Question about API choice

Hello there,

Currently I’m struggling with understanding witch API should I choose for building “ai chatbot for clients support service” I would like to believe that openai api platform could handle this request

There are few questions actually:

  1. Should I pick assistant api or simple chat api? Does assistant API capable of fine-tunning? I would like to “train” assistant with prompt and data, and then assist each individual client in separate threads
  2. As far as I get the idea, the completion api aren’t capable to work with context and continuous dialogs in threads, am I right?
  3. Retrievals. Each time I add files through web assistant menu paper clip icon, the actual thread run asks me about this files, should it checkout them or not, Am I right in considering this Retrievals as knowledge base to “instructed” assistant, or I’m terribly wrong in abusing assistant at all? Should I use separate thread for teaching or how should this idea work at all?

Thanks in advance, the openai’s chat custom gpt works perfectly, just can’t get the idea how to build the same without context in completion api or how to teach assistant api right way.

1 Like

Heya! Welcome to the forum.

The quickstart and cook book are great places to start.

Assistant API has more limitations (at the moment) but is a bit easier to set-up. Regular Chat Completion has you doing more of the RAG (retrieval augmented generation) on your own so you have more control.

Basically, yeah. You can trick completion into doing it, but ChatML and Chat Completion are the way forward, and I would recommend using them.

1 Like