Fine tunning a model to use a base context

I have a faqs (frequently asked questions) and want to create a chatbot to answer questions about that faqs. I have tested with embeddings and it works well…but doesnt have sense to send always the embeddings adding the token cost for a simple question

So I have tested with fine tunning a model, where I finne tunning sending a file with the question in the prompt and the answer in the completion…since it is a basic faq it has arround 50 entries…but doesnt work, when I ask something to the new model it doesnt use my faqs to answer…

I have searched about that and it is explained in different ways: context, memory …and speaking that currently is not possible if you dont send the context inside the new prompt

But I dont understand so if anyone can help I would be grateful.

Why I cant to create a new model with more knowledge? if the davinci model understands if I sent in the context “The price of the service is $15” or summarize a text…why i cant to create a new model equals to that but with those knowledge inside so when I start it and ask about any question of the previous summarized text it answers using it?