cognitiveservices.azure.com/contentsafety/text:shieldPrompt?api-version=2024-09-01"
+    )

tests/test_litellm/proxy/guardrails/guardrail_hooks/azure/test_azure_text_moderation.py

@@ -4,6 +4,7 @@ import pytest
 from fastapi import HTTPException
 
 from litellm.proxy._types import UserAPIKeyAuth
+from litellm.proxy.guardrails.guardrail_registry import InMemoryGuardrailHandler
 from litellm.proxy.guardrails.guardrail_hooks.azure.text_moderation import (
     AzureContentSafetyTextModerationGuardrail,
 )
@@ -463,3 +464,33 @@ async def test_apply_guardrail_handles_missing_texts_key():
 
     mock_post.assert_not_called()
     assert result == {"images": ["x"]}
+
+
+@pytest.mark.asyncio
+async def test_config_without_api_version_calls_documented_azure_api_version():
+    """A config.yaml entry that omits api_version must reach Azure at the documented
+    default. LitellmParams inherits every provider's config model, so a sibling
+    provider's api_version default used to leak into the Azure URL and 404."""
+    handler = InMemoryGuardrailHandler()
+    registered = handler.initialize_guardrail(
+        guardrail={
+            "guardrail_name": "azure-text-moderation-no-api-version",
+            "litellm_params": {
+                "guardrail": "azure/text_moderations",
+                "mode": "pre_call",
+                "api_key": "azure_text_moderation_api_key",
+                "api_base": "https://example.cognitiveservices.azure.com",
+            },
+        }
+    )
+    assert registered is not None
+    guardrail = handler.guardrail_id_to_custom_guardrail[registered["guardrail_id"]]
+    assert isinstance(guardrail, AzureContentSafetyTextModerationGuardrail)
+
+    with patch.object(guardrail.async_handler, "post", return_value=_moderation_response(0)) as mock_post:
+        result = await guardrail.apply_guardrail(inputs={"texts": ["hello"]}, request_data={}, input_type="request")
+
+    assert result == {"texts": ["hello"]}
+    assert mock_post.call_args.kwargs["url"] == (
+        "https://example.cognitiveservices.azure.com/contentsafety/text:analyze?api-version=2024-09-01"
+    )

tests/test_litellm/proxy/guardrails/guardrail_hooks/test_javelin.py

@@ -0,0 +1,42 @@
+from unittest.mock import Mock, patch
+
+import pytest
+
+from litellm.proxy.guardrails.guardrail_hooks.javelin.javelin import JavelinGuardrail
+from litellm.proxy.guardrails.guardrail_registry import InMemoryGuardrailHandler
+from litellm.types.guardrails import GuardrailEventHooks
+
+
+@pytest.mark.asyncio
+async def test_config_without_api_version_calls_javelin_v1():
+    """Javelin's v1 default no longer lives in the shared LitellmParams model (it
+    leaked into every other provider), so the Javelin initializer has to supply
+    it itself when the config omits api_version."""
+    handler = InMemoryGuardrailHandler()
+    registered = handler.initialize_guardrail(
+        guardrail={
+            "guardrail_name": "javelin-no-api-version",
+            "litellm_params": {
+                "guardrail": "javelin",
+                "mode": "pre_call",
+                "api_key": "javelin_api_key",
+                "api_base": "https://javelin.example",
+                "guard_name": "trustsafety",
+            },
+        }
+    )
+    assert registered is not None
+    guardrail = handler.guardrail_id_to_custom_guardrail[registered["guardrail_id"]]
+    assert isinstance(guardrail, JavelinGuardrail)
+    assessments = [{"trustsafety": {"request_reject": False}}]
+    response = Mock()
+    response.json.return_value = {"assessments": assessments}
+
+    with patch.object(guardrail.async_handler, "post", return_value=response) as mock_post:
+        result = await guardrail.call_javelin_guard(
+            request={"input": {"text": "hello"}, "config": None, "metadata": None},
+            event_type=GuardrailEventHooks.pre_call,
+        )
+
+    assert result == {"assessments": assessments}
+    assert mock_post.call_args.kwargs["url"] == "https://javelin.example/v1/guardrail/trustsafety/apply"

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -31795,9 +31795,8 @@ export interface components {
             /**
              * Api Version
              * @description API version for Javelin service
-             * @default v1
              */
-            api_version: string | null;
+            api_version?: string | null;
             /**
              * Application
              * @description Application name for Javelin service