Skip to content




Table of Content

letta

MemGPT

https://huggingface.co/papers/2310.08560

https://tersesystems.com/blog/2025/02/14/adding-memory-to-llms-with-letta/

  • user prompt to a model
    • a model equipped with "core memory", core context, "message memory" with the recent messages, "archival memory" on a vector database
  • organize message memory to free up memory, to store old memory to the vector database as necessary
  • load context
  • load recent messages
  • search relevant memory from a long term memory, the vector database
    • and recall memory as appropriate
  • process all memory streams together
  • generate response and get back to user
  • store interaction, do memory management

letta setup

https://docs.letta.com/

Letta enables developers to build and deploy stateful AI agents - agents that maintain memory and context across long-running conversations

letta ADE

Use Google Chrome to access https://app.letta.com/.