Skip to content

karenina.adapters.langchain.models

models

LangChain model classes for custom endpoints.

This module contains custom LangChain model classes for providers that need special handling (OpenRouter, custom OpenAI-compatible endpoints).

These classes are adapter-internal and should not be imported directly. Use the LLMPort interface via get_llm() instead.

Classes

ChatOpenAIEndpoint

Bases: ChatOpenAI

ChatOpenAI wrapper for user-provided custom endpoints.

Unlike ChatOpenRouter, this does NOT automatically read from environment. API key must be explicitly provided by the user.

This class is used for connecting to OpenAI-compatible endpoints like: - vLLM servers - SGLang servers - Ollama with OpenAI compatibility layer - Text Generation Inference (TGI) - Local LLM servers

Source code in src/karenina/adapters/langchain/models.py 
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
class ChatOpenAIEndpoint(ChatOpenAI):
    """ChatOpenAI wrapper for user-provided custom endpoints.

    Unlike ChatOpenRouter, this does NOT automatically read from environment.
    API key must be explicitly provided by the user.

    This class is used for connecting to OpenAI-compatible endpoints like:
    - vLLM servers
    - SGLang servers
    - Ollama with OpenAI compatibility layer
    - Text Generation Inference (TGI)
    - Local LLM servers
    """

    openai_api_key: SecretStr | None = Field(alias="api_key", default=None)

    @property
    def lc_secrets(self) -> dict[str, str]:
        # Return empty dict - we don't want LangChain trying to read from env
        return {}

    def __init__(
        self,
        base_url: str,
        openai_api_key: str | SecretStr | None = None,
        **kwargs: Any,
    ) -> None:
        # Do NOT fallback to environment - require explicit API key
        if openai_api_key is None:
            raise ValueError(
                "API key is required for openai_endpoint interface. "
                "This interface does not automatically read from environment variables."
            )

        if isinstance(openai_api_key, str):
            openai_api_key = SecretStr(openai_api_key)

        # Normalize URL to ensure it ends with /v1
        normalized_url = _normalize_openai_endpoint_url(base_url)

        super().__init__(
            base_url=normalized_url,
            api_key=openai_api_key,
            **kwargs,
        )

ChatOpenRouter

Bases: ChatOpenAI

LangChain ChatOpenAI wrapper for OpenRouter API.

OpenRouter provides a unified API for accessing multiple LLM providers. This class automatically configures the correct base URL and API key handling for OpenRouter.

The API key is read from the OPENROUTER_API_KEY environment variable if not explicitly provided.

Source code in src/karenina/adapters/langchain/models.py 
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
class ChatOpenRouter(ChatOpenAI):
    """LangChain ChatOpenAI wrapper for OpenRouter API.

    OpenRouter provides a unified API for accessing multiple LLM providers.
    This class automatically configures the correct base URL and API key
    handling for OpenRouter.

    The API key is read from the OPENROUTER_API_KEY environment variable
    if not explicitly provided.
    """

    openai_api_key: SecretStr | None = Field(alias="api_key", default=None)

    @property
    def lc_secrets(self) -> dict[str, str]:
        return {"openai_api_key": "OPENROUTER_API_KEY"}

    def __init__(self, openai_api_key: str | None = None, **kwargs: Any) -> None:
        openai_api_key = openai_api_key or os.environ.get("OPENROUTER_API_KEY")
        super().__init__(
            base_url="https://openrouter.ai/api/v1",
            api_key=SecretStr(openai_api_key) if openai_api_key else None,
            **kwargs,
        )