I’m trying to test my custom GTP on a local server. I have my frontend (Postman) and backend (Flask app) working and I’m getting a 200 OK and a “Hello: How can I help you today?” message. In my app.py code I’m using gpt-4 as the “model” - BUT - as soon as I change it to the custom GPT “model or ID” provided under the GPT name in playground assistant it says “does not exist or you don’t have access to it”. Am I not using the correct ID? What am I doing wrong? Anyone out there have a similar problem?
Perhaps you are simply using Chat Completions, where you must specify the model name from an existing model. Then define behavior with a system message, and provide other messages for contextual chat history and new user input.
There is no such thing as “Custom GPT”.
If you are talking about an Assistant on the Assistants endpoint, that you created through the web UI, it needs a complete code solution for the multi-step process of creating threads, creating messages within threads, running the thread against the Assistant ID that has “instructions”, polling or capturing events to await completion…
https://platform.openai.com/docs/assistants/overview
Apart from the internal file search and code interpreter tools, Assistants is basically server-side conversation history. You can replicate “instructions” by just putting that as initial “system message” in your chat completions API call.
1 Like