Make LLM directly use the response of a GET API from the API Loader without the need of a Vector Store

[fabiopipitone]

Hi there,

I'd like to extract some documents from my vector store (let's say Elasticsearch) via custom API (so I can build the body of the query) and use the response to directly instruct the LLM prompt.

Let's say I have an ES index "myrecipes" with docs containing some recipes and structured as follows:

{
  "recipe_id": "123",
  "recipe_name": "foo",
  "recipe_desc": "let the water boil before putting the pasta in it"
}

Let's say I query the ES index via the API Loader and extract the top 3 recipes. Then I want to pass it to a Prompt Template (or Chain Prompt Template) that I'm building to instruct the LLM as a food critic that has to make a review of each of the retrieved recipes.
How can I use the documents retrieved by the API Loader in the Prompt Template? Cause the output of the API Loader is of Document type and it seems to only match the Document Store input.

Here's where I'm stuck

Any ideas?
Thank you 🙏

P.S. actually, what I'd need to do is to let the user write an ingredient, make an API GET request to elasticsearch using that ingredient as query string, take the top 3 results and then proceed as mentioned above, but let's take one step at a time :)