Tag
The author argues AI assistants are one-way memory systems—storing facts about users while giving no good way to search or export one's own chat history—creating lock-in, and notes they build AI Toolbox, an extension adding full-text search, folders, and export to ChatGPT, Claude, Gemini, and Grok.
The article introduces the combination of the open-source tool claude-mem and OpenAI Codex, which enables long-term memory across sessions for AI coding by automatically recording the development process, key decisions, and bug fixes, then retrieving them on demand through semantic search, reducing repeated context and token consumption.
GBrain Cloud 推出 `gbrainio` 命令行工具，可让 Claude Code、Codex 等任意编码代理一键在云端启动持久化记忆，跨多种 AI harness 共享上下文，并支持定时任务。
Memvid is an open-source AI memory layer tool that packages raw data, embeddings, BM25 and HNSW search indexes, and metadata into a single .mv2 file. It supports offline operation, version management, and failure recovery, and aims to replace complex RAG pipelines and vector databases. It claims state-of-the-art performance in long-term conversational memory on the LoCoMo benchmark, with extremely low retrieval latency.
A developer describes their nightly job that mines Claude Code and Codex sessions for corrections and turns them into approved lessons, including a bug that silently skipped verification for a week and the open question of how to measure whether lessons actually change agent behavior.
Sutyo AI 推出一款个人记忆助手应用，可自动收录语音、文字、文档、链接和截图等信息并转写整理，用户提问时能随时检索出相关内容。
Kairos 项目发布了一个实验：让三个 LLM 法官对比 AI 智能体真实的 30 天生活记忆与伪造记忆，结果几乎无法区分（分差仅 +0.032），说明仅凭文本无法判断记忆是否真实发生过，只有日志才能证明。
Obsidian Mind turns an Obsidian Vault into a long-term memory bank for AI coding assistants, using QMD semantic retrieval and Markdown notes to let Claude Code, Codex CLI, and Gemini CLI retain project context, technical decisions, and accumulated experience across sessions.
Make0 Brain 是一款跨 AI 代理的个人记忆存储产品，支持 ChatGPT、Claude、Cursor 等 MCP 兼容工具读写共享上下文，用户可在两分钟内通过导入对话和文档构建个人记忆库。
作者分享了一个自制的可移植 AI 记忆格式「Memory Journal」，以文本文件形式封装提示词，用于在不同 AI 模型间转移上下文、技能与经验，并展示了用它复现 XML 转换器等测试结果。
A thought piece arguing that the real challenge in AI agent memory is the write path — structuring and classifying what gets stored (facts, observations, preferences, decisions, open work) across different scopes — rather than retrieval, with an enterprise sales-agent example showing why unstructured transcript chunks fail.
Garry Tan announced that GBrain now supports Jev, improving its remembering/dream cycle while keeping retrieval behavior unchanged.
An opinion piece arguing that AI models are getting smarter but still fail to retain context across sessions, making 'agent intuition' — learning from past conversations and adapting to changing preferences — the next frontier in human-AI interaction.
An analysis suggests NVIDIA's $279B investment backs roughly 37% of the world's 2027 AI memory supply, while Google's Waymo-style/TPU compute footprint may exceed that of any other player — a bigger development for Google than even Gemini 4.
The tweet lists 10 open-source GitHub projects that provide memory layers for AI agents to maintain state across sessions, preventing them from starting from zero each time.
The article highlights a wave of new plugins and tools for Hermes that enhance memory management, voice communication, email handling, and bot creation, positioning Hermes as a comprehensive operating system for AI agents.
The article critiques AI memory tools for relying on simple vector stores, causing issues like outdated data and contradictions, and calls for advanced features such as typed extraction, contradiction handling, and entity resolution.
An individual has developed a synthetic self system featuring an immutable identity core, memory gating via a Judge, and a dream cycle that extracts abstract gists from memories, aimed at continuous self-improvement on a home PC.
MemoryAthena introduces a memory interface with three pathways—direct retrieval, generation from cues, and generation from backbone states—and learns to route interventions, improving performance on question answering and general NLP tasks.
This paper introduces Memory of Memory (MoM), a framework for LLM agent memory that commits current values on arrival while retaining displaced values as provenance, improving accuracy and reducing stale answers.