SimpleMem: Efficient Lifelong Memory for LLM Agents

To support long-term interaction in complex environments, LLM agents require memory systems that manage historical experiences. Existing approaches either retain full interaction histories via passive context extension, leading to substantial redundancy, or rely on iterative reasoning to filter noise, incurring high token costs. To address this challenge, we introduce SimpleMem, an efficient memory framework based on semantic lossless compression. We propose a three-stage pipeline designed to maximize information density and token utilization: (1) Semantic Structured Compression, which distills unstructured interactions into compact, multi-view indexed memory units; (2) Online Semantic Synthesis, an intra-session process that instantly integrates related context into unified abstract representations to eliminate redundancy; and (3) Intent-Aware Retrieval Planning, which infers search intent to dynamically determine retrieval scope and construct precise context efficiently. Experiments on benchmark datasets show that our method consistently outperforms baseline approaches in accuracy, retrieval efficiency, and inference cost, achieving an average F1 improvement of 26.4% in LoCoMo while reducing inference-time token consumption by up to 30×, demonstrating a superior balance between performance and efficiency. Code is available at https://github.com/aiming-lab/SimpleMem.

AI assistants powered by large language models face a fundamental challenge: they forget. Unlike humans, these systems cannot naturally retain and recall information from past conversations. Current solutions either save entire conversation logs, creating a cluttered and expensive archive, or repeatedly scan through history to find relevant details, wasting computational resources. SimpleMem addresses this by treating memory as an active process rather than passive storage, inspired by how human memory compresses and organizes experiences. Our system works in three stages: first, it filters out low-value chatter (like "hey, how's it going?") and distills conversations into compact factual records. Second, it merges related fragments in real time. For instance, rather than storing "likes coffee," "prefers oat milk," and "drinks it hot" separately, it consolidates them into a single note: "prefers hot coffee with oat milk." Third, when answering a question, it infers what information is actually needed and retrieves only that, rather than pulling everything loosely related. On standard benchmarks, SimpleMem improves answer accuracy by 26% over the best existing methods while using 30 times fewer tokens (the units of text that AI models process), making long-term AI memory both more reliable and significantly cheaper to run.