A-MEM

by AGI Research / Rutgers

System Card

OrganizationAGI Research / Rutgers

Released2025-02

Architectureagentic-workflow / Zettelkasten-inspired dynamic linked notes

DetailsGenerates structured notes with contextual descriptions, keywords, and tags for each memory, then dynamically identifies connections to existing memories. Memory evolves as new entries trigger updates to historical memories, producing self-organizing knowledge networks.

Parameters—

Domainagent-memorylifelong-learning

Open SourceYes

PaperView Paper

zettelkastendynamic-linkingagenticself-organizing

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

86.497p

77.564p

Cross-Session Memory

1/1

83.193p

Multi-Hop QA

2/3

73.636p

MultiHop-RAG

no data

7784p

Agent Task Memory

1/1

7226p

Personalization

0/1

PerLTQA

no data

Factuality / Grounding

0/1

RAGAS

no data

Sources:A-MEM paper (arXiv:2502.12110); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)A-MEM paper (arXiv:2502.12110); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)A-MEM paper (arXiv:2502.12110); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)A-MEM paper (arXiv:2502.12110); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)A-MEM paper (arXiv:2502.12110); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)A-MEM paper (arXiv:2502.12110); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)