Back to Arena

LangMem

by LangChain

System Card

OrganizationLangChain
Released2025-02
Architecturehierarchical-summary / Episodic + semantic + procedural layers
DetailsThree-layer memory store on top of LangGraph: episodic snippets, semantic facts, and procedural prompts, each with its own write and retrieval policy. Background summarization compresses old episodes.
Parameters—
Domainagent-memoryepisodic-sessionpersonalization
Open SourceYes
WebsiteVisit
CodeRepository
langchainlanggraphtieredsummary

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
6114p
MemoryBank
71.123p
Cross-Session Memory
1/1
LongMemEval
58.37p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
7143p
HotpotQA
67.838p
Agent Task Memory
1/1
AgentBench-Mem
62.79p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:LangMem launch postLangMem (langchain-ai/langmem); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)LangMem (langchain-ai/langmem); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)LangMem (langchain-ai/langmem); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)LangMem (langchain-ai/langmem); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)