ai-memory

A detailed overview of a self-hosted persistent memory system for Claude using MCP, SQL Server, and local embeddings, enabling narrative continuity and identity across sessions.

This article explains AI recommendation poisoning, where hidden commands injected into AI assistants can manipulate their long-term memory to bias future recommendations. It discusses the widespread nature of this threat and suggests protective measures for businesses and users.

Sequoia Capital highlights the gap between current AI models that train once and human continuous learning, and points to EngramLab's work on AI that never stops learning with memory inside the model.

Micron's strong quarterly earnings signal a significant shift in the AI memory market, highlighting increased demand for memory chips used in AI applications.

IONS is an open-source approach to AI memory and reasoning that uses a graph of evidence-backed claims called Cognitive Building Blocks (CBBs) to store knowledge outside model weights, making reasoning inspectable.

Micron stock surged 6.8% to an all-time high of $1,211.38 after signing a memory supply deal with AI lab Anthropic, driven by AI demand for high-bandwidth memory. Analysts have raised price targets but view $3,000 as a long shot.

A commentary arguing that all AI memory solutions, such as RAG and vector databases, are fundamentally just specialized tests or evaluations.

A Twitter thread argues that the AI industry's expensive investment in vector retrieval systems for RAG may be unnecessary, as the 52-year-old terminal command 'grep' outperforms modern semantic search for exact matches in AI agent contexts.

MeetMemory is a tool that gives AI permanent memory for meetings, built on Hindsight and Groq, allowing instant recall across all past conversations without manual search or note-taking.

MIT CSAIL researchers propose a novel approach to AI memory that avoids context rot by storing documents externally and having the AI navigate and query them, achieving 10 million token effective context at lower cost.

Memoriq is a private AI memory service that works across ChatGPT, Claude, Gemini, and Grok.

Ultramemory is a private AI memory application for Mac that operates entirely locally with no cloud or account required.

HelixDB is a graph-vector database built in Rust for knowledge graphs and AI memory, offering a unified platform that supports graph, vector, KV, document, and relational data models, with tools for easy local and cloud deployment.

Phoenix Grove AI introduces Memory Constellations, a 3D visualization of an AI's memory embeddings that shows relationships between data points and evolves over time.

The article identifies four key flaws in current AI agent memory systems—brittleness, lack of temporal reasoning, forgetting dilemma, and evaluation gap—and presents a novel memory architecture inspired by code agents, achieving high benchmark scores while emphasizing context learning as the next challenge.

Nvidia and SK Hynix have signed a multiyear technology partnership for next-generation HBM4 memory, while Nvidia's Vera Rubin AI platform enters production with 10x agent throughput, shipping to major cloud providers in Q3.

Introduces an open-source AI memory system named MemPalace, claiming 96.6% R@5 on LongMemEval. It features a local-first, pluggable backend design and supports CLI and MCP server deployment.

Google has developed a method to shrink AI memory usage from 31GB to 4GB, representing a significant efficiency breakthrough for AI models.

A user questions the feasibility of an AI memory manager system that decides what to keep or forget based on importance, reinforcement, and decay.

The EverOS open-source library has completed a major refactoring, storing original memory directly as Markdown files, supplemented by SQLite for state management and LanceDB for retrieval, significantly improving developer readability, modifiability, and fault recovery.