well now that I have created an assistant, for example, an appointment manager, I have to allow a doctor, for example, to use it for his patients. I see two paths, I can manage the dialogue with the patient in my own interface and therefore I would use the assistant only for natural language understanding which seems a bit limiting to me. The second way that I don’t know how to implement is to use the assistant 100%, and I would limit myself to managing function calls for example. I tried the link to the assistant https://platform.openai.com/playground?assistant=asst_J00a337… but it presents me with the OpenAi interface and asks to log in or register. Instead, I would like to show an interface that says “Welcome to Dr. Mario Bianchi’s assistant …”. Then my application would authenticate to OpenAI.
Use of assistants will typically integrate a lot of your own code - starting with exactly what you’d need to do yourself without the assistant agent that only offers more difficulty and higher unmanaged cost than direct model interaction with chat completions.
assistants still requires:
- user interface
- database (customer, billing, moderation/bans, limits)
- database (customer’s chat lists, titles, enabled features)
- database (customer file management, file results for recall/download)
- backend that secures customer credentials and your own API credentials
- …
These don’t go away with assistants. Interacting with API just becomes that much more tedious and less responsive.
Then of course the logic behind talking to an AI about appointments database — instead of just having a GUI web interface for the same, with more useful search boxes and visual calendar display.
It sounds like you don’t want to develop any of this. You’d likely want to bill for the work of developing a GPT for ChatGPT Plus, where the AI costs are paid by the client’s ChatGPT plus subscription, and then any custom external API that the GPT needs for scheduling actions (tool calls) and authentication you can make your own ongoing billing.