My Assistant Doesn’t Behave The Same

It is possible that assistants do not behave in the same exact way for a given prompt. Additionally, their behavior may change when called from one environment to another (OpenAI/API). You can perform a simple test by providing a name through instructions and checking if the output is coherent between the playground and the AI Engine.

One prompt might never yield the same answer, so don't expect consistent behavior. Here is an example:

You can also check that the context inside of the thread isn't lost. For this, a simple test would be to ask a sequence of questions and then ask what your first question was. This way, you ensure that the previous messages are indeed part of the thread.

You can also check the official OpenAI documentation and call your assistant yourself using curl commands. This way you can ensure that the behaviours you experience on your own machine and AI Engine are similar.