feat: Add reasoning_effort parameter support for OpenAI reasoning models

[ron-42]

Description

This PR adds support for the reasoning_effort parameter for OpenAI's reasoning models (o1, o3, and gpt-5 series). This parameter allows users to control the trade-off between reasoning depth and response speed.

What is reasoning_effort?

The reasoning_effort parameter controls how much computational effort the model puts into reasoning before generating a response. It accepts three values:

• "low" - Faster responses with less reasoning
• "medium" - Balanced approach (default behavior)
• "high" - More thorough reasoning, slower responses

This parameter is only applicable to OpenAI's reasoning models: o1, o1-preview, o1-mini, o3, o3-mini, gpt-5, gpt-5o, gpt-5o-mini, and gpt-5o-micro.

Changes Made

• Added reasoning_effort parameter to BaseLlmConfig class with type safety (Literal["low", "medium", "high"])
• Updated all LLM configuration classes to support the new parameter:
  • OpenAIConfig
  • AzureOpenAIConfig
  • AnthropicConfig
  • OllamaConfig
  • DeepSeekConfig
  • LMStudioConfig
  • VllmConfig
  • AWSBedrockConfig
• Added validation logic in _validate_config() to ensure only valid values are accepted
• Added smart filtering in _get_supported_params() to only pass the parameter to reasoning models
• Added warning when reasoning_effort is set for non-reasoning models
• Created comprehensive test suite with 9 unit tests covering:
  • Parameter passing to reasoning models (o1, o3, gpt-5)
  • Validation of valid and invalid values
  • Filtering for non-reasoning models
  • Optional live API integration test
• Updated documentation:
  • docs/components/llms/config.mdx - Updated parameter table
  • docs/components/llms/models/openai.mdx - Added usage example
  • docs/components/llms/models/azure_openai.mdx - Added usage example

Usage Example

OpenAI

from mem0 import Memory

config = {
    "llm": {
        "provider": "openai",
        "config": {
            "model": "o1-preview",
            "temperature": 1,
            "max_tokens": 2000,
            "reasoning_effort": "high"  # New parameter
        }
    }
}

m = Memory.from_config(config)
m.add("I love programming in Python", user_id="user123")

Azure OpenAI

from mem0 import Memory

config = {
    "llm": {
        "provider": "azure_openai",
        "config": {
            "model": "o1-preview",
            "azure_kwargs": {
                "api_version": "2024-09-01-preview",
                "azure_endpoint": "https://your-endpoint.openai.azure.com/",
                "azure_deployment": "your-o1-deployment"
            },
            "reasoning_effort": "medium"  # New parameter
        }
    }
}

m = Memory.from_config(config)

Testing

To run the new tests:

# Run all reasoning_effort tests
pytest tests/llms/test_openai_reasoning_effort.py -v

# Run a specific test
pytest tests/llms/test_openai_reasoning_effort.py::test_reasoning_effort_with_o1 -v

# Run with live API (requires OPENAI_API_KEY)
pytest tests/llms/test_openai_reasoning_effort.py::test_reasoning_effort_live_api -v

Checklist

• Added reasoning_effort parameter to all config classes
• Implemented validation for parameter values
• Implemented smart filtering (only pass to reasoning models)
• Added warning for non-reasoning models
• Created comprehensive test suite
• Updated documentation
• Followed existing code style and patterns
• All tests pass

Related Issues

Fixes #3651