longterm memory · AI Automation Society

longterm memory

Hey! I'm exploring ways to build long-term memory for my assistant, storing every input and output in a Pinecone vector database. The idea is to capture each message as a vector, tagged with metadata like sender, timestamp, and session ID, so the assistant can access and reference past interactions based on semantic similarity. How have others approached creating long-term memory for assistants? Any best practices or tips on structuring the workflow efficiently?

6 comments