Skip to main content

portia.config

Configuration module for the SDK.

This module defines the configuration classes and enumerations used in the SDK, including settings for storage, API keys, LLM providers, logging, and agent options. It also provides validation for configuration values and loading mechanisms for default settings.

StorageClass Objects​

class StorageClass(Enum)

Enum representing locations plans and runs are stored.

Attributes:

  • MEMORY - Stored in memory.
  • DISK - Stored on disk.
  • CLOUD - Stored in the cloud.

Model Objects​

class Model(NamedTuple)

Provider and model name tuple.

DEPRECATED Use new model configuration options on Config class instead.

Attributes:

  • provider - The provider of the model.
  • model_name - The name of the model in the provider's API.

LLMModel Objects​

class LLMModel(Enum)

Enum for supported LLM models.

DEPRECATED Use new model configuration options on Config class instead.

Models are grouped by provider, with the following providers:

  • OpenAI
  • Anthropic
  • MistralAI
  • Google Generative AI
  • Azure OpenAI

Attributes:

  • GPT_4_O - GPT-4 model by OpenAI.

  • GPT_4_O_MINI - Mini GPT-4 model by OpenAI.

  • GPT_3_5_TURBO - GPT-3.5 Turbo model by OpenAI.

  • CLAUDE_3_5_SONNET - Claude 3.5 Sonnet model by Anthropic.

  • CLAUDE_3_5_HAIKU - Claude 3.5 Haiku model by Anthropic.

  • CLAUDE_3_OPUS - Claude 3.0 Opus model by Anthropic.

  • CLAUDE_3_7_SONNET - Claude 3.7 Sonnet model by Anthropic.

  • MISTRAL_LARGE - Mistral Large Latest model by MistralAI.

  • GEMINI_2_0_FLASH - Gemini 2.0 Flash model by Google Generative AI.

  • GEMINI_2_0_FLASH_LITE - Gemini 2.0 Flash Lite model by Google Generative AI.

  • GPT_4_O_MINI0 - Gemini 1.5 Flash model by Google Generative AI.

  • GPT_4_O_MINI1 - GPT-4 model by Azure OpenAI.

  • GPT_4_O_MINI2 - Mini GPT-4 model by Azure OpenAI.

  • GPT_4_O_MINI3 - O3 Mini model by Azure OpenAI.

    Can be instantiated from a string with the following format:

    • provider/model_name [e.g. LLMModel("openai/gpt-4o")]
    • model_name [e.g. LLMModel("gpt-4o")]

    In the cases where the model name is not unique across providers, the earlier values in the enum definition will take precedence.

_missing_​

@classmethod
def _missing_(cls, value: object) -> LLMModel

Get the LLM model from the model name.

api_name​

@property
def api_name() -> str

Override the default value to return the model name.

provider​

def provider() -> LLMProvider

Get the associated provider for the model.

Returns:

  • LLMProvider - The provider associated with the model.

to_model_string​

def to_model_string() -> str

Get the model string for the model.

Returns:

  • str - The model string.

_AllModelsSupportedWithDeprecation Objects​

class _AllModelsSupportedWithDeprecation(Container)

A type that returns True for any contains check.

__contains__​

def __contains__(item: object) -> bool

Check if the item is in the container.

ExecutionAgentType Objects​

class ExecutionAgentType(Enum)

Enum for types of agents used for executing a step.

Attributes:

  • ONE_SHOT - The one-shot agent.
  • DEFAULT - The default agent.

PlanningAgentType Objects​

class PlanningAgentType(Enum)

Enum for planning agents used for planning queries.

Attributes:

  • DEFAULT - The default planning agent.

LogLevel Objects​

class LogLevel(Enum)

Enum for available log levels.

Attributes:

  • DEBUG - Debug log level.
  • INFO - Info log level.
  • WARNING - Warning log level.
  • ERROR - Error log level.
  • CRITICAL - Critical log level.

parse_str_to_enum​

def parse_str_to_enum(value: str | E, enum_type: type[E]) -> E

Parse a string to an enum or return the enum as is.

Arguments:

  • value str | E - The value to parse.
  • enum_type type[E] - The enum type to parse the value into.

Raises:

  • InvalidConfigError - If the value cannot be parsed into the enum.

Returns:

  • E - The corresponding enum value.

GenerativeModelsConfig Objects​

class GenerativeModelsConfig(BaseModel)

Configuration for a Generative Models.

These models do not all need to be specified manually. If an LLM provider is configured, Portia will use default models that are selected for the particular use-case.

Attributes:

  • default_model - The default generative model to use. This model is used as the fallback model if no other model is specified. It is also used by default in the Portia SDK tool that require an LLM.

  • planning_model - The model to use for the PlanningAgent. Reasoning models are a good choice here, as they are able to reason about the problem and the possible solutions. If not specified, the default_model will be used.

  • execution_model - The model to use for the ExecutionAgent. This model is used for the distilling context from the plan run into tool calls. If not specified, the default_model will be used.

  • introspection_model - The model to use for the IntrospectionAgent. This model is used to introspect the problem and the plan. If not specified, the default_model will be used.

  • summarizer_model - The model to use for the SummarizerAgent. This model is used to summarize output from the plan run. If not specified, the default_model will be used.

parse_models​

@model_validator(mode="before")
@classmethod
def parse_models(cls, data: dict[str, Any]) -> dict[str, Any]

Convert legacy LLMModel values to str with deprecation warning.

Config Objects​

class Config(BaseModel)

General configuration for the SDK.

