diff --git a/litellm/llms/zai/messages/__init__.py b/litellm/llms/zai/messages/__init__.py new file mode 100644 index 00000000000..8b137891791 --- /dev/null +++ b/litellm/llms/zai/messages/__init__.py @@ -0,0 +1 @@ + diff --git a/litellm/llms/zai/messages/transformation.py b/litellm/llms/zai/messages/transformation.py new file mode 100644 index 00000000000..91ae5ff70cc --- /dev/null +++ b/litellm/llms/zai/messages/transformation.py @@ -0,0 +1,72 @@ +""" +Z.AI Anthropic-compatible messages transformation config. +""" + +from collections.abc import Mapping +from typing import Final + +import litellm +from litellm.llms.anthropic.experimental_pass_through.messages.transformation import ( + AnthropicMessagesConfig, +) +from litellm.secret_managers.main import get_secret_str + + +class ZAIAnthropicMessagesConfig(AnthropicMessagesConfig): + """ + Z.AI exposes an Anthropic-compatible Messages API at + https://api.z.ai/api/anthropic (see + https://docs.z.ai/guides/llm/glm-5.3). + + The endpoint accepts the native Anthropic Messages conversation shape + and authenticates with the Z.AI API key sent as the Anthropic + ``x-api-key`` header. + """ + + @property + def custom_llm_provider(self) -> str | None: + return "zai" + + @staticmethod + def get_api_key(api_key: str | None = None) -> str | None: + return api_key or get_secret_str("ZAI_API_KEY") or litellm.api_key + + @staticmethod + def get_api_base(api_base: str | None = None) -> str: + return api_base or "https://api.z.ai/api/anthropic" + + def validate_anthropic_messages_environment( + self, + headers: Mapping[str, str], + model: str, + messages: list[Mapping[str, object]], # mutable-ok: matches the pass-through handler's message list contract + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + api_key: str | None = None, + api_base: str | None = None, + ) -> tuple[dict[str, str], str | None]: # mutable-ok: the handler owns and mutates the returned headers dict + return super().validate_anthropic_messages_environment( + headers=headers, + model=model, + messages=messages, + optional_params=optional_params, + litellm_params=litellm_params, + api_key=self.get_api_key(api_key=api_key), + api_base=api_base, + ) + + def get_complete_url( + self, + api_base: str | None, + api_key: str | None, + model: str, + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + stream: bool | None = None, + ) -> str: + raw_base_url: Final = self.get_api_base(api_base=api_base).rstrip("/") + root_url: Final = raw_base_url.removesuffix("/v1/messages").removesuffix("/v1").removesuffix("/beta") + + if root_url.endswith("/anthropic") or "/anthropic/" in root_url: + return f"{root_url}/v1/messages" + return f"{root_url}/anthropic/v1/messages" diff --git a/litellm/llms/zai/responses/__init__.py b/litellm/llms/zai/responses/__init__.py new file mode 100644 index 00000000000..8b137891791 --- /dev/null +++ b/litellm/llms/zai/responses/__init__.py @@ -0,0 +1 @@ + diff --git a/litellm/llms/zai/responses/transformation.py b/litellm/llms/zai/responses/transformation.py new file mode 100644 index 00000000000..cfcedc2b868 --- /dev/null +++ b/litellm/llms/zai/responses/transformation.py @@ -0,0 +1,64 @@ +""" +Z.AI OpenAI-compatible Responses API transformation config. +""" + +from collections.abc import Mapping +from typing import Final + +import litellm +from litellm.llms.openai.responses.transformation import OpenAIResponsesAPIConfig +from litellm.secret_managers.main import get_secret_str +from litellm.types.router import GenericLiteLLMParams +from litellm.types.utils import LlmProviders + + +class ZAIResponsesAPIConfig(OpenAIResponsesAPIConfig): + """ + Z.AI exposes an OpenAI-compatible Responses API at + https://api.z.ai/api/v1 (see https://docs.z.ai/guides/llm/glm-5.3). + + The endpoint authenticates with the Z.AI API key sent as an + ``Authorization: Bearer`` header. + """ + + _ZAI_CHAT_API_BASE_SUFFIXES: Final = ("/api/paas/v4", "/api/coding/paas/v4") + + @property + def custom_llm_provider(self) -> LlmProviders: + return LlmProviders.ZAI + + @staticmethod + def get_api_key(api_key: str | None = None) -> str | None: + return api_key or get_secret_str("ZAI_API_KEY") or litellm.api_key + + def validate_environment( + self, + headers: Mapping[str, str], + model: