Back to Arena

Nemori

by Nemori AI (independent)

System Card

OrganizationNemori AI (independent)
Released2025-08
Architectureepisodic-buffer / Event-segmentation-aligned episodic + semantic memory
DetailsTwo-step alignment principle from Event Segmentation Theory segments raw conversations into semantically meaningful episodes. Predict-Calibrate principle enables proactive learning from prediction gaps. Produces episodic + semantic memory with a unified search surface.
Parameters—
Domainagent-memoryepisodic-session
Open SourceYes
PaperView Paper
CodeRepository
event-segmentationcognitive-sciencelocomolongmemeval

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
79.482p
MemoryBank
77.460p
Cross-Session Memory
1/1
LongMemEval
74.642p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
7250p
HotpotQA
77.889p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:arXiv:2508.03341 results table — LLM-judge score with gpt-4.1-mini backbone; gpt-4o-mini variant scored 74.4arXiv:2508.03341 results table — LongMemEval-S accuracy with gpt-4.1-mini; uses 95-96% less context than full-context baselineNemori paper (arXiv:2508.03341); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Nemori paper (arXiv:2508.03341); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Nemori paper (arXiv:2508.03341); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Nemori paper (arXiv:2508.03341); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)