Tag
Raymond Chen explains how to use std::unique_ptr to manage a registration cookie in a fake agile wrapper, discussing integer-to-pointer round-tripping and the wil::unique_any alternative.
TencentDB Agent Memory releases a 2.0.0 beta introducing Team Memory, a shared memory hub that lets agents and teammates reuse chat, docs, and code assets. The repo hit #1 on GitHub's TypeScript trending this week.
Share Codex usage tips: avoid excessive memory usage by setting a prompt for scheduled cleanup of memory and cache, and reference old conversations to quickly get context and save tokens.
VerMem is a framework for unified memory management in LLM agents, using local and global verifiers with reinforcement learning to jointly control long-term and short-term memory. It outperforms strong baselines across five benchmarks with improved efficiency-performance trade-offs.
AgentMemBench is a systematic benchmark that evaluates five long-term memory management strategies for conversational AI agents across three datasets, finding that external key-value store retrieval dominates on quality but incurs a larger memory footprint.
Introducing Memmy, an open-source AI memory tool that remembers projects, skills, and business process diagrams across tools. Through real-world tests, it demonstrates the ability to locate local projects, invoke Skills, and connect to business agents, emphasizing its core concept of letting multiple AI tools share a single working memory.
A project claims to run Kimi K3, a 2.78T parameter MoE model, on CPU with 8GB RAM by streaming expert weights from NVMe SSD and using MXFP4 compression, trading speed for memory efficiency.
Memmy is an open-source project that can uniformly scan and manage conversation logs and memories from common AI Agents like Codex and Claude Code. It supports Mac/Windows and TUI/CLI, allowing different Agents to share context.
UniMem proposes a self-routing framework that combines episodic and parametric memory for LLM agents, enabling adaptive memory management in boundary-agnostic task streams without task labels.
ARC improves retrieval efficiency and accuracy in long-context AI agents by separating append-only storage from bounded citation views, achieving near-perfect recall with lower latency and bandwidth.
MemOS is an open-source memory operating system for developers building LLM applications and AI agents, providing persistent, editable memory through a unified API with features like multimodal memory and async ingestion.
oh-my-hermes is an all-in-one suite for Hermes Agent that adds memory management, coding orchestration, wiki building, and more, simplifying agent workflows and reducing the need for multiple plugins.
A developer benchmarks Rote, a memory manager for browser agents that sends page diffs instead of full re-renders, showing a 37% reduction in token growth compared to Browser Use, but with trade-offs on short tasks.
Go 1.26 makes Green Tea the default garbage collector, improving cache-friendliness. This article visualizes heap allocation with Go and C#, and discusses challenges with non-moving collectors and sparse pages.
This article explores the tradeoffs between stateful and stateless agent design for scalable AI systems, providing implementation examples using Groq API and Llama 3.1 8B Instant model.
Discusses building an open MCP memory server to address real issues with agent memory, identifying what is broken and how to fix it.
The article argues that for effective agent memory, it is crucial to forget bad examples, as retaining them degrades performance.
The article explains why making the entire stack out of guard pages is problematic, as it could lead to unbounded memory allocation and system hangs. Instead, fixed guard pages are preferred for bounded fault handling.
Proposes MemDefrag, a training-free framework that uses a middle-layer tracing signal to defragment latent memory in LLMs, achieving significantly better knowledge retention than existing methods like MemoryLLM and M+.
PraMem proposes a paradigm shift for long-horizon behavior prediction by using a lengthy historical sequence as a resource to build experiential memory via practice, improving LLM performance on behavior prediction tasks.