Is there any configuration that could be used to make the chat history limitation when calling the LLM?

[quzhixue-Kimi]

hi there,

There has been one flowise chatbot I configured with upstash redis-backed chat memory. And, langsmith has been added to track the whole process and parameters. (It is just a chat bot flow with memory, no vector store and record manager)

I found that when I talked to my LLM, all the histories of chat message will be added as 'chat_history' parameter.

It will cost the LLM's token a lot. Even, some content of the chat history has no relation with my current topic.

It is appreciated that you may help and guide me to make some configuration about the 'chat_history' limitation to reduce the LLM's token cost.

BR Kimi

[toi500]

Are u the only user that uses this chatbot?

[quzhixue-Kimi]

Hi there,

I would like to publish the chatbot for registered users. So, there might be a number of users who will use the chatbot.

BR Kimi

[HenryHengZJ]

There are different ways to help with that, have a look at the docs: https://docs.flowiseai.com/integrations/langchain/memory/buffer-window-memory https://docs.flowiseai.com/integrations/langchain/memory/conversation-summary-memory https://docs.flowiseai.com/integrations/langchain/memory/conversation-summary-buffer-memory

[quzhixue-Kimi]

hi Henry,

Thank you for your reply. I am using the upstash redis-backed chat memory in my chatbot. It seems that there is no configuration about it.

What I want to do is to record each user's chat history in upstash redis, and the session id in redis is just the unique user name.

BR Kimi

[HenryHengZJ]

I'm afraid for now you'll have to use one of the memory nodes mentioned above, it's similar to upstash. Messages are stored with session if into your Flowise database

[quzhixue-Kimi]

Hi there,

Thank you for your kind help.

BR Kimi