plexus.cli.procedure.model_config module

Model Configuration System for SOP Agent

This module provides a model-agnostic configuration system that allows different LLM models (GPT-4, GPT-5, O3, etc.) to be configured with their specific parameters without hardcoding model-specific logic.

The configuration system: 1. Allows dynamic parameter passing through model_kwargs 2. Supports different parameter sets for different models 3. Maintains backward compatibility with existing GPT-4 configurations 4. Provides clean separation between model configuration and business logic

class plexus.cli.procedure.model_config.ModelConfig(model: str = 'gpt-5', temperature: float | None = None, max_tokens: int | None = None, reasoning_effort: str | None = None, verbosity: str | None = None, model_kwargs: ~typing.Dict[str, ~typing.Any] = <factory>, openai_api_key: str | None = None, stream: bool = False)

Bases: object

Model configuration that supports any LLM model with dynamic parameters.

This approach lets configuration determine what parameters are valid rather than hardcoding model-specific parameter validation in code.

__init__(model: str = 'gpt-5', temperature: float | None = None, max_tokens: int | None = None, reasoning_effort: str | None = None, verbosity: str | None = None, model_kwargs: ~typing.Dict[str, ~typing.Any] = <factory>, openai_api_key: str | None = None, stream: bool = False) → None

create_langchain_llm() → ChatOpenAI

Create a LangChain ChatOpenAI instance with this configuration.

This method handles all the parameter passing complexity so that the calling code doesn’t need to know about model differences.

classmethod from_dict(config_dict: Dict[str, Any]) → ModelConfig

Create ModelConfig from a dictionary (e.g., loaded from YAML/JSON).

This allows configuration to be externalized to config files.

max_tokens: int | None = None

model: str = 'gpt-5'

model_kwargs: Dict[str, Any]

openai_api_key: str | None = None

reasoning_effort: str | None = None

stream: bool = False

temperature: float | None = None

to_langchain_kwargs() → Dict[str, Any]

Convert this config to kwargs suitable for LangChain ChatOpenAI.

This method builds the parameter dict dynamically, only including parameters that are actually set. This allows: - GPT-4 configs to include temperature - GPT-5 configs to include reasoning_effort and verbosity - Any model to include custom parameters via model_kwargs - Invalid parameters to cause errors at the OpenAI API level

verbosity: str | None = None

class plexus.cli.procedure.model_config.ModelConfigs

Bases: object

Predefined model configurations for common scenarios.

These serve as examples and can be customized or extended.

static from_environment() → ModelConfig

Create configuration from environment variables.

This allows runtime configuration without code changes: - MODEL_NAME=gpt-5 - MODEL_TEMPERATURE=0.3 - MODEL_REASONING_EFFORT=high - MODEL_VERBOSITY=medium - etc.

static gpt_4o_default() → ModelConfig

Standard GPT-4o configuration.

static gpt_4o_precise() → ModelConfig

GPT-4o with low temperature for precise tasks.

static gpt_5_default() → ModelConfig

Standard GPT-5 configuration with supported parameters.

static gpt_5_high_reasoning() → ModelConfig

GPT-5 configured for complex reasoning tasks.

static gpt_5_mini_fast() → ModelConfig

GPT-5 mini for fast responses.

plexus.cli.procedure.model_config.create_configured_llm(model_config: ModelConfig | Dict[str, Any] | str | None = None, **override_kwargs) → ChatOpenAI

Convenience function to create a configured LLM.

Args:

model_config: Can be:

• ModelConfig instance

• Dict to create ModelConfig from

• String model name (uses defaults)

• None (uses environment or defaults)

**override_kwargs: Override any parameters

Returns:

Configured ChatOpenAI instance