Back to Arena

Haystack Memory

by deepset

System Card

Organizationdeepset
Released2023-12
Architecturevector-rag / Pipeline-based memory store
DetailsMemory components inside the Haystack 2.x pipeline framework. Stores conversation turns and retrievable documents in a unified document store; integrates with major vector backends.
Parameters—
Domainrag-retrievalagent-memory
Open SourceYes
WebsiteVisit
CodeRepository
haystackdeepsetpipelineproduction

Capability Profile

Benchmark Scores

7 of 14 benchmarks
Long-Context Retrieval
1/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
603p
∞Bench
no data
Multi-Turn Recall
1/2
LoCoMo
62.816p
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
63.217p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
72.153p
HotpotQA
51.23p
Agent Task Memory
1/1
AgentBench-Mem
69.721p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
1/1
RAGAS
70.452p
Sources:Haystack evalsHaystack benchmarkHaystack Memory (deepset-ai/haystack); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)Haystack Memory (deepset-ai/haystack); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Haystack Memory (deepset-ai/haystack); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Haystack Memory (deepset-ai/haystack); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)Haystack Memory (deepset-ai/haystack); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)