GitHub - kingjulio8238/memary: Longterm Memory for Autonomous Agents.

3 days reading through the code of the paper that solved this, here are some notes:

- Long term interactions: the promise of LLM agents is to potentially run over weeks. A lot of the ideas in the paper tackle exactly this problem

- Memory: It’s easy to write down interactions of agents. But how do we track these over time? How do these change with ti... See more

For any conversational use case, such as chatbots, we need conversational memory. The idea behind conversational memory is simple: rather than sending just the most recent interaction to our LLM, we send a history of interactions + the most recent interaction to our LLM — typically in a chat log style format.

With and without conversational memory.

D... See more

LLM agents rely on their Long-Term Memory (LTM) to store and retrieve crucial memories. Conversations, thoughts, plans, actions, observations, skills, and behaviors are all stored within a vector database. The LTM enables pre-processing and post-processing of memories, ensuring optimal retrieval. By considering factors such as context, recency, imp... See more