Is response_format in LiteLLM essentially implemented as an implicit tool call for structured output?

[Nial4]

Hi, I’m currently exploring structured outputs in LLM applications, and I have a question about how response_format is implemented in LiteLLM — particularly in relation to tool calls.

Is LiteLLM’s response_format internally implemented using the model's tool calling mechanism, by injecting an implicit tool based on the schema?

As a reference point:
LangChain’s with_structured_output(SchemaClass) uses a similar approach. Internally, it binds the Pydantic schema as a tool via bind_tools, essentially converting structured output into a tool-calling use case.