Back to Arena

Memoro

by MIT Media Lab

System Card

OrganizationMIT Media Lab
Released2024-03
Architecturevector-rag / Wearable semantic memory with query/queryless modes
DetailsWearable bone-conduction audio assistant using an LLM to infer memory needs mid-conversation. Semantic search over user memories; two modes (explicit voice queries and on-demand predictive assistance).
Parameters—
Domainpersonalizationagent-memory
Open SourceNo
PaperView Paper
WebsiteVisit
chi-2024wearableaudiobone-conductionreal-time

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
55.14p
MemoryBank
61.68p
Cross-Session Memory
1/1
LongMemEval
60.614p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
602p
HotpotQA
61.818p
Agent Task Memory
1/1
AgentBench-Mem
63.210p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Memoro paper (arXiv:2403.02135); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Memoro paper (arXiv:2403.02135); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Memoro paper (arXiv:2403.02135); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Memoro paper (arXiv:2403.02135); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Memoro paper (arXiv:2403.02135); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Memoro paper (arXiv:2403.02135); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)