memU

by NevaMind-AI

System Card

OrganizationNevaMind-AI

Released2025-01

Architecturehierarchical-summary / Filesystem-style (Resources / Items / Categories)

DetailsHierarchical memory for 24/7 proactive agents. Continuously captures user intent in the background, auto-extracts facts, and organizes them into a three-layer file-system metaphor with pgvector and in-memory backends.

Parameters—

Domainagent-memorypersonalizationlifelong-learning

Open SourceYes

WebsiteVisit

CodeRepository

proactive-agenthierarchicalalways-onpgvectorcontext-cache

Capability Profile

Benchmark Scores

6 of 14 benchmarks

Long-Context Retrieval

0/5

RULER

no data

NIAH

no data

LooGLE

no data

LongBench

no data

∞Bench

no data

Multi-Turn Recall

2/2

LoCoMo

92.199p

MemoryBank

75.147p

Cross-Session Memory

1/1

LongMemEval

8289p

Multi-Hop QA

1/3

BABILong

78.380p

MultiHop-RAG

no data

HotpotQA

no data

Agent Task Memory

1/1

AgentBench-Mem

7226p

Personalization

1/1

PerLTQA

73.227p

Factuality / Grounding

0/1

RAGAS

no data

Sources:memu.pro/benchmark + github.com/NevaMind-AI/memU README — Self-reported hybrid retrieval (semantic + keyword + contextual)third-party launch coverage (X/Twitter) — LongMemEval-S; weaker sourcing, not on official page memU (NevaMind-AI/memU); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)memU (NevaMind-AI/memU); evaluated on PerLTQA: A Personal Long-Term Memory Question Answering Dataset (PolyU, 2402)memU (NevaMind-AI/memU); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)memU (NevaMind-AI/memU); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)