ActiveMem: Distributed Active Memory for Long-Horizon LLM Reasoning

In three linesActiveMem introduces a distributed architecture decoupling agent memory from LLM reasoning. A high-level Planner executes reasoning with distilled semantic gists while a lightweight parallel memory system actively accumulates and consolidates these gists. State-of-the-art results on BrowseComp-Plus and GAIA with reduced overhead.

Summary generated by Claude — human-verified