2 ms·

I assume the model is probably generating a search query, then the results get inserted as context and it uses that to formulate a response.