As with the GPT-4 Turbo model, is it possible to obtain an API from GPTs that have been created and trained on custom data to use in my own application?
You will have to build an assistant using Assistants api and feed it with the custom instructions of your trained GPT to convert to an api
3 Likes
But that means that if I want many users to interact with the assistant I would need to spin a new version of the assistant for each user, correct?
That wouldn’t save much in terms of time and tokens.
Is there a way to have such an assistant or to have a pre-prompt always in place instead of having to create one every time a conversation is started?
Having an existing set of instructions that is always acted upon as the behavior of the AI is the whole point of assistants.
One assistant that you set up can interact with multiple users.
A user’s conversation is stored in a thread. A thread has messages specific to the user. A new user input into a thread has an assistant response written to it by starting a run. You simply tell the run which assistant and which thread.