str, + litellm_params: GenericLiteLLMParams | None, + ) -> dict[str, str]: # mutable-ok: the responses handler owns and mutates the returned headers dict + request_api_key: Final = litellm_params.api_key if litellm_params is not None else None + resolved_params: Final = GenericLiteLLMParams(api_key=self.get_api_key(api_key=request_api_key)) + + return super().validate_environment( + headers=headers, + model=model, + litellm_params=resolved_params, + ) + + def get_complete_url( + self, + api_base: str | None, + litellm_params: Mapping[str, object], + ) -> str: + # ``litellm_params.api_base`` can carry the Z.AI chat-completions base + # (``/api/paas/v4``) when the generic provider resolver pre-fills it from + # the chat config. Z.AI serves Responses on a different base, so ignore + # the chat-only bases and use the Responses base instead. + normalized_api_base: Final = (api_base or "").rstrip("/") + chat_base_passed_in: Final = normalized_api_base.endswith(self._ZAI_CHAT_API_BASE_SUFFIXES) + base_url: Final = normalized_api_base if api_base and not chat_base_passed_in else "https://api.z.ai/api/v1" + + if base_url.endswith("/responses"): + return base_url + return f"{base_url}/responses" diff --git a/litellm/utils.py b/litellm/utils.py index dd35c17809f..cc6715e5023 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -8851,6 +8851,12 @@ class ProviderConfigManager: ) return TencentAnthropicMessagesConfig() + elif litellm.LlmProviders.ZAI == provider: + from litellm.llms.zai.messages.transformation import ( + ZAIAnthropicMessagesConfig, + ) + + return ZAIAnthropicMessagesConfig() elif litellm.LlmProviders.GITHUB_COPILOT == provider: if "claude" in model_lower: from litellm.llms.github_copilot.messages.transformation import ( @@ -9094,6 +9100,12 @@ class ProviderConfigManager: return litellm.BedrockMantleResponsesAPIConfig( use_openai_path=mantle_base_segment(model, litellm.model_cost) == "openai/v1" ) + elif litellm.LlmProviders.ZAI == provider: + from litellm.llms.zai.responses.transformation import ( + ZAIResponsesAPIConfig, + ) + + return ZAIResponsesAPIConfig() return None @staticmethod diff --git a/tests/test_litellm/llms/zai/test_zai_anthropic_messages_transformation.py b/tests/test_litellm/llms/zai/test_zai_anthropic_messages_transformation.py new file mode 100644 index 00000000000..cdc6f46e357 --- /dev/null +++ b/tests/test_litellm/llms/zai/test_zai_anthropic_messages_transformation.py @@ -0,0 +1,96 @@ +import litellm +from litellm.llms.anthropic.experimental_pass_through.messages.transformation import ( + AnthropicMessagesConfig, +) +from litellm.llms.zai.messages.transformation import ZAIAnthropicMessagesConfig +from litellm.utils import ProviderConfigManager + + +def test_zai_provider_uses_anthropic_messages_config(): + config = ProviderConfigManager.get_provider_anthropic_messages_config( + model="glm-5.3", + provider=litellm.LlmProviders.ZAI, + ) + + assert isinstance(config, ZAIAnthropicMessagesConfig) + assert config.custom_llm_provider == "zai" + + +def test_anthropic_provider_keeps_default_config_for_zai_named_model(): + config = ProviderConfigManager.get_provider_anthropic_messages_config( + model="glm-5.3", + provider=litellm.LlmProviders.ANTHROPIC, + ) + + assert isinstance(config, AnthropicMessagesConfig) + assert not isinstance(config, ZAIAnthropicMessagesConfig) + + +def test_zai_anthropic_messages_config_defaults(): + config = ZAIAnthropicMessagesConfig() + + assert config.custom_llm_provider == "zai" + assert config.get_api_base() == "https://api.z.ai/api/anthropic" + + +def test_zai_anthropic_messages_url_defaults_to_anthropic_endpoint(): + config = ZAIAnthropicMessagesConfig() + + url_cases = { + None: "https://api.z.ai/api/anthropic/v1/messages", + "https://api.z.ai/api/anthropic": "https://api.z.ai/api/anthropic/v1/messages", + "https://api.z.ai/api/anthropic/v1": "https://api.z.ai/api/anthropic/v1/messages", + "https://api.z.ai/api/anthropic/v1/messages": "https://api.z.ai/api/anthropic/v1/messages", + "https://api.z.ai/api": "https://api.z.ai/api/anthropic/v1/messages", + } + + for api_base, expected_url in url_cases.items(): + assert ( + config.get_complete_url( + api_base=api_base, + api_key=None, + model="glm-5.3", + optional_params={}, + litellm_params={}, + ) + == expected_url + ) + + +def test_zai_anthropic_messages_headers_use_zai_key(): + config = ZAIAnthropicMessagesConfig() + + headers, api_base = config.validate_anthropic_messages_environment( + headers={}, + model="glm-5.3", + messages=[], + optional_params={}, + litellm_params={}, + api_key="sk-zai", + api_base="https://example.test/anthropic", + ) + + assert api_base == "https://example.test/anthropic" + assert headers["x-api-key"] == "sk-zai" + assert headers["anthropic-version"] == "2023-06-01" + assert headers["content-type"] == "application/json" + + +def test_zai_anthropic_messages_respects_existing_case_insensitive_auth_headers(): + config = ZAIAnthropicMessagesConfig() + + headers, _ = config.validate_anthropic_messages_environment( + headers={"Authorization": "Bearer caller-token"}, + model="glm-5.3", + messages=[], + optional_params={}, + litellm_params={}, + api_key="sk-zai", + api_base="https://api.z.ai/api/anthropic", + ) + + assert headers == { + "Authorization": "Bearer caller-token", + "anthropic-version": "2023-06-01", + "content-type": "application/json", + } diff --git a/tests/test_litellm/llms/zai/test_zai_responses_transformation.py b/tests/test_litellm/llms/zai/test_zai_responses_transformation.py new file mode 100644 index 00000000000..40f69a77943 --- /dev/null +++ b/tests/test_litellm/llms/zai/test_zai_responses_transformation.py @@ -0,0 +1,96 @@ +import litellm +from litellm.llms.zai.responses.transformation import ZAIResponsesAPIConfig +from litellm.types.router import GenericLiteLLMParams +from litellm.types.utils import LlmProviders +from litellm.utils import ProviderConfigManager + + +def test_zai_provider_uses_responses_api_config(): + config = ProviderConfigManager.get_provider_responses_api_config( + model="glm-5.3", + provider=litellm.LlmProviders.ZAI, + ) + + assert isinstance(config, ZAIResponsesAPIConfig) + assert config.custom_llm_provider == LlmProviders.ZAI + + +def test_zai_responses_url_defaults_to_responses_endpoint(): + config = ZAIResponsesAPIConfig() + + url_cases = { + None: "https://api.z.ai/api/v1/responses", + "https://api.z.ai/api/v1": "https://api.z.ai/api/v1/responses", + "https://api.z.ai/api/v1/": "https://api.z.ai/api/v1/responses", + "https://api.z.ai/api/v1/responses": "https://api.z.ai/api/v1/responses", + } + + for api_base, expected_url in url_cases.items(): + assert config.get_complete_url(api_base=api_base, litellm_params={}) == expected_url + + +def test_zai_responses_url_ignores_chat_completions_api_base(): + config = ZAIResponsesAPIConfig() + + chat_bases = ( + "https://api.z.ai/api/paas/v4", + "https://api.z.ai/api/paas/v4/", + "https://api.z.ai/api/coding/paas/v4", + ) + + for chat_base in chat_bases: + assert config.get_complete_url(api_base=chat_base, litellm_params={}) == "https://api.z.ai/api/v1/responses" + + +def test_zai_responses_url_keeps_custom_api_base(): + config = ZAIResponsesAPIConfig() + + assert ( + config.get_complete_url( + api_base="https://gateway.example.com/openai/v1", + litellm_params={}, + ) + == "https://gateway.example.com/openai/v1/responses" + ) + + +def test_zai_responses_headers_use_bearer_token(): + config = ZAIResponsesAPIConfig() + litellm_params = GenericLiteLLMParams(api_key="sk-zai") + + headers = config.validate_environment( + headers={}, + model="glm-5.3", + litellm_params=litellm_params, + ) + + assert headers["Authorization"] == "Bearer sk-zai" + assert headers["Content-Type"] == "application/json" + + +def test_zai_responses_headers_fall_back_to_environment_key(monkeypatch): + monkeypatch.setenv("ZAI_API_KEY", "sk-zai-env") + config = ZAIResponsesAPIConfig() + + headers = config.validate_environment( + headers={}, + model="glm-5.3", + litellm_params=GenericLiteLLMParams(), + ) + + assert headers["Authorization"] == "Bearer sk-zai-env" + + +def test_zai_responses_headers_prefer_zai_key_over_global_key(monkeypatch): + monkeypatch.setenv("ZAI_API_KEY", "sk-zai-env") + monkeypatch.setattr(litellm, "api_key", "sk-global-other-provider", raising=False) + + config = ZAIResponsesAPIConfig() + + headers = config.validate_environment( + headers={}, + model="glm-5.3", + litellm_params=GenericLiteLLMParams(), + ) + + assert headers["Authorization"] == "Bearer sk-zai-env"