Sleep Time Agents Letta In letta, you can create special sleep time agents that share the memory of your primary agents, but run in the background and can modify the memory asynchronously. This document describes the sleep time agent pattern, a specialized multi agent architecture where background agents periodically process conversation history to optimize memory blocks without interrupting the main agent's responsiveness.
Sleep Time Agents Letta This repo contains code to reproduce the empirical aime gsm results in the sleep time compute research paper. if you are interested in building agents that use sleep time compute, view the official sleep time agents developer docs. The “sleep time compute” concept introduced by letta proposes that agents benefit from scheduled downtime, during which they process chat histories, reorganize context, and essentially “refresh” their internal models. We introduce sleep time compute, which allows models to "think" offline about contexts before queries are presented: by anticipating what queries users might ask and pre computing useful quantities, we can significantly reduce the compute requirements at test time. We expect to eventually default to the new sleep time architecture for all agents – excited to have people try it out!.
Sleep Time Agents Letta We introduce sleep time compute, which allows models to "think" offline about contexts before queries are presented: by anticipating what queries users might ask and pre computing useful quantities, we can significantly reduce the compute requirements at test time. We expect to eventually default to the new sleep time architecture for all agents – excited to have people try it out!. I compare test time and sleep time compute via exam night prep, then detail letta’s `rethink memory ()` pattern and trade offs for production agents. Build ai agents with long term memory, advanced reasoning, and custom tools inside a visual environment using the agent development environment, or with python and node.js sdks. The sleep time agent automatically reviews your conversation history every few turns (you can configure the frequency). it summarizes the dialogue, identifies the user's personality, tracks key decisions, and updates memory blocks like human or persona. This repo contains code to reproduce the empirical aime gsm results in the sleep time compute research paper. if you are interested in building agents that use sleep time compute, view the official sleep time agents developer docs.
Comments are closed.