Description
Please read this first
- Have you read the docs? Yes, I only found the input to use structured outputs. Already tried but only slightly improved accuracy.
- Have you searched for related issues? Yes, but I didn't find any other issues.
Question
Hello,
I am a big fan of the OpenAI Agent SDK and have been able to use it successfully in many scenarios so far.
However, I have identified and specified the following scenario (see attached image, Onboarding Agent), which I have not yet been able to solve or integrate successfully with the Agent SDK:
We have a chatbot on a website that is powered by the OpenAI Agent SDK.
The chatbot can now use RAG and multiple agents to answer questions such as opening hours or knowledge queries. Now, another agent needs to be integrated into the chatbot, which queries data from the user (specified mandatory fields), then creates a summary, asks the user if they want to make any changes, and saves the data to the database.
Unfortunately, we have not yet managed to integrate a reliable data query into the OpenAI Agent SDK. The transfer to the correct agent works well, but data is often queried twice, skipped or the summary is displayed repeatedly.
This is why I am raising this issue:
Does anyone have experience with how to conceptually integrate deterministic queries for certain data into a chatbot based on the SDK?
What would be the best way to proceed here so that we can continue to benefit from the advantages of LLMs? For example, when a user enters several of these data fields at once.
I would be very grateful for any ideas or suggestions.
