Back to Arena

RecallM

by Cisco Research / independent (Kynoch & Latapie)

System Card

OrganizationCisco Research / independent (Kynoch & Latapie)
Released2023-07
Architecturehybrid / Graph + vector with temporal belief updating
DetailsHybrid memory combining graph databases and vector stores to support temporal reasoning, integrating structured relational data with embedding-based retrieval to enable dynamic belief updating and sequential knowledge tracking.
Parameters—
Domainlifelong-learningknowledge-graphepisodic-session
Open SourceYes
PaperView Paper
CodeRepository
temporal-reasoningbelief-updatehybridgraph

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
78.276p
MemoryBank
73.935p
Cross-Session Memory
1/1
LongMemEval
76.755p
Multi-Hop QA
2/3
BABILong
78.177p
MultiHop-RAG
no data
HotpotQA
77.486p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:RecallM paper (arXiv:2307.02738); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)RecallM paper (arXiv:2307.02738); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)RecallM paper (arXiv:2307.02738); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)RecallM paper (arXiv:2307.02738); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)RecallM paper (arXiv:2307.02738); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)RecallM paper (arXiv:2307.02738); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)