[Question]: How Do I prevent the Assistant changing the user's input when calling an action? #2976
-
What is your question?I am using an OpenAI assistant which has an action configured to access long-term memory. I would like what the user entered to be passed to the action, but looking at the action debug in the Librechat UI, I see it calls the action and summarizes the user's input. This breaks the memory retrieval. Is there a way to just send exactly what the user entered or at least have some control over the LLM which is changing the intent please? More Details. What is the main subject of your question?No response ScreenshotsNo response Code of Conduct
|
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
You would have to hack the code a bit to force the user's input as the input for your desired action, every time. Otherwise, the only other way is through very specific prompting, because OpenAI Assistants will always form their own inputs for tools (but can always be overriden programmatically) |
Beta Was this translation helpful? Give feedback.
-
|
OK, thanks! |
Beta Was this translation helpful? Give feedback.
You would have to hack the code a bit to force the user's input as the input for your desired action, every time.
Otherwise, the only other way is through very specific prompting, because OpenAI Assistants will always form their own inputs for tools (but can always be overriden programmatically)