A suggestion about assistant's instruction

Hmm, I feel it’s difficult to use the assistant in real products now. It seems somewhat useful, but not completely so.
From my own needs, I most hope that the thread can automatically and dynamically adjust the prompt (not me to request again, or a function call). because that will save lots of request time. For example:

Prompt_A: check reverse-engineer…
if Prompt_A return false, it can automatically run with prompt_B
Prompt_B: the main instruction…

If it always need to return the result to me and then need I to initiate the request again, I feel it is simpler and more intuitive to use the original ChatCompletion.

or if an Assistant can call another assistant in ‘run’,and automatic handle the response. It may also have the same effect.

Thank you for your hard working, I appreciate it

1 Like

you may want to check out the amazing world of agents :thinking:

Agreed, sounds like Langchain Multi Chain Router could be useful to you:

Hmm… the Langchain is just a wrapper for the OpenAI API, and it does not reduce the actual number of requests.

Won’t reduce the amount of requests, but based on what you initially talked about in terms of letting the agent perform ReACT reasoning, you may want to check out their Agents, it sounds like what you were first talking about.