Testing using qdrant database for memories.

#4
by aA34543534 - opened

Just a quick note to mention that the latest model handles 'memories' better. My system uses qdrant with a "You remember (username) said (content) on (date)" and the RAG dataset provides a significant improvement over the base SciPhi 32k.

SciPhi-AI org

Awesome, thanks for sharing.

Do you have more details on this? It sounds quite interesting.

https://github.com/jason-brian-anderson/long_term_memory_with_qdrant/blob/main/long_term_memory.py using a fork of that class. But I tend to add in a check for both the vector search of the ai response and the user response. I also have a 'dream' function where it takes a vector search on a topic and writes a summary which saves it back to the vector database. So during conversations, the ai has access to all of the memories during content generation.

Sign up or log in to comment