Implements native support for OpenAI's Responses API (/v1/responses) as a new
LLM provider in CrewAI. This addresses feature request #4152.
Key features:
- New OpenAIResponsesCompletion class extending BaseLLM
- Support for both explicit provider parameter and model prefix routing
- Message conversion from CrewAI format to Responses API format
- Tool/function calling support
- Streaming support (sync and async)
- Structured output via Pydantic models
- Token usage tracking
- Support for o-series reasoning models with reasoning_effort parameter
- Support for stateful conversations via previous_response_id
Usage:
# Option 1: Using provider parameter
llm = LLM(model='gpt-4o', provider='openai_responses')
# Option 2: Using model prefix
llm = LLM(model='openai_responses/gpt-4o')
Includes comprehensive test coverage for:
- Provider routing
- Message conversion
- Tool conversion
- API calls
- Parameter preparation
- Context window sizes
- Feature support methods
- Token usage extraction
Co-Authored-By: João <joao@crewai.com>
9c5b68f1c5