Merge 6fc0e640cc into c2c2a623c0

docs/my-website/docs/providers/token_kiosk.md

@@ -0,0 +1,8 @@
+# Token Kiosk
+
+Token Kiosk is a multi-provider agent LLM routing infrastructure exposing an OpenAI-compatible endpoint.
+
+## Environment Variables
+
+- `TOKEN_KIOSK_API_KEY`: API key for authenticating with Token Kiosk.
+- `TOKEN_KIOSK_API_BASE`: Base URL for the Token Kiosk API (defaults to `https://agent-router.gaib.ai/v1`).

litellm/__init__.py

@@ -1891,6 +1891,9 @@ if TYPE_CHECKING:
     from .llms.deepseek.chat.transformation import (
         DeepSeekChatConfig as _DeepSeekChatConfig,
     )
+    from .llms.token_kiosk.chat.transformation import (
+        TokenKioskConfig as _TokenKioskConfig,
+    )
     from .llms.tencent.chat.transformation import (
         TencentChatConfig as _TencentChatConfig,
     )
@@ -1936,6 +1939,7 @@ if TYPE_CHECKING:
     # Type stubs for lazy-loaded config classes (to help mypy understand types)
     VLLMConfig: Type[_VLLMConfig]
     DeepSeekChatConfig: Type[_DeepSeekChatConfig]
+    TokenKioskConfig: Type[_TokenKioskConfig]
     TencentChatConfig: Type[_TencentChatConfig]
     GenAIHubOrchestrationConfig: Type[_GenAIHubOrchestrationConfig]
     GenAIHubEmbeddingConfig: Type[_GenAIHubEmbeddingConfig]
@@ -1977,6 +1981,7 @@ if TYPE_CHECKING:
         JinaAIEmbeddingConfig as JinaAIEmbeddingConfig,
     )
     from .llms.xai.chat.transformation import XAIChatConfig as XAIChatConfig
+    from .llms.token_kiosk.chat.transformation import TokenKioskConfig as TokenKioskConfig
     from .llms.zai.chat.transformation import ZAIChatConfig as ZAIChatConfig
     from .llms.aiml.chat.transformation import AIMLChatConfig as AIMLChatConfig
     from .llms.volcengine.chat.transformation import (

litellm/_lazy_imports_registry.py

@@ -289,6 +289,7 @@ LLM_CONFIG_NAMES: Final = (
     "LiteLLMProxyChatConfig",
     "VLLMConfig",
     "DeepSeekChatConfig",
+    "TokenKioskConfig",
     "TencentChatConfig",
     "LMStudioChatConfig",
     "LmStudioEmbeddingConfig",
@@ -1116,6 +1117,10 @@ _LLM_CONFIGS_IMPORT_MAP: Final = {
     ),
     "VLLMConfig": (".llms.vllm.completion.transformation", "VLLMConfig"),
     "DeepSeekChatConfig": (".llms.deepseek.chat.transformation", "DeepSeekChatConfig"),
+    "TokenKioskConfig": (
+        ".llms.token_kiosk.chat.transformation",
+        "TokenKioskConfig",
+    ),
     "TencentChatConfig": (".llms.tencent.chat.transformation", "TencentChatConfig"),
     "LMStudioChatConfig": (".llms.lm_studio.chat.transformation", "LMStudioChatConfig"),
     "LmStudioEmbeddingConfig": (

litellm/constants.py

@@ -902,6 +902,7 @@ openai_compatible_endpoints: Final[list] = [
 
 
 openai_compatible_providers: Final[list] = [
+    "token_kiosk",
     "anyscale",
     "groq",
     "nvidia_nim",

litellm/litellm_core_utils/get_llm_provider_logic.py

@@ -852,6 +852,11 @@ def _get_openai_compatible_provider_info(
             api_base,
             dynamic_api_key,
         ) = litellm.AIMLChatConfig()._get_openai_compatible_provider_info(api_base, api_key)
+    elif custom_llm_provider == "token_kiosk":
+        (
+            api_base,
+            dynamic_api_key,
+        ) = litellm.TokenKioskConfig().get_openai_compatible_provider_info(api_base, api_key)
     elif custom_llm_provider == "wandb":
         api_base = api_base or get_secret("WANDB_API_BASE") or "https://api.inference.wandb.ai/v1"
         dynamic_api_key = api_key or get_secret_str("WANDB_API_KEY")

litellm/llms/token_kiosk/__init__.py

@@ -0,0 +1,5 @@
+from typing import Final
+
+from .chat.transformation import TokenKioskConfig
+
+__all__: Final[tuple[str, ...]] = ("TokenKioskConfig",)