This class holds the configuration for the SDK, including API keys, LLM settings, logging options, and storage settings. It also provides validation for configuration consistency and offers methods for loading configuration from files or default values.

Attributes:

  • portia_api_endpoint - The endpoint for the Portia API.
  • portia_api_key - The API key for Portia.
  • openai_api_key - The API key for OpenAI.
  • anthropic_api_key - The API key for Anthropic.
  • mistralai_api_key - The API key for MistralAI.
  • google_api_key - The API key for Google Generative AI.
  • azure_openai_api_key - The API key for Azure OpenAI.
  • azure_openai_endpoint - The endpoint for Azure OpenAI.
  • llm_provider - The LLM provider. If set, Portia uses this to select the best models for each agent. Can be None if custom models are provided.
  • models - A configuration for the LLM models for Portia to use.
  • portia_api_key0 - The storage class used (e.g., MEMORY, DISK, CLOUD).
  • portia_api_key1 - The directory for storage, if applicable.
  • portia_api_key2 - The default log level (e.g., DEBUG, INFO).
  • portia_api_key3 - The default destination for logs (e.g., sys.stdout).
  • portia_api_key4 - Whether to serialize logs in JSON format.
  • portia_api_key5 - The planning agent type.
  • portia_api_key6 - The execution agent type.
  • portia_api_key7 - A dictionary of feature flags for the SDK.

parse_feature_flags​

@model_validator(mode="after")
def parse_feature_flags() -> Self

Add feature flags if not provided.

setup_cache​

@model_validator(mode="after")
def setup_cache() -> Self

Set up LLM cache if Redis URL is provided.

parse_storage_class​

@field_validator("storage_class", mode="before")
@classmethod
def parse_storage_class(cls, value: str | StorageClass) -> StorageClass

Parse storage class to enum if string provided.

parse_default_log_level​

@field_validator("default_log_level", mode="before")
@classmethod
def parse_default_log_level(cls, value: str | LogLevel) -> LogLevel

Parse default_log_level to enum if string provided.

parse_execution_agent_type​

@field_validator("execution_agent_type", mode="before")
@classmethod
def parse_execution_agent_type(
        cls, value: str | ExecutionAgentType) -> ExecutionAgentType

Parse execution_agent_type to enum if string provided.

parse_planning_agent_type​

@field_validator("planning_agent_type", mode="before")
@classmethod
def parse_planning_agent_type(
        cls, value: str | PlanningAgentType) -> PlanningAgentType

Parse planning_agent_type to enum if string provided.

exceeds_output_threshold​

def exceeds_output_threshold(value: str | list[str | dict]) -> bool

Determine whether the provided output value exceeds the large output threshold.

get_agent_default_model​

def get_agent_default_model(
        agent_key: str,
        llm_provider: LLMProvider | None = None
) -> GenerativeModel | str | None

Get the default model for the given agent key.

fill_default_models​

@model_validator(mode="after")
def fill_default_models() -> Self

Fill in default models for the LLM provider if not provided.

check_config​

@model_validator(mode="after")
def check_config() -> Self

Validate Config is consistent.

from_default​

@classmethod
def from_default(cls, **kwargs) -> Config

Create a Config instance with default values, allowing overrides.

Returns:

  • Config - The default config

has_api_key​

def has_api_key(name: str) -> bool

Check if the given API Key is available.

must_get_api_key​

def must_get_api_key(name: str) -> SecretStr

Retrieve the required API key for the configured provider.

Raises:

  • ConfigNotFoundError - If no API key is found for the provider.

Returns:

  • SecretStr - The required API key.

must_get​

def must_get(name: str, expected_type: type[T]) -> T

Retrieve any value from the config, ensuring its of the correct type.

Arguments:

  • name str - The name of the config record.
  • expected_type type[T] - The expected type of the value.

Raises:

  • ConfigNotFoundError - If no API key is found for the provider.
  • InvalidConfigError - If the config isn't valid

Returns:

  • T - The config value

get_default_model​

def get_default_model() -> GenerativeModel

Get or build the default model from the config.

The default model will always be present. It is a general purpose model that is used for the SDK's LLM-based Tools, such as the ImageUnderstandingTool and the LLMTool.

Additionally, unless specified all other specific agent models will default to this model.

get_planning_model​

def get_planning_model() -> GenerativeModel

Get or build the planning model from the config.

See the GenerativeModelsConfig class for more information

get_execution_model​

def get_execution_model() -> GenerativeModel

Get or build the execution model from the config.

See the GenerativeModelsConfig class for more information

get_introspection_model​

def get_introspection_model() -> GenerativeModel

Get or build the introspection model from the config.

See the GenerativeModelsConfig class for more information

get_summarizer_model​

def get_summarizer_model() -> GenerativeModel

Get or build the summarizer model from the config.

See the GenerativeModelsConfig class for more information

get_generative_model​

def get_generative_model(
        model: str | GenerativeModel | None) -> GenerativeModel | None

Get a GenerativeModel instance.

Arguments:

  • model str | GenerativeModel | None - The model to get, either specified as a string in the form of "provider/model_name", or as a GenerativeModel instance. Also accepts None, in which case None is returned.

Returns:

GenerativeModel | None: The model instance or None.

llm_provider_default_from_api_keys​

def llm_provider_default_from_api_keys(**kwargs) -> LLMProvider | None

Get the default LLM provider from the API keys.

Returns:

  • LLMProvider - The default LLM provider.
  • None - If no API key is found.

default_config​

def default_config(**kwargs) -> Config

Return default config with values that can be overridden.

Returns:

  • Config - The default config