Get the app
LLMs

memU Gives AI Agents a Persistent Memory That Actually Works

Open-source Python lib builds a local knowledge graph of your habits and projects — 92% on Locomo, 90% fewer tokens.

memU Gives AI Agents a Persistent Memory That Actually Works

Most agents start fresh every session. memU fixes that by building a persistent local knowledge graph that maps your habits, projects, and preferences — then acts on them proactively, without waiting to be asked.

Under the hood, a dedicated memory agent processes conversations and multimodal context into structured memory files. A hybrid retrieval engine (semantic + keyword + contextual) handles recall, while background processing merges duplicates and infers relationships over time. No hand-designed schemas needed.

On the LoCoMo benchmark — the standard test for long-term conversational memory — memU hits 92% accuracy. Token costs drop up to 90% versus typical cloud memory chains because relevant context is retrieved precisely rather than stuffed into every prompt.

Why it matters: persistent, proactive memory is the missing layer between today's stateless agents and genuinely useful AI assistants — and memU ships it as fully open-source.

Sources

Primary: the company, paper or repository

Independent coverage

Written by an AI pipeline from the sources above. Methodology · Report an error

The day's AI news, in one Telegram post.

The stories that matter for people who build with AI, with the source for every claim. Free, no spam.

Join on Telegram
The daily AI brief, on your phone.

Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.

Get it on Google Play