Hi all,
For those integrating GPTs into their products via API—how do you handle testing without affecting the production version? Did you duplicate the app for dedicated sandbox environment? as it wont be the same- would love to hear some tips how do you manage GPT application versions.
Thank you
Hi @asafr93 and welcome to the community!
Can you elaborate on “integrating GPTs into their products via API”? Do you mean integrating custom GPTs? If yes, there is actually no way to integrate API towards custom GPTs. If you mean by using chat completions API, then a standard way is to have a dev and production environment, or in some cases dev, staging and production. Basically just using standard CI/CD principles.
Hi and thank you for your response.
I created a custom gpt, and now I would like to connect my product through API to my custom gpt (as written in the image attached).
Once I establish the connection, users will be able to send requests to the custom gpt through my product.
so now my question is- once my custom gpt connect to my product and bascelliy in production environment. how can I create a sandbox environment?
Let me know if something not clear, maybe I still didn’t explain myself correctly. Thank you
Ok I see - you want to use actions within your custom GPT to talk to your APIs, and your users will interact with your custom GPT directly. Great.
Unfortunately there is no CI/CD process for custom GPTs. There is certainly no way to have a sandboxed environment for it.
I see. So what are my options? dupliace my custom gpt with the exactly same instructions and make it my “sandbox environment”?
If I understand correctly- even though I will duplicate, the gpt wont provide the same responses as the first custom gpt is more trained so his answer will be more accurate. Is this correct?
thanks
Correct, you won’t get the same responses.
My recommendation (how I do things) is to not use custom GPT as any kind of a “product”. I create a custom GPT either for very quick prototyping, or as a way for my team to internally use it for some quick tasks (e.g. a GPT for creating prompts for specific use cases).
For production systems and proper products, I advise to either use Assistant API, or just straight chat completions API, and then you are in full control over how you want to implement, test and deploy your system.
My team at my company has decided to develop a custom GPT and integrate our product with it.
What are our current options if we still want to create an artificial sandbox environment?
Is duplicating the existing custom gpt the only option, even though the output may not be identical?
Looking forward to your response.
Thank you!
