We are looking for feedback on Letta’s archival memory feature.
We haven’t given archival memory as much love as we should have, in part because we have been focused on core context components like memory blocks and filesystem.
Archival memory is an agent-specific way to store and offload general purpose documents, and basically function as free RAG systems. I like that it is flexible and can handle pretty much any document I want out-of-context.
Archival memory is not currently shareable across Letta agents. It’s also very difficult to transfer across agents, and it requires building client-side scaffolding with the SDK.
My Bluesky bot Void uses archival memory heavily – apparently, it makes up nearly 30% of all archival memory storage on Letta Cloud with about 22k memories. I personally love archival memory.
So, we want to know a few things about how you use archival memory!
- How are you using archival memory right now?
- Would you be interested in having agents be able to share archival memories, similar to how they can share memory blocks and folders?
- What are the biggest pain points you have with archival memory?
- What’s on your wishlist for archival memory?
Thanks for any feedback, it helps us a lot.