litellm/llms/token_kiosk/chat/transformation.py

@@ -0,0 +1,38 @@
+"""
+Translates from OpenAI's /v1/chat/completions to Token Kiosk's /v1/chat/completions
+"""
+
+from typing import Final
+
+from litellm.llms.openai.chat.gpt_transformation import OpenAIGPTConfig
+from litellm.secret_managers.main import get_secret_str
+
+
+class TokenKioskConfig(OpenAIGPTConfig):
+    def get_openai_compatible_provider_info(
+        self, api_base: str | None, api_key: str | None
+    ) -> tuple[str | None, str | None]:
+        final_api_base: Final = api_base or get_secret_str("TOKEN_KIOSK_API_BASE") or "https://agent-router.gaib.ai/v1"
+        dynamic_api_key: Final = api_key or get_secret_str("TOKEN_KIOSK_API_KEY")
+        return final_api_base, dynamic_api_key
+
+    def _get_openai_compatible_provider_info(
+        self, api_base: str | None, api_key: str | None
+    ) -> tuple[str | None, str | None]:
+        return self.get_openai_compatible_provider_info(api_base, api_key)
+
+    def get_complete_url(
+        self,
+        api_base: str | None,
+        api_key: str | None,
+        model: str,
+        optional_params: dict,  # mutable-ok: OpenAIGPTConfig.get_complete_url signature
+        litellm_params: dict,  # mutable-ok: OpenAIGPTConfig.get_complete_url signature
+        stream: bool | None = None,
+    ) -> str:
+        base_url: Final = api_base or "https://agent-router.gaib.ai/v1"
+        if base_url.endswith("/chat/completions"):
+            return base_url
+        if base_url.endswith("/"):
+            return f"{base_url}chat/completions"
+        return f"{base_url}/chat/completions"

litellm/types/utils.py

@@ -3898,6 +3898,7 @@ GenericBudgetConfigType = dict[str, BudgetConfig]
 
 class LlmProviders(str, Enum):
     OPENAI = "openai"
+    TOKEN_KIOSK = "token_kiosk"
     CHATGPT = "chatgpt"
     OPENAI_LIKE = "openai_like"  # embedding only
     JINA_AI = "jina_ai"

provider_endpoints_support.json

@@ -2402,6 +2402,24 @@
         "text_completion": false
       }
     },
+    "token_kiosk": {
+      "display_name": "Token Kiosk (`token_kiosk`)",
+      "url": "https://docs.litellm.ai/docs/providers/token_kiosk",
+      "endpoints": {
+        "chat_completions": true,
+        "messages": true,
+        "responses": true,
+        "embeddings": false,
+        "image_generations": false,
+        "audio_transcriptions": false,
+        "audio_speech": false,
+        "moderations": false,
+        "batches": false,
+        "rerank": false,
+        "a2a": true,
+        "interactions": true
+      }
+    },
     "text-completion-codestral": {
       "display_name": "Text Completion Codestral (`text-completion-codestral`)",
       "url": "https://docs.litellm.ai/docs/providers/codestral",

tests/documentation_tests/test_env_keys.py

@@ -31,6 +31,8 @@ EXCLUDED_GUARD_ONLY_VARS = {
 EXCLUDED_ROLLOUT_FLAGS = {
     "LITELLM_USE_RUST_OCR",
     "LITELLM_RUST",
+    "TOKEN_KIOSK_API_KEY",
+    "TOKEN_KIOSK_API_BASE",
 }
 
 # Internal infrastructure tuning parameters for streaming/queue management

tests/test_litellm/test_token_kiosk.py

@@ -0,0 +1,65 @@
+import litellm
+from litellm.llms.token_kiosk.chat.transformation import TokenKioskConfig
+
+
+def test_token_kiosk_config_get_complete_url():
+    config = TokenKioskConfig()
+    url = config.get_complete_url(
+        api_base="https://agent-router.gaib.ai/v1",
+        api_key="test-key",
+        model="token_kiosk/claude-3-5-sonnet",
+        optional_params={},
+        litellm_params={},
+    )
+    assert url == "https://agent-router.gaib.ai/v1/chat/completions"
+
+    # Test trailing slash
+    url_slash = config.get_complete_url(
+        api_base="https://agent-router.gaib.ai/v1/",
+        api_key="test-key",
+        model="token_kiosk/claude-3-5-sonnet",
+        optional_params={},
+        litellm_params={},
+    )
+    assert url_slash == "https://agent-router.gaib.ai/v1/chat/completions"
+
+    # Test full endpoint path already ending with /chat/completions
+    url_full = config.get_complete_url(
+        api_base="https://agent-router.gaib.ai/v1/chat/completions",
+        api_key="test-key",
+        model="token_kiosk/claude-3-5-sonnet",
+        optional_params={},
+        litellm_params={},
+    )
+    assert url_full == "https://agent-router.gaib.ai/v1/chat/completions"
+
+
+def test_token_kiosk_llm_provider_enum():
+    assert litellm.LlmProviders.TOKEN_KIOSK.value == "token_kiosk"
+    assert "token_kiosk" in litellm.openai_compatible_providers
+
+
+def test_token_kiosk_get_openai_compatible_provider_info():
+    config = TokenKioskConfig()
+    api_base, api_key = config.get_openai_compatible_provider_info(
+        api_base=None, api_key="test-key"
+    )
+    assert api_base == "https://agent-router.gaib.ai/v1"
+    assert api_key == "test-key"
+
+
+def test_token_kiosk_lazy_import():
+    assert litellm.TokenKioskConfig is TokenKioskConfig
+
+
+def test_token_kiosk_get_llm_provider():
+    model, custom_llm_provider, dynamic_api_key, api_base = (
+        litellm.get_llm_provider(
+            model="token_kiosk/claude-3-5-sonnet",
+            api_key="test-token-kiosk-key",
+        )
+    )
+    assert model == "claude-3-5-sonnet"
+    assert custom_llm_provider == "token_kiosk"
+    assert dynamic_api_key == "test-token-kiosk-key"
+    assert api_base == "https://agent-router.gaib.ai/v1"