Long Term Memory for AI assistant
shahar lotan
4 replies
Hey Builders,
I'm currently working on an AI assistant app and have been thinking a lot about how to integrate a long-term memory feature (on top of the current conversation history). The goal is to have the assistant "remember" user preferences, past interactions, and other contextual information over time, so it can provide more personalized and context-aware assistance.
I’d love to get some insights from anyone who has tackled something similar or has experience with long-term memory in AI systems.
Any advice, resources, or pointers to relevant tools/libraries would be greatly appreciated!
Replies
Chirotpal@cpd_2k5
@shahar_lotan We've already built an infrastructure to take care of memory for AI agents, we just launched our private beta on PH - check this out it might really help you. For early adopters we're providing generous free usage credit too to get started.
https://www.producthunt.com/post...
Share
Yescribe.ai
Yep, vector databases like Pinecone or Weaviate are a great option for storing embeddings to enable long-term memory for AI assistants. They let you efficiently store and retrieve relevant context from past interactions. Claude + a vector DB is a powerful combo for building AI with memory.
If you're using LLMs, which it sounds like you are in context to remembering user preferences and past interactions over time. Have you looked into vector databases?
Finalle.ai
Thanks @baba_nadimpalli! I'm using pinecone, the hard part for me is designing the memory itself- which data should i save, how should i save it? (full context or summary?), which data should i forget etc.