Why does Guard.parse use an LLM API in LangChain integration?

[jamesbraza]

Reading the current (guardrails-ai==0.2.9) LangChain integration:

1. langchain.llms.OpenAI gets called
2. Then Guard.parse takes an llm_api argument

Why does Guard.parse accept an llm_api argument? Will it invoke the LLM a second time?

[irgolic]

Hiya, it accepts llm_api because Guardrails validators have the ability to "reask" the LLM if the validator fails. The supplied llm_api will be used for this.

For an example of reasking, see here: https://docs.guardrailsai.com/defining_guards/pydantic/#structured-output-with-validation