I just created a GPT with Actions. I am using it as a user interface to my existing RAG API. In other words, instead of uploading the knowledge to the GPT, I use an action call for the GPT to request the knowledge from my API. Overall, I really like what I’m able to do. But, I do have some problems with / questions about the process.
- 45 second timeout is WAY too low. My API uses gpt-4.5-turbo-preview, and I don’t have to tell you how slow it can be sometimes. No way to increase this timeout?
- Is there a specific OpenAI IP block used for GPT action calls? If so, how do I find out what it is? I would like to know the calls are coming from OpenAI servers, even if they are public calls.
- Is there some sort of logging mechanism so we can see what the GPT has sent and the API server is sending back when a call fails? Troubleshooting failed calls right now is really difficult.
- Is there some way to pass instructions to GPT users on how to use the GPT other than in the “description” field or entering into the GPT instructions? Like a Help screen? If not, there should be!
- There appears to be some sort of caching mechanism. When I go in and make changes to OAS specifications or instructions, then save, then go back to GPT screen – it doesn’t appear to always recognize the changes, especially those that pertain to the schema.
- Is there a way to send information via the API call that will tell me what ChatGPT user is actually using the GPT at that moment?
- How do I create a link to my website in the GPT by line?
Overall this is what I have observed so far. It’s probably an issue for Support (how do we contact them?), but since they take a while to get back to us, I thought I’d ask the knowledgeable folks here who’ve been doing this for a while.
Thanks!