Merge b7b401eda5 into f4a2e2b3af

litellm/llms/openai_like/dynamic_config.py

@@ -79,6 +79,17 @@ def create_config_class(provider: SimpleProviderConfig):
             # Resolve API key
             resolved_key: Final = api_key or get_secret_str(provider.api_key_env)
 
+            if provider.special_handling.get("require_api_key") and not resolved_key:
+                # Fail closed: without this the caller's OPENAI_API_KEY would be used as a fallback
+                # further down the dispatch path and sent to this provider.
+                from litellm.exceptions import AuthenticationError
+
+                raise AuthenticationError(
+                    message=f"{provider.slug}: no API key. Pass api_key or set {provider.api_key_env}.",
+                    llm_provider=provider.slug,
+                    model="",
+                )
+
             return resolved_base, resolved_key
 
         def get_openai_compatible_provider_info(

litellm/llms/openai_like/providers.json

@@ -302,5 +302,15 @@
       "preserve_cache_control": true
     },
     "supported_endpoints": ["/v1/chat/completions"]
+  },
+  "sagg": {
+    "base_url": "https://api.privatedeskai.com/v1",
+    "api_key_env": "SAGG_API_KEY",
+    "param_mappings": {
+      "max_completion_tokens": "max_tokens"
+    },
+    "special_handling": {
+      "require_api_key": true
+    }
   }
 }

litellm/types/utils.py

@@ -4709,6 +4709,7 @@ class LlmProviders(str, Enum):
     HOSTINGER_ROUTER = "hostinger_router"
     ZEROGPU = "zerogpu"
     DARKBLOOM = "darkbloom"
+    SAGG = "sagg"
     META = "meta"
     SAIL = "sail"
     REKA = "reka"

provider_endpoints_support.json

@@ -3578,6 +3578,22 @@
         "generateContent": false
       }
     },
+    "sagg": {
+      "display_name": "SAGG (`sagg`)",
+      "url": "https://docs.litellm.ai/docs/providers/sagg",
+      "endpoints": {
+        "chat_completions": true,
+        "messages": false,
+        "responses": false,
+        "embeddings": false,
+        "image_generations": false,
+        "audio_transcriptions": false,
+        "audio_speech": false,
+        "moderations": false,
+        "batches": false,
+        "rerank": false
+      }
+    },
     "searxng": {
       "display_name": "SearXNG (`searxng`)",
       "url": "https://docs.litellm.ai/docs/search/searxng",

tests/unit/llms/openai_like/test_sagg_provider.py

@@ -0,0 +1,108 @@
+"""
+Tests for SAGG provider configuration and integration.
+
+SAGG (https://api.privatedeskai.com) is an OpenAI-compatible LLM
+inference gateway with automatic multi-provider failover.
+"""
+
+import litellm
+
+
+class TestSaggProviderConfig:
+    def test_sagg_in_provider_list(self):
+        from litellm import LlmProviders
+
+        assert hasattr(LlmProviders, "SAGG")
+        assert LlmProviders.SAGG.value == "sagg"
+        assert "sagg" in litellm.provider_list
+
+    def test_sagg_json_config_exists(self):
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+
+        assert JSONProviderRegistry.exists("sagg")
+
+        sagg = JSONProviderRegistry.get("sagg")
+        assert sagg is not None
+        assert sagg.base_url == "https://api.privatedeskai.com/v1"
+        assert sagg.api_key_env == "SAGG_API_KEY"
+        assert sagg.param_mappings.get("max_completion_tokens") == "max_tokens"
+
+    def test_sagg_provider_resolution(self):
+        """The real model id contains its own slash
+        (deepseek-ai/DeepSeek-V4-Flash-0731), so this also confirms
+        provider-prefix splitting only splits on the FIRST slash, same
+        as e.g. pinstripes/ps/glm-4.5-air elsewhere in this registry."""
+        from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+        model, provider, api_key, api_base = get_llm_provider(
+            model="sagg/deepseek-ai/DeepSeek-V4-Flash-0731",
+            custom_llm_provider=None,
+            api_base=None,
+            api_key="test-key",
+        )
+
+        assert model == "deepseek-ai/DeepSeek-V4-Flash-0731"
+        assert provider == "sagg"
+        assert api_base == "https://api.privatedeskai.com/v1"
+
+    def test_sagg_router_config(self):
+        from litellm import Router
+
+        router = Router(
+            model_list=[
+                {
+                    "model_name": "sagg-deepseek",
+                    "litellm_params": {
+                        "model": "sagg/deepseek-ai/DeepSeek-V4-Flash-0731",
+                        "api_key": "test-key",
+                    },
+                }
+            ]
+        )
+
+        assert len(router.model_list) == 1
+        assert router.model_list[0]["model_name"] == "sagg-deepseek"
+
+    def test_sagg_parameter_mapping(self):
+        """SAGG's own request parser only recognizes max_tokens, so a
+        caller sending max_completion_tokens needs this mapping to
+        actually take effect server-side rather than being dropped."""
+        from litellm.llms.openai_like.dynamic_config import create_config_class
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+
+        provider = JSONProviderRegistry.get("sagg")
+        config_class = create_config_class(provider)
+        config = config_class()
+
+        optional_params = {}
+        non_default_params = {"max_completion_tokens": 100, "temperature": 0.7}
+        result = config.map_openai_params(
+            non_default_params, optional_params, "deepseek-ai/DeepSeek-V4-Flash-0731", False
+        )
+
+        assert "max_tokens" in result
+        assert result["max_tokens"] == 100
+        assert "max_completion_tokens" not in result
+        assert result["temperature"] == 0.7
+
+
+class TestSaggRequiresItsOwnKey:
+    def _config(self):
+        from litellm.llms.openai_like.dynamic_config import create_config_class
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+
+        return create_config_class(JSONProviderRegistry.get("sagg"))()
+
+    def test_missing_key_fails_closed_even_with_openai_key_set(self, monkeypatch):
+        import pytest
+
+        monkeypatch.delenv("SAGG_API_KEY", raising=False)
+        monkeypatch.setenv("OPENAI_API_KEY", "sk-not-for-sagg")
+        with pytest.raises(litellm.AuthenticationError):
+            self._config()._get_openai_compatible_provider_info(None, None)
+
+    def test_explicit_key_and_env_key_still_work(self, monkeypatch):
+        monkeypatch.delenv("SAGG_API_KEY", raising=False)
+        assert self._config()._get_openai_compatible_provider_info(None, "k1")[1] == "k1"
+        monkeypatch.setenv("SAGG_API_KEY", "k2")
+        assert self._config()._get_openai_compatible_provider_info(None, None)[1] == "k2"