Expand description
StepFun provider backed by the native Responses API
(POST {base}/responses).
StepFun’s /responses surface is wire-compatible with the OpenAI Responses
API: history is replayed as input items (message, function_call,
function_call_output), the system prompt travels as top-level
instructions, reasoning effort is encoded as reasoning.effort, and the
output-token budget uses max_output_tokens.
This replaces the previous chat-completions implementation. The provider
keeps the registration contract expected by impl_standard_provider_constructor!
(type name + 7-argument from_config) and reuses the shared Responses
streaming adapter so normalized stream events stay consistent with the native
OpenAI provider.