Conversation Memory

Memory within a conversation, LangChain Memory Docs

Development plan:

Required functionality: [ ]- Simple Buffer (keep n most recent messages) - https://python.langchain.com/docs/expression_language/cookbook/memory

Future development: [ ]- Token Buffer (keep a buffer of messages that is less than n tokens) [ ]- Summary Buffer (keep a buffer of messages that is less than n tokens, but also summarize the messages) [ ]- Time weighted Vectorstore (keep a vectorstore of messages, weighted by time) - I would want to implement something like this, but implement something like: - All messages tied to the User are in the vector store - Retriever accesses this vector store based on similarity to current Couplet Contextinversely weighted by time (recent messages advantaged) - Messages within Chat context given a (very?) big advantage over messages outside of context (Tunable via UserInterface) - Give a (very strong) advantage to 'in-chat' (ensuring they will be selected in the RAG model),

or like, sigmoid or whatever so that recent messages within the conversation are weighted more heavily than older messages. - https://js.langchain.com/docs/modules/data_connection/retrievers/time_weighted_vectorstore