Merge remote-tracking branch 'origin/litellm_internal_staging' into litellm_daily_any_cleanup_08_07_2026_3Mn0k

.github/workflows/test-unit-proxy-endpoints.yml

@@ -44,6 +44,7 @@ jobs:
         tests/test_litellm/proxy/vector_store_endpoints
         tests/test_litellm/proxy/agent_endpoints
         tests/test_litellm/proxy/a2a
+        tests/test_litellm/proxy/credential_endpoints
         tests/test_litellm/proxy/discovery_endpoints
         tests/test_litellm/proxy/health_endpoints
         tests/test_litellm/proxy/shutdown

litellm/integrations/azure_sentinel/azure_sentinel.py

@@ -78,8 +78,8 @@ class AzureSentinelLogger(CustomBatchLogger):
                 If not provided, will use AZURE_SENTINEL_AUDIT_STREAM_NAME env var or the standard stream name.
             authority_host (str, optional): Microsoft Entra authority host that issues the OAuth2 token,
                 e.g. "https://login.microsoftonline.us" for Azure Government. If not provided, will use
-                AZURE_AUTHORITY_HOST env var or default to the Azure Public Cloud authority. The Azure
-                Monitor audience is derived from it.
+                AZURE_SENTINEL_AUTHORITY_HOST or AZURE_AUTHORITY_HOST env vars, or default to the Azure
+                Public Cloud authority. The Azure Monitor audience is derived from it.
         """
         self.async_httpx_client = get_async_httpx_client(llm_provider=httpxSpecialProvider.LoggingCallback)
 
@@ -95,7 +95,10 @@ class AzureSentinelLogger(CustomBatchLogger):
             client_secret or os.getenv("AZURE_SENTINEL_CLIENT_SECRET") or os.getenv("AZURE_CLIENT_SECRET")
         )
         resolved_authority_host: Final = self._normalize_authority_host(
-            authority_host or os.getenv("AZURE_AUTHORITY_HOST") or DEFAULT_AZURE_AUTHORITY_HOST
+            authority_host
+            or os.getenv("AZURE_SENTINEL_AUTHORITY_HOST")
+            or os.getenv("AZURE_AUTHORITY_HOST")
+            or DEFAULT_AZURE_AUTHORITY_HOST
         )
 
         if not resolved_dcr_immutable_id:

litellm/proxy/management_endpoints/key_management_endpoints.py

@@ -3528,11 +3528,34 @@ async def info_key_fn(
 ):
     """
     Retrieve information about a key.
+
     Parameters:
-        key: Optional[str] = Query parameter representing the key in the request
-        user_api_key_dict: UserAPIKeyAuth = Dependency representing the user's API key
+    - key: str | None (query parameter) - The key to look up. Accepts the plaintext key or its hash.
+      Defaults to the key in the Authorization header.
+
     Returns:
-        Dict containing the key and its associated information
+    - key: str - The key that was looked up, echoed back as it was passed in
+    - info: dict - The key's row, minus the hashed token
+        - key_alias: str | None - User-friendly key alias
+        - spend: float - Amount spent by the key. When budget_duration is set this covers only the
+          current budget window, not the key's lifetime
+        - max_budget: float | None - Max budget for the key, enforced against spend
+        - budget_duration: str | None - Budget reset period ("30d", "1h", etc.)
+        - budget_reset_at: datetime | None - When the current budget window ends and spend is next
+          reset to 0, not when it was last reset. Reset times snap to standard boundaries in the
+          configured timezone (30d and 1mo land on the 1st of the month, 7d on Monday, 1h on the
+          hour), so subtracting budget_duration from it does not give the window's start
+        - model_max_budget: dict - Per-model budgets, e.g. {"gpt-4": {"budget_limit": 0.0005, "time_period": "30d"}}
+        - model_max_budget_usage: dict | None - Current-window spend per model, present only when
+          the key has per-model budgets
+        - models: list - Model_name's the key is allowed to call
+        - tpm_limit / rpm_limit: int | None - Tokens and requests per minute limits
+        - metadata: dict - Metadata for the key, e.g. {"team": "core-infra"}
+        - blocked: bool | None - Whether the key is blocked
+        - expires: datetime | None - When the key stops authenticating requests
+        - last_active: datetime | None - When the key was last used
+        - object_permission: dict | None - Resolved vector store / MCP permissions when the key has
+          an object_permission_id
 
     Example Curl:
     ```

tests/test_litellm/integrations/test_azure_sentinel.py

@@ -313,6 +313,7 @@ def _build_logger(**overrides):
 
 @pytest.fixture
 def _no_authority_host_env(monkeypatch):
+    monkeypatch.delenv("AZURE_SENTINEL_AUTHORITY_HOST", raising=False)
     monkeypatch.delenv("AZURE_AUTHORITY_HOST", raising=False)
 
 
@@ -389,3 +390,38 @@ async def test_azure_sentinel_token_request_uses_sovereign_authority_and_audienc
     assert len(token_calls) == 1
     assert token_calls[0].kwargs["url"] == "https://login.microsoftonline.us/test-tenant-id/oauth2/v2.0/token"
     assert token_calls[0].kwargs["data"]["scope"] == "https://monitor.azure.us/.default"
+
+
+def test_azure_sentinel_authority_host_prefers_the_sentinel_scoped_env_var(_no_authority_host_env, monkeypatch):
+    """AZURE_AUTHORITY_HOST is shared with Azure OpenAI and the azure_storage callback, so a deployment
+    whose Sentinel workspace lives in a different cloud than the rest of its Azure resources needs a
+    Sentinel-scoped override. This mirrors how tenant, client id and secret already resolve."""
+    monkeypatch.setenv("AZURE_AUTHORITY_HOST", "https://login.microsoftonline.com")
+    monkeypatch.setenv("AZURE_SENTINEL_AUTHORITY_HOST", "https://login.microsoftonline.us")
+
+    logger = _build_logger()
+
+    assert logger.authority_host == "https://login.microsoftonline.us"
+    assert logger.oauth_scope == "https://monitor.azure.us/.default"
+
+
+def test_azure_sentinel_falls_back_to_the_shared_authority_host(_no_authority_host_env, monkeypatch):
+    """With no Sentinel-scoped override the shared variable still applies, which is the behavior
+    shipped in the original fix."""
+    monkeypatch.setenv("AZURE_AUTHORITY_HOST", "https://login.microsoftonline.us")
+
+    logger = _build_logger()
+
+    assert logger.authority_host == "https://login.microsoftonline.us"
+    assert logger.oauth_scope == "https://monitor.azure.us/.default"
+
+
+def test_azure_sentinel_authority_host_argument_outranks_the_scoped_env_var(_no_authority_host_env, monkeypatch):
+    """An explicit constructor argument is the most specific source and has to win, otherwise a
+    deployment that exports the scoped variable silently overrides an SDK caller."""
+    monkeypatch.setenv("AZURE_SENTINEL_AUTHORITY_HOST", "https://login.microsoftonline.us")
+
+    logger = _build_logger(authority_host="https://login.microsoftonline.com")
+
+    assert logger.authority_host == "https://login.microsoftonline.com"
+    assert logger.oauth_scope == "https://monitor.azure.com/.default"