daidocs launches open plain-text .dai file format for AI memory with MCP server integration

An arXiv post evaluates an auditable long-term memory system on LongMemEval-S, combining hybrid candidate retrieval, cross-encoder reranking, coverage-first packet compilation, and deterministic scaffolds, with an LLM only as a replaceable final reader. Two 500-question passes with a Claude Opus reader score 479/500 and 475/500, straddling Chronos High's published 478/500 without establishing superiority. The authors release artifacts for re-scoring while flagging major caveats: no held-out evaluation, modified scoring prompts, and unresolved reader-context effects.

Products·launched

MemBukkit: open-source long-term memory layer for LLM/RAG apps

MemBukkit is an open-source long-term memory layer for LLM/RAG applications that converts conversations and documents into dated atomic facts, enabling temporal memory and efficient retrieval. It achieves 92.6% on LongMemEval-S using the official GPT-4o judge, with a reader seeing only ~3.2k tokens per question instead of ~100k, solving the problem of context window limits and improving accuracy over full-context reading.

MemBukkit is an open-source long-term memory layer for LLM/RAG applications that converts conversations and documents into dated atomic facts, enabling temporal memory and efficient retrieval. It achieves 92.6% on LongMemEval-S using the official GPT-4o judge, with a reader seeing only ~3.2k tokens per question instead of ~100k, solving the problem of context window limits and improving accuracy over full-context reading.

A developer open-sourced Memory Engine MCP, a local-first long-term memory system for AI coding tools that models information as typed atoms connected by typed bonds. It uses a hybrid ranking pipeline combining SQLite FTS5 full-text search and semantic similarity via Ollama, aiming to connect and curate data rather than just store it.

Products·launched

samemind: AI agent memory as plain markdown in git, no cloud or API keys

samemind stores AI agent memory as plain markdown files in a git repository, eliminating the need for databases, cloud services, or API keys. It works across 12 AI coding engines including Claude Code, Cursor, and opencode, providing a portable and version-controlled memory layer for AI agents.

samemind stores AI agent memory as plain markdown files in a git repository, eliminating the need for databases, cloud services, or API keys. It works across 12 AI coding engines including Claude Code, Cursor, and opencode, providing a portable and version-controlled memory layer for AI agents.

Memmy is a personal memory hub that provides a shared long-term memory layer across multiple AI agents like Cursor, Claude Code, and Codex. It distills user knowledge, preferences, and project experience into a single memory that evolves through collaboration, solving the problem of fragmented context across different AI tools. Available as a desktop app, CLI, and API.