Skip to main content

portia.model

LLM provider model classes for Portia Agents.

Message Objects​

class Message(BaseModel)

Portia LLM message class.

from_langchain​

@classmethod
def from_langchain(cls, message: BaseMessage) -> Message

Create a Message from a LangChain message.

Arguments:

  • message BaseMessage - The LangChain message to convert.

Returns:

  • Message - The converted message.

to_langchain​

def to_langchain() -> BaseMessage

Convert to LangChain BaseMessage sub-type.

Returns:

  • BaseMessage - The converted message, subclass of LangChain's BaseMessage.

LLMProvider Objects​

class LLMProvider(Enum)

Enum for supported LLM providers.

Attributes:

  • OPENAI - OpenAI provider.
  • ANTHROPIC - Anthropic provider.
  • MISTRALAI - MistralAI provider.
  • GOOGLE_GENERATIVE_AI - Google Generative AI provider.
  • AZURE_OPENAI - Azure OpenAI provider.

GenerativeModel Objects​

class GenerativeModel(ABC)

Base class for all generative model clients.

__init__​

def __init__(model_name: str) -> None

Initialize the model.

Arguments:

  • model_name - The name of the model.

get_response​

@abstractmethod
def get_response(messages: list[Message]) -> Message

Given a list of messages, call the model and return its response as a new message.

Arguments:

  • messages list[Message] - The list of messages to send to the model.

Returns:

  • Message - The response from the model.

get_structured_response​

@abstractmethod
def get_structured_response(messages: list[Message],
                            schema: type[BaseModelT]) -> BaseModelT

Get a structured response from the model, given a Pydantic model.

Arguments:

  • messages list[Message] - The list of messages to send to the model.
  • schema type[BaseModelT] - The Pydantic model to use for the response.

Returns:

  • BaseModelT - The structured response from the model.

__str__​

def __str__() -> str

Get the string representation of the model.

__repr__​

def __repr__() -> str

Get the string representation of the model.

to_langchain​

@abstractmethod
def to_langchain() -> BaseChatModel

Get the LangChain client.

LangChainGenerativeModel Objects​

class LangChainGenerativeModel(GenerativeModel)

Base class for LangChain-based models.

__init__​

def __init__(client: BaseChatModel, model_name: str) -> None

Initialize with LangChain client.

Arguments:

  • client - LangChain chat model instance
  • model_name - The name of the model

to_langchain​

def to_langchain() -> BaseChatModel

Get the LangChain client.

get_response​

def get_response(messages: list[Message]) -> Message

Get response using LangChain model.

get_structured_response​

def get_structured_response(messages: list[Message], schema: type[BaseModelT],
                            **kwargs: Any) -> BaseModelT

Get structured response using LangChain model.

Arguments:

  • messages list[Message] - The list of messages to send to the model.
  • schema type[BaseModelT] - The Pydantic model to use for the response.
  • **kwargs - Additional keyword arguments to pass to the with_structured_output method.

Returns:

  • BaseModelT - The structured response from the model.

OpenAIGenerativeModel Objects​

class OpenAIGenerativeModel(LangChainGenerativeModel)

OpenAI model implementation.

__init__​

def __init__(*,
             model_name: str,
             api_key: SecretStr,
             seed: int = 343,
             max_retries: int = 3,
             temperature: float = 0,
             **kwargs: Any) -> None

Initialize with OpenAI client.

Arguments:

  • model_name - OpenAI model to use
  • api_key - API key for OpenAI
  • seed - Random seed for model generation
  • max_retries - Maximum number of retries
  • temperature - Temperature parameter
  • **kwargs - Additional keyword arguments to pass to ChatOpenAI

get_structured_response​

def get_structured_response(messages: list[Message], schema: type[BaseModelT],
                            **kwargs: Any) -> BaseModelT

Call the model in structured output mode targeting the given Pydantic model.

Arguments:

  • messages list[Message] - The list of messages to send to the model.
  • schema type[BaseModelT] - The Pydantic model to use for the response.
  • **kwargs - Additional keyword arguments to pass to the model.

Returns:

  • BaseModelT - The structured response from the model.

get_structured_response_instructor​

def get_structured_response_instructor(messages: list[Message],
                                       schema: type[BaseModelT]) -> BaseModelT

Get structured response using instructor.

AzureOpenAIGenerativeModel Objects​

class AzureOpenAIGenerativeModel(LangChainGenerativeModel)

Azure OpenAI model implementation.

__init__​

def __init__(*,
             model_name: str,
             api_key: SecretStr,
             azure_endpoint: str,
             api_version: str = "2025-01-01-preview",
             seed: int = 343,
             max_retries: int = 3,
             temperature: float = 0,
             **kwargs: Any) -> None

Initialize with Azure OpenAI client.

Arguments:

  • model_name - OpenAI model to use
  • azure_endpoint - Azure OpenAI endpoint
  • api_version - Azure API version
  • seed - Random seed for model generation
  • api_key - API key for Azure OpenAI
  • max_retries - Maximum number of retries
  • temperature - Temperature parameter (defaults to 1 for O_3_MINI, 0 otherwise)
  • **kwargs - Additional keyword arguments to pass to AzureChatOpenAI

get_structured_response​

def get_structured_response(messages: list[Message], schema: type[BaseModelT],
                            **kwargs: Any) -> BaseModelT

Call the model in structured output mode targeting the given Pydantic model.

Arguments:

  • messages list[Message] - The list of messages to send to the model.
  • schema type[BaseModelT] - The Pydantic model to use for the response.
  • **kwargs - Additional keyword arguments to pass to the model.

Returns:

  • BaseModelT - The structured response from the model.

get_structured_response_instructor​

def get_structured_response_instructor(messages: list[Message],
                                       schema: type[BaseModelT]) -> BaseModelT

Get structured response using instructor.

AnthropicGenerativeModel Objects​

class AnthropicGenerativeModel(LangChainGenerativeModel)

Anthropic model implementation.

__init__​

def __init__(*,
             model_name: str = "claude-3-5-sonnet-latest",
             api_key: SecretStr,
             timeout: int = 120,
             max_retries: int = 3,
             max_tokens: int = 8096,
             **kwargs: Any) -> None

Initialize with Anthropic client.

Arguments:

  • model_name - Name of the Anthropic model
  • timeout - Request timeout in seconds
  • max_retries - Maximum number of retries
  • max_tokens - Maximum number of tokens to generate
  • api_key - API key for Anthropic
  • **kwargs - Additional keyword arguments to pass to ChatAnthropic

get_structured_response​

def get_structured_response(messages: list[Message], schema: type[BaseModelT],
                            **kwargs: Any) -> BaseModelT

Call the model in structured output mode targeting the given Pydantic model.

Arguments:

  • messages list[Message] - The list of messages to send to the model.
  • schema type[BaseModelT] - The Pydantic model to use for the response.
  • **kwargs - Additional keyword arguments to pass to the model.

Returns:

  • BaseModelT - The structured response from the model.

get_structured_response_instructor​

def get_structured_response_instructor(messages: list[Message],
                                       schema: type[BaseModelT]) -> BaseModelT

Get structured response using instructor.

map_message_to_instructor​

def map_message_to_instructor(message: Message) -> ChatCompletionMessageParam

Map a Message to ChatCompletionMessageParam.

Arguments:

  • message Message - The message to map.

Returns:

  • ChatCompletionMessageParam - Message in the format expected by instructor.