agent-memory

The paper introduces CLR-Bench, a benchmark curriculum of hidden rules for evaluating whether self-evolving LLM agents can learn latent rules from feedback, finding that most current self-evolution methods barely learn unless given a fixed lesson or a stronger base model.

A preprint tests five agent memory systems (Graphiti, Zep, mem0, langmem, cognee) for policy revocation, finding no system reliably prevents agents from retrieving and acting on superseded harmful policies, with revoked records sometimes ranked first and unsafe actions taken in over 40% of trials.

FAIR at Meta 提出 Routed Continual Learning：将过往轨迹按簇划分、每个簇搭配固定记忆库中的一种记忆系统，并由 LLM 路由器决定复用或新建簇，在交错任务流上相对全局 ICL 历史提升 15–30% 归一化奖励并降低 65–72% API 成本。

A developer shares how 'beads,' an agent-native ticket system, solves common multi-agent workflow problems like context preservation, agent delegation, and parallel coordination by providing a shared, persistent ticket database that all agents on a network can access simultaneously.

This article introduces Obelisk, an open-source system that indexes historical sessions from coding agents like Claude Code and Codex into SQLite for direct retrieval by agents. Its design is inspired by the paper "Is Complex Agentic Memory Always Needed?", which found that a simple RAW-RAG approach (retaining raw conversation history and retrieving on demand) outperformed complex agentic memory schemes across multiple memory benchmarks.

MemSteer, developed at Peking University and Microsoft Research Asia, converts user histories into automatically generated supervision tests for agent memory, and MemSteer-Opt uses the resulting evidence to optimize storage, retrieval, and presentation without hand-labeled data or fine-tuning, improving BM25 baselines on ATM-Bench and RHELM.

LangChain highlights an experiment pairing openwiki with a lightweight evidence graph, allowing AI agents to trace generated knowledge back to its supporting evidence for more reliable, domain-specific responses.

LangChain explains how nearly all proactive, always-on agents are built on cron-style scheduled runs—covering memory "dreaming," proactivity, triage, follow-ups, and orchestration—and shows how to build them with Managed Deep Agents on LangSmith, where over 70% of runs are already scheduled rather than user-initiated.

TrawMem reframes LLM agent memory systems as task-oriented role-adaptive workspaces, organizing interaction histories into threads and selecting complementary memories by their task roles, achieving consistent macro-F1 gains of 2.23–3.14 points over representative memory systems on LoCoMo and MemGallery benchmarks.

This paper shows that retrieving the correct fact is insufficient for memory agents, as an obsolete competing statement harms accuracy more than unrelated evidence of the same size, evaluated on MemoryAgentBench's FactConsolidation task with Qwen2.5 and Qwen3 models.

brain.md is a file-based persistent memory layer tool that records project decisions, constraints, and context in pure Markdown within a code repository, letting multiple AI coding agents such as Claude Code, Codex, and Cursor share project knowledge across sessions, so every new session doesn't start from scratch.

Jenova AI founder Boris Wang announces the product crossed a $2M annualized run rate while profitable, driven by users who build persistent AI-narrated story worlds, with dedicated users generating ~700,000-word stories and spending 4x more than other paying users.

Breakdown of memt-1.0's memory-native architecture, which uses gated DeltaNet-2 with a mixture-of-storage layer to internalize procedural experience into model states, delivering 1.86x inference speedup and 66x smaller storage at 128k context versus full KV caching.

The paper introduces a framework of 'memory benefit' and 'memory tax' to measure the net value of dedicated agent memory systems over naive RAG, finding that many memory systems underperform direct retrieval across benchmarks. It analyzes how memory processes can remove or obscure evidence and calls for evaluating the net value of memory rather than standalone performance.

Kent C. Dodds 推荐 Kody.codes，一个位于各家 AI 编码代理（Cursor、Claude Code、ChatGPT 等）之下的共享层，提供记忆、包、触发器、凭证与应用五大原语，让团队在不同代理之间保持上下文与工作流的可移植性。

A developer is asking the community how they actually solve agent memory in production, whether through vector DBs with RAG, knowledge graphs, or human-in-the-loop approval for a shared knowledge base.

LangChain is hosting a virtual workshop on November 3rd covering agent memory and context engineering, featuring a 5-minute context engineering explainer by Jess Weng.

Redis promoted three capabilities for reducing AI application costs: semantic caching via LangCache, persistent agent memory, and RAM-to-SSD storage tiering with Redis Flex, framed as a sponsored post by @svpino.

The paper introduces REM (Reusable Experience Memory), a framework that manages experience memory for coding agents through store, recall, and forget decisions inside the think-act loop, outperforming state-of-the-art memory systems on SWE-ContextBench Lite with 68.7% match@1 and 27.6% task resolution at lower cost.

The MemTensor team releases MemT 1.0, pushing memory mechanisms from bolt-on RAG and Agent Memory into the core of model computation, allowing models to accumulate experience while solving tasks. Official tests show improvements of over 5% in Coding and Working scenarios, and the source code is now open-sourced on Hugging Face.