My Chatbot is remembering old files, how can I delete the memory?

[MrMrch]

I need to reset the memory of the chatbot, it keeps using an older version of the file which was a test, and now I can't make it stop.
Im using the QA Conversational Retrieval as per this tutorial https://www.youtube.com/watch?v=riXpu1tHzl0&ab_channel=DaveEbbelaar

[trendion]

same problem here

[HenryHengZJ]

Have you specified filter or namespace or combination of both for the Pinecone node?

[MrMrch]

Mmh I think only the namespace, but I'm not at my pc right now

[MrMrch]

@HenryHengZJ sorry so I added a new file instead of the previous one, but still not seeing any change. The new informations are not received or used when asking the questions

[LeandroTecware]

I have the same problem. As a first question, from a new chat it is possible to notice that the chat has accumulated content or returns an error message:

This model's maximum context length is 4097 tokens. However, you requested 4140 tokens (3140 in the messages, 1000 in the completion). Please reduce the length of the messages or completion.

[rvy474]

Upserting new data does not mean the old data is deleted. You will first need to delete the old data. Below is some python code that I ran on Google Collab.

#install pinecone client
!pip install pinecone-client

#get the project connected
import pinecone
pinecone.init(api_key="your key", environment="your environment")
index = pinecone.Index("your index")

#delete the index
index.delete(delete_all=True, namespace='')

If you have not added a namespace then use '' else use the actual name of the namespace.

Deleting the old data, saving your chat flow and then upserting the new data should solve the problem

[HenryHengZJ]

you can either delete the old data, or store the new data to a new namespace/collection, basically concept of another folder. That way you can prevent the mix of knowledge file/data

[mwmdev]

What you need is a Record Manager node in your flow, in particular setting the « Cleanup » option in additional parameters.